lint

Signed-off-by: Duncan Moss <djm.moss@gmail.com>
add comments
2025-11-04 19:54:45 +00:00 · 2025-11-04 19:54:43 +00:00 · 2025-11-04 19:53:57 +00:00 · 2025-11-04 19:52:25 +00:00
1956 changed files with 57613 additions and 163203 deletions
@@ -1,24 +0,0 @@
-name: vllm_ci
-job_dirs:
-  - ".buildkite/test_areas"
-  - ".buildkite/image_build"
-run_all_patterns:
-  - "docker/Dockerfile"
-  - "CMakeLists.txt"
-  - "requirements/common.txt"
-  - "requirements/cuda.txt"
-  - "requirements/build.txt"
-  - "requirements/test.txt"
-  - "setup.py"
-  - "csrc/"
-  - "cmake/"
-run_all_exclude_patterns:
-  - "docker/Dockerfile."
-  - "csrc/cpu/"
-  - "csrc/rocm/"
-  - "cmake/hipify.py"
-  - "cmake/cpu_extension.cmake"
-registries: public.ecr.aws/q9t5s3a7
-repositories:
-  main: "vllm-ci-postmerge-repo"
-  premerge: "vllm-ci-test-repo"
@@ -0,0 +1,46 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+import argparse
+import os
+
+template = """<!DOCTYPE html>
+<html>
+    <body>
+    <h1>Links for vLLM</h1/>
+        <a href="../{x86_wheel_html_escaped}">{x86_wheel}</a><br/>
+        <a href="../{arm_wheel_html_escaped}">{arm_wheel}</a><br/>
+    </body>
+</html>
+"""
+
+parser = argparse.ArgumentParser()
+parser.add_argument("--wheel", help="The wheel path.", required=True)
+args = parser.parse_args()
+
+filename = os.path.basename(args.wheel)
+
+with open("index.html", "w") as f:
+    print(f"Generated index.html for {args.wheel}")
+    # sync the abi tag with .buildkite/scripts/upload-wheels.sh
+    if "x86_64" in filename:
+        x86_wheel = filename
+        arm_wheel = filename.replace("x86_64", "aarch64").replace(
+            "manylinux1", "manylinux2014"
+        )
+    elif "aarch64" in filename:
+        x86_wheel = filename.replace("aarch64", "x86_64").replace(
+            "manylinux2014", "manylinux1"
+        )
+        arm_wheel = filename
+    else:
+        raise ValueError(f"Unsupported wheel: {filename}")
+    # cloudfront requires escaping the '+' character
+    f.write(
+        template.format(
+            x86_wheel=x86_wheel,
+            x86_wheel_html_escaped=x86_wheel.replace("+", "%2B"),
+            arm_wheel=arm_wheel,
+            arm_wheel_html_escaped=arm_wheel.replace("+", "%2B"),
+        )
+    )
@@ -1,56 +0,0 @@
-#!/bin/bash
-set -e
-
-if [[ $# -lt 8 ]]; then
-  echo "Usage: $0 <registry> <repo> <commit> <branch> <vllm_use_precompiled> <vllm_merge_base_commit> <cache_from> <cache_to>"
-  exit 1
-fi
-
-REGISTRY=$1
-REPO=$2
-BUILDKITE_COMMIT=$3
-BRANCH=$4
-VLLM_USE_PRECOMPILED=$5
-VLLM_MERGE_BASE_COMMIT=$6
-CACHE_FROM=$7
-CACHE_TO=$8
-
-# authenticate with AWS ECR
-aws ecr-public get-login-password --region us-east-1 | docker login --username AWS --password-stdin $REGISTRY
-aws ecr get-login-password --region us-east-1 | docker login --username AWS --password-stdin 936637512419.dkr.ecr.us-east-1.amazonaws.com
-
-# docker buildx 
-docker buildx create --name vllm-builder --driver docker-container --use
-docker buildx inspect --bootstrap
-docker buildx ls
-
-# skip build if image already exists
-if [[ -z $(docker manifest inspect $REGISTRY/$REPO:$BUILDKITE_COMMIT) ]]; then
-  echo "Image not found, proceeding with build..."
-else
-  echo "Image found"
-  exit 0
-fi
-
-if [[ "${VLLM_USE_PRECOMPILED:-0}" == "1" ]]; then
-  merge_base_commit_build_args="--build-arg VLLM_MERGE_BASE_COMMIT=${VLLM_MERGE_BASE_COMMIT}"
-else
-  merge_base_commit_build_args=""
-fi
-
-# build
-docker buildx build --file docker/Dockerfile \
-  --build-arg max_jobs=16 \
-  --build-arg buildkite_commit=$BUILDKITE_COMMIT \
-  --build-arg USE_SCCACHE=1 \
-  --build-arg TORCH_CUDA_ARCH_LIST="8.0 8.9 9.0 10.0" \
-  --build-arg FI_TORCH_CUDA_ARCH_LIST="8.0 8.9 9.0a 10.0a" \
-  --build-arg VLLM_USE_PRECOMPILED="${VLLM_USE_PRECOMPILED:-0}" \
-  ${merge_base_commit_build_args} \
-  --cache-from type=registry,ref=${CACHE_FROM},mode=max \
-  --cache-to type=registry,ref=${CACHE_TO},mode=max \
-  --tag ${REGISTRY}/${REPO}:${BUILDKITE_COMMIT} \
-  $( [[ "${BRANCH}" == "main" ]] && echo "--tag ${REGISTRY}/${REPO}:latest" ) \
-  --push \
-  --target test \
-  --progress plain .
@@ -1,57 +0,0 @@
-group: Abuild
-steps:
-  - label: ":docker: Build image"
-    key: image-build
-    depends_on: []
-    commands:
-    - .buildkite/image_build/image_build.sh $REGISTRY $REPO $BUILDKITE_COMMIT $BRANCH $VLLM_USE_PRECOMPILED $VLLM_MERGE_BASE_COMMIT $CACHE_FROM $CACHE_TO
-    retry:
-      automatic:
-        - exit_status: -1  # Agent was lost
-          limit: 2
-        - exit_status: -10  # Agent was lost
-          limit: 2
-
-  - label: ":docker: Build CPU image"
-    key: image-build-cpu
-    depends_on: []
-    commands:
-    - .buildkite/image_build/image_build_cpu.sh $REGISTRY $REPO $BUILDKITE_COMMIT
-    env:
-      DOCKER_BUILDKIT: "1"
-    retry:
-      automatic:
-        - exit_status: -1  # Agent was lost
-          limit: 2
-        - exit_status: -10  # Agent was lost
-          limit: 2
-
-  - label: ":docker: Build HPU image"
-    soft_fail: true
-    depends_on: []
-    key: image-build-hpu
-    commands:
-    - .buildkite/image_build/image_build_hpu.sh $REGISTRY $REPO $BUILDKITE_COMMIT
-    env:
-      DOCKER_BUILDKIT: "1"
-    retry:
-      automatic:
-        - exit_status: -1  # Agent was lost
-          limit: 2
-        - exit_status: -10  # Agent was lost
-          limit: 2
-  
-  - label: ":docker: Build CPU arm64 image"
-    key: cpu-arm64-image-build
-    depends_on: []
-    optional: true
-    commands:
-    - .buildkite/image_build/image_build_cpu_arm64.sh $REGISTRY $REPO $BUILDKITE_COMMIT
-    env:
-      DOCKER_BUILDKIT: "1"
-    retry:
-      automatic:
-        - exit_status: -1  # Agent was lost
-          limit: 2
-        - exit_status: -10  # Agent was lost
-          limit: 2
@@ -1,36 +0,0 @@
-#!/bin/bash
-set -e
-
-if [[ $# -lt 3 ]]; then
-  echo "Usage: $0 <registry> <repo> <commit>"
-  exit 1
-fi
-
-REGISTRY=$1
-REPO=$2
-BUILDKITE_COMMIT=$3
-
-# authenticate with AWS ECR
-aws ecr-public get-login-password --region us-east-1 | docker login --username AWS --password-stdin $REGISTRY
-
-# skip build if image already exists
-if [[ -z $(docker manifest inspect $REGISTRY/$REPO:$BUILDKITE_COMMIT-cpu) ]]; then
-  echo "Image not found, proceeding with build..."
-else
-  echo "Image found"
-  exit 0
-fi
-
-# build
-docker build --file docker/Dockerfile.cpu \
-  --build-arg max_jobs=16 \
-  --build-arg buildkite_commit=$BUILDKITE_COMMIT \
-  --build-arg VLLM_CPU_AVX512BF16=true \
-  --build-arg VLLM_CPU_AVX512VNNI=true \
-  --build-arg VLLM_CPU_AMXBF16=true \
-  --tag $REGISTRY/$REPO:$BUILDKITE_COMMIT-cpu \
-  --target vllm-test \
-  --progress plain .
-
-# push
-docker push $REGISTRY/$REPO:$BUILDKITE_COMMIT-cpu
@@ -1,33 +0,0 @@
-#!/bin/bash
-set -e
-
-if [[ $# -lt 3 ]]; then
-  echo "Usage: $0 <registry> <repo> <commit>"
-  exit 1
-fi
-
-REGISTRY=$1
-REPO=$2
-BUILDKITE_COMMIT=$3
-
-# authenticate with AWS ECR
-aws ecr-public get-login-password --region us-east-1 | docker login --username AWS --password-stdin $REGISTRY
-
-# skip build if image already exists
-if [[ -z $(docker manifest inspect $REGISTRY/$REPO:$BUILDKITE_COMMIT-cpu) ]]; then
-  echo "Image not found, proceeding with build..."
-else
-  echo "Image found"
-  exit 0
-fi
-
-# build
-docker build --file docker/Dockerfile.cpu \
-  --build-arg max_jobs=16 \
-  --build-arg buildkite_commit=$BUILDKITE_COMMIT \
-  --tag $REGISTRY/$REPO:$BUILDKITE_COMMIT-cpu \
-  --target vllm-test \
-  --progress plain .
-
-# push
-docker push $REGISTRY/$REPO:$BUILDKITE_COMMIT-cpu
@@ -1,34 +0,0 @@
-#!/bin/bash
-set -e
-
-if [[ $# -lt 3 ]]; then
-  echo "Usage: $0 <registry> <repo> <commit>"
-  exit 1
-fi
-
-REGISTRY=$1
-REPO=$2
-BUILDKITE_COMMIT=$3
-
-# authenticate with AWS ECR
-aws ecr-public get-login-password --region us-east-1 | docker login --username AWS --password-stdin $REGISTRY
-
-# skip build if image already exists
-if [[ -z $(docker manifest inspect $REGISTRY/$REPO:$BUILDKITE_COMMIT-hpu) ]]; then
-  echo "Image not found, proceeding with build..."
-else
-  echo "Image found"
-  exit 0
-fi
-
-# build
-docker build \
-  --file tests/pytorch_ci_hud_benchmark/Dockerfile.hpu \
-  --build-arg max_jobs=16 \
-  --build-arg buildkite_commit=$BUILDKITE_COMMIT \
-  --tag $REGISTRY/$REPO:$BUILDKITE_COMMIT-hpu \
-  --progress plain \
-  https://github.com/vllm-project/vllm-gaudi.git
-
-# push
-docker push $REGISTRY/$REPO:$BUILDKITE_COMMIT-hpu
@@ -8,4 +8,3 @@ tasks:
    value: 0.80
 limit: 250 # will run on 250 * 14 subjects = 3500 samples
 num_fewshot: 5
-rtol: 0.05
@@ -0,0 +1,12 @@
+# For vllm script, with -t option (tensor parallel size).
+# bash .buildkite/lm-eval-harness/run-lm-eval-gsm-vllm-baseline.sh -m nm-testing/Qwen2-1.5B-Instruct-W8A16-Channelwise -b "auto" -l 1000 -f 5 -t 1
+model_name: "nm-testing/Qwen2-1.5B-Instruct-W8A16-Channelwise"
+tasks:
+- name: "gsm8k"
+  metrics:
+  - name: "exact_match,strict-match"
+    value: 0.595
+  - name: "exact_match,flexible-extract"
+    value: 0.582
+limit: 1000
+num_fewshot: 5
@@ -1 +0,0 @@
-Meta-Llama-4-Maverick-17B-128E-Instruct-FP8.yaml
@@ -2,7 +2,7 @@
 # We can use this script to compute baseline accuracy on chartqa for vllm.
 #
 # Make sure you have lm-eval-harness installed:
-#   pip install "lm-eval[api]>=0.4.9.2"
+#   pip install lm-eval==0.4.9

 usage() {
    echo``
@@ -2,7 +2,7 @@
 # We can use this script to compute baseline accuracy on GSM for transformers.
 #
 # Make sure you have lm-eval-harness installed:
-#   pip install "lm-eval[api]>=0.4.9.2"
+#   pip install git+https://github.com/EleutherAI/lm-evaluation-harness.git@206b7722158f58c35b7ffcd53b035fdbdda5126d#egg=lm-eval[api]

 usage() {
    echo``
@@ -3,7 +3,7 @@
 # We use this for fp8, which HF does not support.
 #
 # Make sure you have lm-eval-harness installed:
-#   pip install "lm-eval[api]>=0.4.9.2"
+#   pip install git+https://github.com/EleutherAI/lm-evaluation-harness.git@206b7722158f58c35b7ffcd53b035fdbdda5126d#egg=lm-eval[api]

 usage() {
    echo``
@@ -3,7 +3,7 @@
 # We use this for fp8, which HF does not support.
 #
 # Make sure you have lm-eval-harness installed:
-#   pip install "lm-eval[api]>=0.4.9.2"
+#   pip install git+https://github.com/EleutherAI/lm-evaluation-harness.git@206b7722158f58c35b7ffcd53b035fdbdda5126d#egg=lm-eval[api]

 usage() {
    echo``
@@ -9,40 +9,11 @@ pytest -s -v test_lm_eval_correctness.py \
    --tp-size=1
 """

-import os
-from contextlib import contextmanager
-
 import lm_eval
 import numpy as np
 import yaml

-DEFAULT_RTOL = 0.08
-
-
-@contextmanager
-def scoped_env_vars(new_env: dict[str, str]):
-    if not new_env:
-        # Fast path: nothing to do
-        yield
-        return
-
-    old_values = {}
-    new_keys = []
-
-    try:
-        for key, value in new_env.items():
-            if key in os.environ:
-                old_values[key] = os.environ[key]
-            else:
-                new_keys.append(key)
-            os.environ[key] = str(value)
-        yield
-    finally:
-        # Restore / clean up
-        for key, value in old_values.items():
-            os.environ[key] = value
-        for key in new_keys:
-            os.environ.pop(key, None)
+RTOL = 0.08


 def launch_lm_eval(eval_config, tp_size):
@@ -61,26 +32,23 @@ def launch_lm_eval(eval_config, tp_size):
        f"trust_remote_code={trust_remote_code},"
        f"max_model_len={max_model_len},"
    )
-
-    env_vars = eval_config.get("env_vars", None)
-    with scoped_env_vars(env_vars):
-        results = lm_eval.simple_evaluate(
-            model=backend,
-            model_args=model_args,
-            tasks=[task["name"] for task in eval_config["tasks"]],
-            num_fewshot=eval_config["num_fewshot"],
-            limit=eval_config["limit"],
-            # TODO(yeq): using chat template w/ fewshot_as_multiturn is supposed help
-            # text models. however, this is regressing measured strict-match for
-            # existing text models in CI, so only apply it for mm, or explicitly set
-            apply_chat_template=eval_config.get(
-                "apply_chat_template", backend == "vllm-vlm"
-            ),
-            fewshot_as_multiturn=eval_config.get("fewshot_as_multiturn", False),
-            # Forward decoding and early-stop controls (e.g., max_gen_toks, until=...)
-            gen_kwargs=eval_config.get("gen_kwargs"),
-            batch_size=batch_size,
-        )
+    results = lm_eval.simple_evaluate(
+        model=backend,
+        model_args=model_args,
+        tasks=[task["name"] for task in eval_config["tasks"]],
+        num_fewshot=eval_config["num_fewshot"],
+        limit=eval_config["limit"],
+        # TODO(yeq): using chat template w/ fewshot_as_multiturn is supposed help
+        # text models. however, this is regressing measured strict-match for
+        # existing text models in CI, so only apply it for mm, or explicitly set
+        apply_chat_template=eval_config.get(
+            "apply_chat_template", backend == "vllm-vlm"
+        ),
+        fewshot_as_multiturn=eval_config.get("fewshot_as_multiturn", False),
+        # Forward decoding and early-stop controls (e.g., max_gen_toks, until=...)
+        gen_kwargs=eval_config.get("gen_kwargs"),
+        batch_size=batch_size,
+    )
    return results


@@ -89,8 +57,6 @@ def test_lm_eval_correctness_param(config_filename, tp_size):

    results = launch_lm_eval(eval_config, tp_size)

-    rtol = eval_config.get("rtol", DEFAULT_RTOL)
-
    success = True
    for task in eval_config["tasks"]:
        for metric in task["metrics"]:
@@ -98,9 +64,8 @@ def test_lm_eval_correctness_param(config_filename, tp_size):
            measured_value = results["results"][task["name"]][metric["name"]]
            print(
                f"{task['name']} | {metric['name']}: "
-                f"ground_truth={ground_truth:.3f} | "
-                f"measured={measured_value:.3f} | rtol={rtol}"
+                f"ground_truth={ground_truth} | measured={measured_value}"
            )
-            success = success and np.isclose(ground_truth, measured_value, rtol=rtol)
+            success = success and np.isclose(ground_truth, measured_value, rtol=RTOL)

    assert success
@@ -7,7 +7,7 @@ vLLM also maintains a continuous performance benchmark under [perf.vllm.ai](http

 ## Performance benchmark quick overview

-**Benchmarking Coverage**: latency, throughput and fix-qps serving on B200, A100, H100, Intel® Xeon® Processors, Intel® Gaudi® 3 Accelerators and Arm® Neoverse™ with different models.
+**Benchmarking Coverage**: latency, throughput and fix-qps serving on B200, A100, H100, Intel® Xeon® Processors and Intel® Gaudi® 3 Accelerators with different models.

 **Benchmarking Duration**: about 1hr.

@@ -23,7 +23,7 @@ bash .buildkite/performance-benchmarks/scripts/run-performance-benchmarks.sh

 Runtime environment variables:

- `ON_CPU`: set the value to '1' on Intel® Xeon® and Arm® Neoverse™ Processors. Default value is 0.
+- `ON_CPU`: set the value to '1' on Intel® Xeon® Processors. Default value is 0.
 - `SERVING_JSON`: JSON file to use for the serving tests. Default value is empty string (use default file).
 - `LATENCY_JSON`: JSON file to use for the latency tests. Default value is empty string (use default file).
 - `THROUGHPUT_JSON`: JSON file to use for the throughout tests. Default value is empty string (use default file).
@@ -34,9 +34,8 @@ Runtime environment variables:

 See [performance-benchmarks-descriptions.md](performance-benchmarks-descriptions.md) for detailed descriptions, and use `tests/latency-tests.json`, `tests/throughput-tests.json`, `tests/serving-tests.json` to configure the test cases.
 > NOTE: For Intel® Xeon® Processors, use `tests/latency-tests-cpu.json`, `tests/throughput-tests-cpu.json`, `tests/serving-tests-cpu.json` instead.
-> For Intel® Gaudi® 3 Accelerators, use `tests/latency-tests-hpu.json`, `tests/throughput-tests-hpu.json`, `tests/serving-tests-hpu.json` instead.
-> For Arm® Neoverse™, use `tests/latency-tests-arm64-cpu.json`, `tests/throughput-tests-arm64-cpu.json`, `tests/serving-tests-arm64-cpu.json` instead.
-
+For Intel® Gaudi® 3 Accelerators, use `tests/latency-tests-hpu.json`, `tests/throughput-tests-hpu.json`, `tests/serving-tests-hpu.json` instead.
+>
 ### Latency test

 Here is an example of one test inside `latency-tests.json`:
@@ -109,65 +108,6 @@ The number of this test is less stable compared to the delay and latency benchma

 WARNING: The benchmarking script will save json results by itself, so please do not configure `--save-results` or other results-saving-related parameters in `serving-tests.json`.

-#### Default Parameters Field
-
-We can specify default parameters in a JSON field with key `defaults`. Parameters defined in the field are applied globally to all serving tests, and can be overridden in test case fields. Here is an example:
-
-<details>
-<summary> An Example of default parameters field </summary>
-
-```json
-{
-  "defaults": {
-    "qps_list": [
-      "inf"
-    ],
-    "server_environment_variables": {
-      "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1
-    },
-    "server_parameters": {
-      "tensor_parallel_size": 1,
-      "dtype": "bfloat16",
-      "block_size": 128,
-      "disable_log_stats": "",
-      "load_format": "dummy"
-    },
-    "client_parameters": {
-      "backend": "vllm",
-      "dataset_name": "random",
-      "random-input-len": 128,
-      "random-output-len": 128,
-      "num_prompts": 200,
-      "ignore-eos": ""
-    }
-  },
-  "tests": [
-    {
-      "test_name": "serving_llama3B_tp2_random_128_128",
-      "server_parameters": {
-        "model": "meta-llama/Llama-3.2-3B-Instruct",
-        "tensor_parallel_size": 2,
-      },
-      "client_parameters": {
-        "model": "meta-llama/Llama-3.2-3B-Instruct",
-      }
-    },
-    {
-      "test_name": "serving_qwen3_tp4_random_128_128",
-      "server_parameters": {
-        "model": "Qwen/Qwen3-14B",
-        "tensor_parallel_size": 4,
-      },
-      "client_parameters": {
-        "model": "Qwen/Qwen3-14B",
-      }
-    },
-  ]
-}
-```
-
-</details>
-
 ### Visualizing the results

 The `convert-results-json-to-markdown.py` helps you put the benchmarking results inside a markdown table, by formatting [descriptions.md](performance-benchmarks-descriptions.md) with real benchmarking results.
@@ -176,6 +116,19 @@ If you do not see the table, please wait till the benchmark finish running.
 The json version of the table (together with the json version of the benchmark) will be also attached to the markdown file.
 The raw benchmarking results (in the format of json files) are in the `Artifacts` tab of the benchmarking.

-#### Performance Results Comparison  
+The `compare-json-results.py` helps to compare benchmark results JSON files converted using `convert-results-json-to-markdown.py`.
+When run, benchmark script generates results under `benchmark/results` folder, along with the `benchmark_results.md` and `benchmark_results.json`.
+`compare-json-results.py` compares two `benchmark_results.json` files and provides performance ratio e.g. for Output Tput, Median TTFT and Median TPOT.  
+If only one benchmark_results.json is passed, `compare-json-results.py` compares different TP and PP configurations in the benchmark_results.json instead.

-Follow the instructions in [performance results comparison](https://docs.vllm.ai/en/latest/benchmarking/dashboard/#performance-results-comparison) to analyze performance results and the sizing guide.
+Here is an example using the script to compare result_a and result_b with Model, Dataset name, input/output length, max concurrency and qps.
+`python3 compare-json-results.py -f results_a/benchmark_results.json -f results_b/benchmark_results.json`
+
+|   | Model | Dataset Name | Input Len | Output Len | # of max concurrency | qps  | results_a/benchmark_results.json | results_b/benchmark_results.json | perf_ratio        |
+|----|---------------------------------------|--------|-----|-----|------|-----|-----------|----------|----------|
+| 0  | meta-llama/Meta-Llama-3.1-8B-Instruct | random | 128 | 128 | 1000 | 1 | 142.633982                             | 156.526018                             | 1.097396 |
+| 1  | meta-llama/Meta-Llama-3.1-8B-Instruct | random | 128 | 128 | 1000 | inf| 241.620334                             | 294.018783                             | 1.216863 |
+
+A comparison diagram will be generated below the table.
+Here is an example to compare between 96c/results_gnr_96c_091_tp2pp3 and 128c/results_gnr_128c_091_tp2pp3
+<img width="1886" height="828" alt="image" src="https://github.com/user-attachments/assets/c02a43ef-25d0-4fd6-90e5-2169a28682dd" />
@@ -1,13 +1,8 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-
-from __future__ import annotations
-
 import argparse
-import html as _html
 import json
 import os
-from dataclasses import dataclass
 from importlib import util

 import pandas as pd
@@ -15,49 +10,27 @@ import pandas as pd
 pd.options.display.float_format = "{:.2f}".format
 plotly_found = util.find_spec("plotly.express") is not None

-DEFAULT_INFO_COLS = [
-    "Model",
-    "Dataset Name",
-    "Input Len",
-    "Output Len",
-    #    "TP Size",
-    #    "PP Size",
-    "# of max concurrency.",
-    "qps",
-]

-# Safety net: if any DataFrame leaks into to_html(), keep precision at 2.
-pd.set_option("display.precision", 2)
-pd.set_option("display.float_format", lambda x: f"{x:.2f}")
-
-
-# -----------------------------
-# Core data compare
-# -----------------------------
 def compare_data_columns(
-    files: list[str],
-    name_column: str,
-    data_column: str,
-    info_cols: list[str],
-    drop_column: str,
-    debug: bool = False,
+    files, name_column, data_column, info_cols, drop_column, debug=False
 ):
    """
    Align concatenation by keys derived from info_cols instead of row order.
    - Pick one canonical key list: subset of info_cols present in ALL files.
    - For each file: set index to those keys, aggregate duplicates
-      (mean for metric, first for names).
+    - (mean for metric, first for names).
    - Concat along axis=1 (indexes align), then reset_index so callers can
-      group by columns.
+    - group by columns.
    - If --debug, add a <file_label>_name column per file.
    """
    print("\ncompare_data_column:", data_column)

    frames = []
-    raw_data_cols: list[str] = []
+    raw_data_cols = []
    compare_frames = []

-    cols_per_file: list[set] = []
+    # 1) choose a canonical key list from info_cols that exists in ALL files
+    cols_per_file = []
    for f in files:
        try:
            df_tmp = pd.read_json(f, orient="records")
@@ -67,20 +40,24 @@ def compare_data_columns(

    key_cols = [c for c in info_cols if all(c in cset for cset in cols_per_file)]
    if not key_cols:
+        # soft fallback: use any info_cols present in the first file
        key_cols = [c for c in info_cols if c in list(cols_per_file[0])]
    if not key_cols:
        raise ValueError(
            "No common key columns found from info_cols across the input files."
        )

+    # 2) build a single "meta" block (keys as columns) once, aligned by the key index
    meta_added = False

    for file in files:
        df = pd.read_json(file, orient="records")

+        # Keep rows that actually have the compared metric (same as original behavior)
        if drop_column in df.columns:
            df = df.dropna(subset=[drop_column], ignore_index=True)

+        # Stabilize numeric key columns (harmless if missing)
        for c in (
            "Input Len",
            "Output Len",
@@ -92,26 +69,32 @@ def compare_data_columns(
            if c in df.columns:
                df[c] = pd.to_numeric(df[c], errors="coerce")

+        # Ensure all key columns exist
        for c in key_cols:
            if c not in df.columns:
                df[c] = pd.NA

+        # Set index = key_cols and aggregate duplicates → unique MultiIndex
        df_idx = df.set_index(key_cols, drop=False)

+        # meta (key columns), unique per key
        meta = df_idx[key_cols]
        if not meta.index.is_unique:
            meta = meta.groupby(level=key_cols, dropna=False).first()

+        # metric series for this file, aggregated to one row per key
        file_label = "/".join(file.split("/")[:-1]) or os.path.basename(file)
        s = df_idx[data_column]
        if not s.index.is_unique:
            s = s.groupby(level=key_cols, dropna=False).mean()
-        s.name = file_label
+        s.name = file_label  # column label like original

+        # add meta once (from first file) so keys are the leftmost columns
        if not meta_added:
            frames.append(meta)
            meta_added = True

+        # (NEW) debug: aligned test-name column per file
        if debug and name_column in df_idx.columns:
            name_s = df_idx[name_column]
            if not name_s.index.is_unique:
@@ -123,19 +106,26 @@ def compare_data_columns(
        raw_data_cols.append(file_label)
        compare_frames.append(s)

+        # Generalize ratio: for any file N>=2, add ratio (fileN / file1)
        if len(compare_frames) >= 2:
            base = compare_frames[0]
            current = compare_frames[-1]
            if "P99" in data_column or "Median" in data_column:
-                ratio = base / current
+                ratio = base / current  # for latency
            else:
                ratio = current / base
-            ratio = ratio.mask(base == 0)
+            ratio = ratio.mask(base == 0)  # avoid inf when baseline is 0
            ratio.name = f"Ratio 1 vs {len(compare_frames)}"
            frames.append(ratio)

-    concat_df = pd.concat(frames, axis=1).reset_index(drop=True)
+    # 4) concat on columns with aligned MultiIndex;
+    # then reset_index to return keys as columns
+    concat_df = pd.concat(frames, axis=1)
+    concat_df = concat_df.reset_index(drop=True).reset_index()
+    if "index" in concat_df.columns:
+        concat_df = concat_df.drop(columns=["index"])

+    # Ensure key/info columns appear first (in your info_cols order)
    front = [c for c in info_cols if c in concat_df.columns]
    rest = [c for c in concat_df.columns if c not in front]
    concat_df = concat_df[front + rest]
@@ -144,15 +134,20 @@ def compare_data_columns(
    return concat_df, raw_data_cols


-# -----------------------------
-# Split helper
-# -----------------------------
 def split_json_by_tp_pp(
    input_file: str = "benchmark_results.json", output_root: str = "."
 ) -> list[str]:
+    """
+    Split a benchmark JSON into separate folders by (TP Size, PP Size).
+
+    Creates: <output_root>/tp{TP}_pp{PP}/benchmark_results.json
+    Returns: list of file paths written.
+    """
+    # Load JSON data into DataFrame
    with open(input_file, encoding="utf-8") as f:
        data = json.load(f)

+    # If the JSON is a dict with a list under common keys, use that list
    if isinstance(data, dict):
        for key in ("results", "serving_results", "benchmarks", "data"):
            if isinstance(data.get(key), list):
@@ -161,6 +156,7 @@ def split_json_by_tp_pp(

    df = pd.DataFrame(data)

+    # Keep only "serving" tests
    name_col = next(
        (c for c in ["Test name", "test_name", "Test Name"] if c in df.columns), None
    )
@@ -169,6 +165,7 @@ def split_json_by_tp_pp(
            df[name_col].astype(str).str.contains(r"serving", case=False, na=False)
        ].copy()

+    # Handle alias column names
    rename_map = {
        "tp_size": "TP Size",
        "tensor_parallel_size": "TP Size",
@@ -179,14 +176,21 @@ def split_json_by_tp_pp(
        columns={k: v for k, v in rename_map.items() if k in df.columns}, inplace=True
    )

+    # Ensure TP/PP columns exist (default to 1 if missing)
    if "TP Size" not in df.columns:
        df["TP Size"] = 1
    if "PP Size" not in df.columns:
        df["PP Size"] = 1

-    df["TP Size"] = pd.to_numeric(df["TP Size"], errors="coerce").fillna(1).astype(int)
-    df["PP Size"] = pd.to_numeric(df["PP Size"], errors="coerce").fillna(1).astype(int)
+    # make sure TP/PP are numeric ints with no NaN
+    df["TP Size"] = (
+        pd.to_numeric(df.get("TP Size", 1), errors="coerce").fillna(1).astype(int)
+    )
+    df["PP Size"] = (
+        pd.to_numeric(df.get("PP Size", 1), errors="coerce").fillna(1).astype(int)
+    )

+    # Split into separate folders
    saved_paths: list[str] = []
    for (tp, pp), group_df in df.groupby(["TP Size", "PP Size"], dropna=False):
        folder_name = os.path.join(output_root, f"tp{int(tp)}_pp{int(pp)}")
@@ -199,9 +203,32 @@ def split_json_by_tp_pp(
    return saved_paths


-# -----------------------------
-# Styling helpers
-# -----------------------------
+def _add_limit_line(fig, y_value, label):
+    # Visible dashed line + annotation
+    fig.add_hline(
+        y=y_value,
+        line_dash="dash",
+        line_color="red" if "ttft" in label.lower() else "blue",
+        annotation_text=f"{label}: {y_value} ms",
+        annotation_position="top left",
+    )
+    # Optional: add a legend item (as a transparent helper trace)
+    if plot and plotly_found:
+        import plotly.graph_objects as go
+
+        fig.add_trace(
+            go.Scatter(
+                x=[None],
+                y=[None],
+                mode="lines",
+                line=dict(
+                    dash="dash", color="red" if "ttft" in label.lower() else "blue"
+                ),
+                name=f"{label}",
+            )
+        )
+
+
 def _find_concurrency_col(df: pd.DataFrame) -> str:
    for c in [
        "# of max concurrency.",
@@ -212,6 +239,7 @@ def _find_concurrency_col(df: pd.DataFrame) -> str:
    ]:
        if c in df.columns:
            return c
+    # Fallback: guess an integer-like column (harmless if unused)
    for c in df.columns:
        if df[c].dtype.kind in "iu" and df[c].nunique() > 1 and df[c].min() >= 1:
            return c
@@ -220,7 +248,8 @@ def _find_concurrency_col(df: pd.DataFrame) -> str:

 def _highlight_threshold(
    df: pd.DataFrame, threshold: float
-) -> pd.io.formats.style.Styler:
+) -> "pd.io.formats.style.Styler":
+    """Highlight numeric per-configuration columns with value <= threshold."""
    conc_col = _find_concurrency_col(df)
    key_cols = [
        c
@@ -231,7 +260,6 @@ def _highlight_threshold(
        c for c in df.columns if c not in key_cols and not str(c).startswith("Ratio")
    ]
    conf_cols = [c for c in conf_cols if pd.api.types.is_numeric_dtype(df[c])]
-
    return df.style.map(
        lambda v: "background-color:#e6ffe6;font-weight:bold;"
        if pd.notna(v) and v <= threshold
@@ -240,264 +268,7 @@ def _highlight_threshold(
    )


-def highlight_ratio_columns(styler: pd.io.formats.style.Styler):
-    ratio_cols = [c for c in styler.data.columns if "ratio" in str(c).lower()]
-    if not ratio_cols:
-        return styler
-
-    styler = styler.apply(
-        lambda _: ["background-color: #fff3b0"] * len(styler.data),
-        subset=ratio_cols,
-        axis=0,
-    )
-
-    styler = styler.set_table_styles(
-        [
-            {
-                "selector": f"th.col_heading.level0.col{i}",
-                "props": [("background-color", "#fff3b0")],
-            }
-            for i, col in enumerate(styler.data.columns)
-            if col in ratio_cols
-        ],
-        overwrite=False,
-    )
-    return styler
-
-
-def _apply_two_decimals(
-    styler: pd.io.formats.style.Styler,
-) -> pd.io.formats.style.Styler:
-    df = styler.data
-    num_cols = df.select_dtypes("number").columns
-    if len(num_cols) == 0:
-        return styler
-    return styler.format({c: "{:.2f}" for c in num_cols}, na_rep="")
-
-
-# -----------------------------
-# Valid max concurrency summary helpers
-# -----------------------------
-def _config_value_columns(df: pd.DataFrame, conc_col: str) -> list[str]:
-    key_cols = [
-        c
-        for c in ["Model", "Dataset Name", "Input Len", "Output Len"]
-        if c in df.columns
-    ]
-    exclude = set(key_cols + [conc_col, "qps", "QPS"])
-
-    cols: list[str] = []
-    for c in df.columns:
-        if c in exclude:
-            continue
-        lc = str(c).lower()
-        if lc.startswith("ratio"):
-            continue
-        if lc.endswith("_name") or lc == "test name" or lc == "test_name":
-            continue
-        if pd.api.types.is_numeric_dtype(df[c]):
-            cols.append(c)
-    return cols
-
-
-def _max_concurrency_ok(
-    df: pd.DataFrame, conc_col: str, cfg_col: str, threshold: float
-):
-    if df is None or conc_col not in df.columns or cfg_col not in df.columns:
-        return pd.NA
-
-    d = df[[conc_col, cfg_col]].copy()
-    d[conc_col] = pd.to_numeric(d[conc_col], errors="coerce")
-    d[cfg_col] = pd.to_numeric(d[cfg_col], errors="coerce")
-    d = d.dropna(subset=[conc_col, cfg_col])
-
-    if d.empty:
-        return pd.NA
-
-    ok = d[d[cfg_col] <= threshold]
-    if ok.empty:
-        return pd.NA
-
-    return ok[conc_col].max()
-
-
-def _value_at_concurrency(df: pd.DataFrame, conc_col: str, cfg_col: str, conc_value):
-    if (
-        df is None
-        or conc_col not in df.columns
-        or cfg_col not in df.columns
-        or pd.isna(conc_value)
-    ):
-        return pd.NA
-
-    d = df[[conc_col, cfg_col]].copy()
-    d[conc_col] = pd.to_numeric(d[conc_col], errors="coerce")
-    d[cfg_col] = pd.to_numeric(d[cfg_col], errors="coerce")
-
-    conc_value = pd.to_numeric(conc_value, errors="coerce")
-    if pd.isna(conc_value):
-        return pd.NA
-
-    hit = d[d[conc_col] == conc_value]
-    if hit.empty:
-        return pd.NA
-    return hit[cfg_col].iloc[0]
-
-
-def build_valid_max_concurrency_summary_html(
-    tput_group_df: pd.DataFrame | None,
-    ttft_group_df: pd.DataFrame | None,
-    tpot_group_df: pd.DataFrame | None,
-    conc_col: str,
-    args,
-) -> str:
-    if ttft_group_df is None and tpot_group_df is None:
-        return ""
-
-    ttft_cols = (
-        _config_value_columns(ttft_group_df, conc_col)
-        if ttft_group_df is not None
-        else []
-    )
-    tpot_cols = (
-        _config_value_columns(tpot_group_df, conc_col)
-        if tpot_group_df is not None
-        else []
-    )
-    tput_cols = (
-        _config_value_columns(tput_group_df, conc_col)
-        if tput_group_df is not None
-        else []
-    )
-
-    if ttft_group_df is not None and tpot_group_df is not None:
-        cfg_cols = [c for c in ttft_cols if c in tpot_cols]
-        if tput_group_df is not None:
-            cfg_cols = [c for c in cfg_cols if c in tput_cols] or cfg_cols
-    else:
-        cfg_cols = ttft_cols or tpot_cols
-
-    if not cfg_cols:
-        cfg_cols = sorted(set(ttft_cols) | set(tpot_cols) | set(tput_cols), key=str)
-
-    rows = []
-    for cfg in cfg_cols:
-        ttft_max = (
-            _max_concurrency_ok(ttft_group_df, conc_col, cfg, args.ttft_max_ms)
-            if ttft_group_df is not None
-            else pd.NA
-        )
-        tpot_max = (
-            _max_concurrency_ok(tpot_group_df, conc_col, cfg, args.tpot_max_ms)
-            if tpot_group_df is not None
-            else pd.NA
-        )
-        both = (
-            pd.NA
-            if (pd.isna(ttft_max) or pd.isna(tpot_max))
-            else min(ttft_max, tpot_max)
-        )
-
-        tput_at_both = (
-            _value_at_concurrency(tput_group_df, conc_col, cfg, both)
-            if tput_group_df is not None
-            else pd.NA
-        )
-        ttft_at_both = (
-            _value_at_concurrency(ttft_group_df, conc_col, cfg, both)
-            if ttft_group_df is not None
-            else pd.NA
-        )
-        tpot_at_both = (
-            _value_at_concurrency(tpot_group_df, conc_col, cfg, both)
-            if tpot_group_df is not None
-            else pd.NA
-        )
-
-        rows.append(
-            {
-                "Configuration": cfg,
-                f"Max {conc_col} (TTFT ≤ {args.ttft_max_ms:g} ms)": ttft_max,
-                f"Max {conc_col} (TPOT ≤ {args.tpot_max_ms:g} ms)": tpot_max,
-                f"Max {conc_col} (Both)": both,
-                "Output Tput @ Both (tok/s)": tput_at_both,
-                "TTFT @ Both (ms)": ttft_at_both,
-                "TPOT @ Both (ms)": tpot_at_both,
-            }
-        )
-
-    summary_df = pd.DataFrame(rows)
-
-    # --- Coerce numeric columns so Styler doesn't miss them due to object dtype ---
-    for c in summary_df.columns:
-        if c == "Configuration":
-            continue
-        summary_df[c] = pd.to_numeric(summary_df[c], errors="coerce")
-
-    both_col = f"Max {conc_col} (Both)"
-
-    # --- Strict 2-decimal formatting for ALL non-Configuration columns ---
-    formatters = {}
-    for c in summary_df.columns:
-        if c == "Configuration":
-            continue
-        # default argument binds per-column formatter correctly
-        formatters[c] = lambda v: "" if pd.isna(v) else f"{float(v):.2f}"
-
-    styler = summary_df.style.format(formatters)
-
-    def _green(v):
-        return "background-color:#e6ffe6;font-weight:bold;" if pd.notna(v) else ""
-
-    if both_col in summary_df.columns:
-        styler = styler.map(_green, subset=[both_col])
-
-    title = (
-        '<div style="font-size: 1.15em; font-weight: 700; margin: 12px 0 6px 0;">'
-        "Valid Max Concurrency Summary"
-        "</div>\n"
-    )
-    return title + styler.to_html(table_attributes='border="1" class="dataframe"')
-
-
-# -----------------------------
-# Plot helper
-# -----------------------------
-def _add_limit_line(fig, y_value: float, label: str):
-    fig.add_hline(
-        y=y_value,
-        line_dash="dash",
-        line_color="red" if "ttft" in label.lower() else "blue",
-        annotation_text=f"{label}: {y_value} ms",
-        annotation_position="top left",
-    )
-    if plotly_found:
-        import plotly.graph_objects as go
-
-        fig.add_trace(
-            go.Scatter(
-                x=[None],
-                y=[None],
-                mode="lines",
-                line=dict(
-                    dash="dash",
-                    color="red" if "ttft" in label.lower() else "blue",
-                ),
-                name=label,
-            )
-        )
-
-
-# -----------------------------
-# Refactored main + group-first report
-# -----------------------------
-@dataclass(frozen=True)
-class MetricPlan:
-    data_cols: list[str]
-    drop_column: str
-
-
-def build_parser() -> argparse.ArgumentParser:
+if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument(
        "-f", "--file", action="append", type=str, help="input file name"
@@ -537,289 +308,149 @@ def build_parser() -> argparse.ArgumentParser:
        default=100.0,
        help="Reference limit for TPOT plots (ms)",
    )
-    return parser

+    args = parser.parse_args()

-def choose_metrics(latency: str) -> MetricPlan:
-    latency = (latency or "").lower()
    drop_column = "P99"
+    name_column = "Test name"
+    info_cols = [
+        "Model",
+        "Dataset Name",
+        "Input Len",
+        "Output Len",
+        "TP Size",
+        "PP Size",
+        "# of max concurrency.",
+        "qps",
+    ]

-    if "median" in latency:
-        return MetricPlan(
-            data_cols=["Output Tput (tok/s)", "Median TTFT (ms)", "Median"],
-            drop_column=drop_column,
-        )
-
-    return MetricPlan(
-        data_cols=["Output Tput (tok/s)", "P99 TTFT (ms)", "P99"],
-        drop_column=drop_column,
-    )
-
-
-def prepare_input_files(args, info_cols: list[str]) -> tuple[list[str], list[str]]:
-    if not args.file:
-        raise ValueError("No input files provided. Use -f/--file.")
+    if "median" in args.latency:
+        data_cols_to_compare = ["Output Tput (tok/s)", "Median TTFT (ms)", "Median"]
+        html_msgs_for_data_cols = [
+            "Compare Output Tokens /n",
+            "Median TTFT /n",
+            "Median TPOT /n",
+        ]
+        drop_column = "P99"
+    elif "p99" in args.latency:
+        data_cols_to_compare = ["Output Tput (tok/s)", "P99 TTFT (ms)", "P99"]
+        html_msgs_for_data_cols = [
+            "Compare Output Tokens /n",
+            "P99 TTFT /n",
+            "P99 TPOT /n",
+        ]

    if len(args.file) == 1:
        files = split_json_by_tp_pp(args.file[0], output_root="splits")
        info_cols = [c for c in info_cols if c not in ("TP Size", "PP Size")]
    else:
        files = args.file
-
-    return files, info_cols
-
-
-def get_y_axis_col(info_cols: list[str], xaxis: str) -> str:
-    y_axis_index = info_cols.index(xaxis) if xaxis in info_cols else 6
-    return info_cols[y_axis_index]
-
-
-def get_group_cols(output_df: pd.DataFrame, info_cols: list[str]) -> list[str]:
-    filtered_info_cols = info_cols[:4]
-    group_cols = [c for c in filtered_info_cols if c in output_df.columns]
-    if not group_cols:
-        raise ValueError(
-            f"No valid group-by columns. Expected subset: {filtered_info_cols}, "
-            f"but DataFrame has: {list(output_df.columns)}"
-        )
-    return group_cols
-
-
-def normalize_group_key(name):
-    return name if isinstance(name, tuple) else (name,)
-
-
-def group_filename(name, prefix: str = "perf_comparison_") -> str:
-    name_vals = normalize_group_key(name)
-    safe = ",".join(map(str, name_vals)).replace(",", "_").replace("/", "-")
-    return f"{prefix}{safe}.html"
-
-
-def build_group_suffix(group_cols: list[str], name) -> str:
-    name_vals = normalize_group_key(name)
-    return " , ".join(f"{col} : [ {val} ] " for col, val in zip(group_cols, name_vals))
-
-
-def render_metric_table_html(
-    display_group: pd.DataFrame,
-    metric_label: str,
-    group_suffix: str,
-    args,
-) -> str:
-    title = (
-        f'<div style="font-size: 1.25em; font-weight: 600; margin: 12px 0;">'
-        f"{_html.escape(metric_label)}"
-        f" — {_html.escape(group_suffix)}"
-        f"</div>\n"
-    )
-
-    metric_name = metric_label.lower()
-    if "ttft" in metric_name:
-        styler = _highlight_threshold(display_group, args.ttft_max_ms)
-    elif ("tpot" in metric_name) or ("median" in metric_name) or ("p99" in metric_name):
-        styler = _highlight_threshold(display_group, args.tpot_max_ms)
-    else:
-        styler = display_group.style
-
-    styler = _apply_two_decimals(styler)
-    styler = highlight_ratio_columns(styler)
-
-    return title + styler.to_html(table_attributes='border="1" class="dataframe"')
-
-
-def maybe_write_plot(
-    main_fh,
-    sub_fh,
-    group_df: pd.DataFrame,
-    raw_data_cols: list[str],
-    metric_label: str,
-    y_axis_col: str,
-    args,
-):
-    if not (args.plot and plotly_found):
-        return
-
-    import plotly.express as px
-
-    df = group_df[raw_data_cols].sort_values(by=y_axis_col)
-    df_melted = df.melt(
-        id_vars=y_axis_col,
-        var_name="Configuration",
-        value_name=metric_label,
-    )
-
-    fig = px.line(
-        df_melted,
-        x=y_axis_col,
-        y=metric_label,
-        color="Configuration",
-        title=f"{metric_label} vs {y_axis_col}",
-        markers=True,
-    )
-
-    # Ensure plot hover + y tick labels are also 2 decimals.
-    fig.update_traces(hovertemplate="%{y:.2f}<extra></extra>")
-    fig.update_yaxes(tickformat=".2f")
-
-    metric_name = metric_label.lower()
-    if "ttft" in metric_name:
-        _add_limit_line(fig, args.ttft_max_ms, "TTFT limit")
-    elif ("tpot" in metric_name) or ("median" in metric_name) or ("p99" in metric_name):
-        _add_limit_line(fig, args.tpot_max_ms, "TPOT limit")
-
-    html = fig.to_html(full_html=True, include_plotlyjs="cdn")
-    main_fh.write(html)
-    sub_fh.write(html)
-
-
-def build_group_keys(
-    df: pd.DataFrame, group_cols: list[str], sort_cols: list[str] | None = None
-):
-    if sort_cols:
-        df = df.sort_values(by=sort_cols)
-    gb = df.groupby(group_cols, dropna=False)
-    return [k for k, _ in gb]
-
-
-def write_report_group_first(
-    files: list[str], info_cols: list[str], plan: MetricPlan, args
-):
-    name_column = "Test name"
-    y_axis_col = get_y_axis_col(info_cols, args.xaxis)
-
    print("comparing : " + ", ".join(files))
-
-    metric_cache: dict[str, tuple[pd.DataFrame, list[str]]] = {}
-    group_cols_canonical: list[str] | None = None
-
-    for metric_label in plan.data_cols:
-        output_df, raw_data_cols = compare_data_columns(
-            files,
-            name_column,
-            metric_label,
-            info_cols,
-            plan.drop_column,
-            debug=args.debug,
-        )
-
-        raw_data_cols = list(raw_data_cols)
-        raw_data_cols.insert(0, y_axis_col)
-
-        group_cols = get_group_cols(output_df, info_cols)
-        if group_cols_canonical is None:
-            group_cols_canonical = group_cols
-        else:
-            group_cols_canonical = [c for c in group_cols_canonical if c in group_cols]
-
-        metric_cache[metric_label] = (
-            output_df.sort_values(by=args.xaxis),
-            raw_data_cols,
-        )
-
-    if not group_cols_canonical:
-        raise ValueError("No canonical group columns found across metrics.")
-
-    first_metric = plan.data_cols[0]
-    first_df_sorted, _ = metric_cache[first_metric]
-    group_keys = build_group_keys(
-        first_df_sorted, group_cols_canonical, sort_cols=[args.xaxis]
-    )
-
-    metric_groupbys = {
-        metric_label: df.groupby(group_cols_canonical, dropna=False)
-        for metric_label, (df, _) in metric_cache.items()
-    }
-
-    with open("perf_comparison.html", "w", encoding="utf-8") as main_fh:
-        main_fh.write('<meta charset="utf-8">\n')
-        for gkey in group_keys:
-            gkey_tuple = normalize_group_key(gkey)
-            suffix = build_group_suffix(group_cols_canonical, gkey_tuple)
-            sub_path = group_filename(gkey_tuple)
-            group_header = (
-                '<div style="font-size: 1.4em; font-weight: 700; '
-                'margin: 18px 0 10px 0;">'
-                f"{_html.escape(suffix)}"
-                "</div>\n"
+    debug = args.debug
+    plot = args.plot
+    # For Plot feature, assign y axis from one of info_cols
+    y_axis_index = info_cols.index(args.xaxis) if args.xaxis in info_cols else 6
+    with open("perf_comparison.html", "w") as text_file:
+        for i in range(len(data_cols_to_compare)):
+            output_df, raw_data_cols = compare_data_columns(
+                files,
+                name_column,
+                data_cols_to_compare[i],
+                info_cols,
+                drop_column,
+                debug=debug,
            )

-            main_fh.write(group_header)
-            with open(sub_path, "w", encoding="utf-8") as sub_fh:
-                sub_fh.write('<meta charset="utf-8">\n')
-                sub_fh.write(group_header)
-                tput_group_df = None
-                ttft_group_df = None
-                tpot_group_df = None
-                conc_col = args.xaxis
+            # For Plot feature, insert y axis from one of info_cols
+            raw_data_cols.insert(0, info_cols[y_axis_index])

-                for metric_label in plan.data_cols:
-                    gb = metric_groupbys[metric_label]
-                    df_sorted, raw_data_cols = metric_cache[metric_label]
+            filtered_info_cols = info_cols[:-2]
+            existing_group_cols = [
+                c for c in filtered_info_cols if c in output_df.columns
+            ]
+            if not existing_group_cols:
+                raise ValueError(
+                    f"No valid group-by columns  "
+                    f"Expected subset: {filtered_info_cols}, "
+                    f"but DataFrame has: {list(output_df.columns)}"
+                )
+            # output_df_sorted = output_df.sort_values(by=existing_group_cols)
+            output_df_sorted = output_df.sort_values(by=args.xaxis)
+            output_groups = output_df_sorted.groupby(existing_group_cols, dropna=False)
+            for name, group in output_groups:
+                group_name = (
+                    ",".join(map(str, name)).replace(",", "_").replace("/", "-")
+                )
+                group_html_name = "perf_comparison_" + group_name + ".html"

-                    try:
-                        group_df = gb.get_group(gkey)
-                    except KeyError:
-                        missing = (
-                            '<div style="font-size: 1.1em; font-weight: 600; '
-                            'margin: 10px 0;">'
-                            f"{_html.escape(metric_label)} — missing for this group"
-                            "</div>\n"
+                metric_name = str(data_cols_to_compare[i]).lower()
+                if "tok/s" in metric_name:
+                    html = group.to_html()
+                elif "ttft" in metric_name:
+                    styler = _highlight_threshold(group, args.ttft_max_ms).format(
+                        {c: "{:.2f}" for c in group.select_dtypes("number").columns},
+                        na_rep="—",
+                    )
+                    html = styler.to_html(
+                        table_attributes='border="1" class="dataframe"'
+                    )
+                elif (
+                    "tpot" in metric_name
+                    or "median" in metric_name
+                    or "p99" in metric_name
+                ):
+                    styler = _highlight_threshold(group, args.tpot_max_ms).format(
+                        {c: "{:.2f}" for c in group.select_dtypes("number").columns},
+                        na_rep="—",
+                    )
+                    html = styler.to_html(
+                        table_attributes='border="1" class="dataframe"'
+                    )
+
+                text_file.write(html_msgs_for_data_cols[i])
+                text_file.write(html)
+                with open(group_html_name, "a+") as sub_text_file:
+                    sub_text_file.write(html_msgs_for_data_cols[i])
+                    sub_text_file.write(html)
+
+                    if plot and plotly_found:
+                        import plotly.express as px
+
+                        df = group[raw_data_cols]
+                        df_sorted = df.sort_values(by=info_cols[y_axis_index])
+                        # Melt DataFrame for plotting
+                        df_melted = df_sorted.melt(
+                            id_vars=info_cols[y_axis_index],
+                            var_name="Configuration",
+                            value_name=data_cols_to_compare[i],
+                        )
+                        title = (
+                            data_cols_to_compare[i] + " vs " + info_cols[y_axis_index]
+                        )
+                        # Create Plotly line chart
+                        fig = px.line(
+                            df_melted,
+                            x=info_cols[y_axis_index],
+                            y=data_cols_to_compare[i],
+                            color="Configuration",
+                            title=title,
+                            markers=True,
                        )

-                        main_fh.write(missing)
-                        sub_fh.write(missing)
-                        continue
+                        # ---- Add threshold lines based on metric name ----
+                        if "ttft" in metric_name:
+                            _add_limit_line(fig, args.ttft_max_ms, "TTFT limit")
+                        elif (
+                            "tpot" in metric_name
+                            or "median" in metric_name
+                            or "p99" in metric_name
+                        ):
+                            _add_limit_line(fig, args.tpot_max_ms, "TPOT limit")

-                    if conc_col not in group_df.columns:
-                        conc_col = _find_concurrency_col(group_df)
-
-                    mn = metric_label.lower().strip()
-                    if "tok/s" in mn:
-                        tput_group_df = group_df
-                    elif "ttft" in mn:
-                        ttft_group_df = group_df
-                    elif mn in ("p99", "median") or "tpot" in mn:
-                        tpot_group_df = group_df
-
-                    display_group = group_df.drop(
-                        columns=group_cols_canonical, errors="ignore"
-                    )
-
-                    html = render_metric_table_html(
-                        display_group, metric_label, suffix, args
-                    )
-                    main_fh.write(html)
-                    sub_fh.write(html)
-
-                    maybe_write_plot(
-                        main_fh,
-                        sub_fh,
-                        group_df=group_df,
-                        raw_data_cols=raw_data_cols,
-                        metric_label=metric_label,
-                        y_axis_col=y_axis_col,
-                        args=args,
-                    )
-
-                summary_html = build_valid_max_concurrency_summary_html(
-                    tput_group_df=tput_group_df,
-                    ttft_group_df=ttft_group_df,
-                    tpot_group_df=tpot_group_df,
-                    conc_col=conc_col,
-                    args=args,
-                )
-                if summary_html:
-                    main_fh.write(summary_html)
-                    sub_fh.write(summary_html)
-
-
-def main():
-    args = build_parser().parse_args()
-    info_cols = list(DEFAULT_INFO_COLS)
-    plan = choose_metrics(args.latency)
-    files, info_cols = prepare_input_files(args, info_cols)
-    write_report_group_first(files, info_cols, plan, args)
-
-
-if __name__ == "__main__":
-    main()
+                        # Export to HTML
+                        text_file.write(
+                            fig.to_html(full_html=True, include_plotlyjs="cdn")
+                        )
+                        sub_text_file.write(
+                            fig.to_html(full_html=True, include_plotlyjs="cdn")
+                        )
@@ -49,11 +49,7 @@ check_cpus() {
    echo "Need at least 1 NUMA to run benchmarking."
    exit 1
  fi
-  if [[ "$(uname -m)" == "aarch64" ]] || [[ "$(uname -m)" == "arm64" ]]; then
-    declare -g gpu_type="arm64-cpu"
-  else
-    declare -g gpu_type="cpu"
-  fi
+  declare -g gpu_type="cpu"
  echo "GPU type is $gpu_type"
 }

@@ -114,8 +110,7 @@ json2envs() {
 wait_for_server() {
  # wait for vllm server to start
  # return 1 if vllm server crashes
-  local timeout_val="1200"
-  timeout "$timeout_val" bash -c '
+  timeout 1200 bash -c '
    until curl -X POST localhost:8000/v1/completions; do
      sleep 1
    done' && return 0 || return 1
@@ -211,8 +206,8 @@ run_latency_tests() {

    # check if there is enough GPU to run the test
    tp=$(echo "$latency_params" | jq -r '.tensor_parallel_size')
-    if [[ "$ON_CPU" == "1" ]]; then
-      pp=$(echo "$latency_params" | jq -r '.pipeline_parallel_size // 1')
+    if [ "$ON_CPU" == "1" ]; then
+      pp=$(echo "$latency_params" | jq -r '.pipeline_parallel_size')
      world_size=$(($tp*$pp))
      if [[ $numa_count -lt $world_size  && -z "${REMOTE_HOST}" ]]; then
        echo "Required world-size $world_size but only $numa_count NUMA nodes found. Skip testcase $test_name."
@@ -280,8 +275,8 @@ run_throughput_tests() {

    # check if there is enough GPU to run the test
    tp=$(echo "$throughput_params" | jq -r '.tensor_parallel_size')
-    if [[ "$ON_CPU" == "1" ]]; then
-      pp=$(echo "$throughput_params" | jq -r '.pipeline_parallel_size // 1')
+    if [ "$ON_CPU" == "1" ]; then
+      pp=$(echo "$throughput_params" | jq -r '.pipeline_parallel_size')
      world_size=$(($tp*$pp))
      if [[ $numa_count -lt $world_size  && -z "${REMOTE_HOST}" ]]; then
        echo "Required world-size $world_size but only $numa_count NUMA nodes found. Skip testcase $test_name."
@@ -321,44 +316,12 @@ run_throughput_tests() {
 run_serving_tests() {
  # run serving tests using `vllm bench serve` command
  # $1: a json file specifying serving test cases
-  #
-  # Supported JSON formats:
-  # 1) Plain format: top-level array
-  #    [ { "test_name": "...", "server_parameters": {...}, ... }, ... ]
-  #
-  # 2) Default parameters field + plain format tests
-  #    {
-  #      "defaults": { ... },
-  #      "tests": [ { "test_name": "...", "server_parameters": {...}, ... }, ... ]
-  #    }

  local serving_test_file
  serving_test_file=$1

  # Iterate over serving tests
-  jq -c '
-    if type == "array" then
-      # Plain format: test cases array
-      .[]
-    elif (type == "object" and has("tests")) then
-      # merge the default parameters into each test cases
-      . as $root
-      | ($root.defaults // {}) as $d
-      | ($root.tests // [])[]
-      # default qps / max_concurrency from defaults if missing
-      | .qps_list = (.qps_list // $d.qps_list)
-      | .max_concurrency_list = (.max_concurrency_list // $d.max_concurrency_list)
-      # merge envs / params: test overrides defaults
-      | .server_environment_variables =
-          (($d.server_environment_variables // {}) + (.server_environment_variables // {}))
-      | .server_parameters =
-          (($d.server_parameters // {}) + (.server_parameters // {}))
-      | .client_parameters =
-          (($d.client_parameters // {}) + (.client_parameters // {}))
-    else
-      error("Unsupported serving test file format: must be array or object with .tests")
-    end
-  ' "$serving_test_file" | while read -r params; do
+  jq -c '.[]' "$serving_test_file" | while read -r params; do
    # get the test name, and append the GPU type back to it.
    test_name=$(echo "$params" | jq -r '.test_name')
    if [[ ! "$test_name" =~ ^serving_ ]]; then
@@ -372,33 +335,28 @@ run_serving_tests() {
      continue
    fi

-    # get client and server arguments (after merged the default parameters)
+    # get client and server arguments
    server_params=$(echo "$params" | jq -r '.server_parameters')
    server_envs=$(echo "$params" | jq -r '.server_environment_variables')
    client_params=$(echo "$params" | jq -r '.client_parameters')
-
    server_args=$(json2args "$server_params")
    server_envs=$(json2envs "$server_envs")
    client_args=$(json2args "$client_params")
-
-    # qps_list
    qps_list=$(echo "$params" | jq -r '.qps_list')
    qps_list=$(echo "$qps_list" | jq -r '.[] | @sh')
    echo "Running over qps list $qps_list"
-
-    # max_concurrency_list (fallback to num_prompts if missing)
    max_concurrency_list=$(echo "$params" | jq -r '.max_concurrency_list')
    if [[ -z "$max_concurrency_list" || "$max_concurrency_list" == "null" ]]; then
-      num_prompts=$(echo "$client_params" | jq -r '.num_prompts')
-      max_concurrency_list="[$num_prompts]"
+        num_prompts=$(echo "$client_params" | jq -r '.num_prompts')
+        max_concurrency_list="[$num_prompts]"
    fi
    max_concurrency_list=$(echo "$max_concurrency_list" | jq -r '.[] | @sh')
    echo "Running over max concurrency list $max_concurrency_list"

    # check if there is enough resources to run the test
    tp=$(echo "$server_params" | jq -r '.tensor_parallel_size')
-    if [[ "$ON_CPU" == "1" ]]; then
-      pp=$(echo "$server_params" | jq -r '.pipeline_parallel_size // 1')
+    if [ "$ON_CPU" == "1" ]; then
+      pp=$(echo "$server_params" | jq -r '.pipeline_parallel_size')
      world_size=$(($tp*$pp))
      if [[ $numa_count -lt $world_size  && -z "${REMOTE_HOST}" ]]; then
        echo "Required world-size $world_size but only $numa_count NUMA nodes found. Skip testcase $test_name."
@@ -500,9 +458,9 @@ run_serving_tests() {
 main() {
  local ARCH
  ARCH=''
-  if [[ "$ON_CPU" == "1" ]]; then
-    check_cpus
-    ARCH="-$gpu_type"
+  if [ "$ON_CPU" == "1" ];then
+     check_cpus
+     ARCH='-cpu'
  else
     check_gpus
     ARCH="$arch_suffix"
@@ -1,26 +0,0 @@
-[
-    {
-        "test_name": "latency_llama8B_tp1",
-        "environment_variables": {
-            "VLLM_RPC_TIMEOUT": 100000,
-            "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
-            "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
-            "VLLM_CPU_KVCACHE_SPACE": 40
-        },
-        "parameters": {
-            "model": "meta-llama/Llama-3.1-8B-Instruct",
-            "tensor_parallel_size": 1,
-            "load_format": "dummy",
-            "dtype": "bfloat16",
-            "distributed_executor_backend": "mp",
-            "block_size": 128,
-            "trust_remote_code": "",
-            "disable_log_stats": "",
-            "enforce_eager": "",
-            "max_num_batched_tokens": 2048,
-            "max_num_seqs": 256,
-            "num_iters_warmup": 5,
-            "num_iters": 15
-        }
-    }
-]
@@ -1,130 +0,0 @@
-{
-  "defaults": {
-    "qps_list": [
-      "inf"
-    ],
-    "max_concurrency_list": [
-      12,
-      16,
-      24,
-      32,
-      64,
-      128,
-      200
-    ],
-    "server_environment_variables": {
-      "VLLM_RPC_TIMEOUT": 100000,
-      "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
-      "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
-      "VLLM_CPU_SGL_KERNEL": 1,
-      "VLLM_CPU_KVCACHE_SPACE": 40
-    },
-    "server_parameters": {
-      "model": "meta-llama/Llama-3.1-8B-Instruct",
-      "tensor_parallel_size": 1,
-      "dtype": "bfloat16",
-      "distributed_executor_backend": "mp",
-      "block_size": 128,
-      "trust_remote_code": "",
-      "disable_log_stats": "",
-      "enforce_eager": "",
-      "max_num_batched_tokens": 2048,
-      "max_num_seqs": 256,
-      "load_format": "dummy"
-    },
-    "client_parameters": {
-      "model": "meta-llama/Llama-3.1-8B-Instruct",
-      "backend": "vllm",
-      "ignore-eos": "",
-      "num_prompts": 200
-    }
-  },
-  "tests": [
-    {
-      "test_name": "serving_llama8B_tp1_sharegpt",
-      "server_parameters": {
-        "tensor_parallel_size": 1
-      },
-      "client_parameters": {
-        "dataset_name": "sharegpt",
-        "dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json"
-      }
-    },
-    {
-      "test_name": "serving_llama8B_tp2_sharegpt",
-      "server_parameters": {
-        "tensor_parallel_size": 2
-      },
-      "client_parameters": {
-        "dataset_name": "sharegpt",
-        "dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json"
-      }
-    },
-    {
-      "test_name": "serving_llama8B_tp1_random_128_128",
-      "server_parameters": {
-        "tensor_parallel_size": 1
-      },
-      "client_parameters": {
-        "dataset_name": "random",
-        "random-input-len": 128,
-        "random-output-len": 128
-      }
-    },
-    {
-      "test_name": "serving_llama8B_tp2_random_128_128",
-      "server_parameters": {
-        "tensor_parallel_size": 2
-      },
-      "client_parameters": {
-        "dataset_name": "random",
-        "random-input-len": 128,
-        "random-output-len": 128
-      }
-    },
-    {
-      "test_name": "serving_llama8B_tp1_random_128_2048",
-      "server_parameters": {
-        "tensor_parallel_size": 1
-      },
-      "client_parameters": {
-        "dataset_name": "random",
-        "random-input-len": 128,
-        "random-output-len": 2048
-      }
-    },
-    {
-      "test_name": "serving_llama8B_tp2_random_128_2048",
-      "server_parameters": {
-        "tensor_parallel_size": 2
-      },
-      "client_parameters": {
-        "dataset_name": "random",
-        "random-input-len": 128,
-        "random-output-len": 2048
-      }
-    },
-    {
-      "test_name": "serving_llama8B_tp1_random_2048_128",
-      "server_parameters": {
-        "tensor_parallel_size": 1
-      },
-      "client_parameters": {
-        "dataset_name": "random",
-        "random-input-len": 2048,
-        "random-output-len": 128
-      }
-    },
-    {
-      "test_name": "serving_llama8B_tp2_random_2048_128",
-      "server_parameters": {
-        "tensor_parallel_size": 2
-      },
-      "client_parameters": {
-        "dataset_name": "random",
-        "random-input-len": 2048,
-        "random-output-len": 128
-      }
-    }
-  ]
-}
@@ -0,0 +1,610 @@
+[
+    {
+        "test_name": "serving_llama8B_bf16_tp1_sharegpt",
+        "qps_list": ["inf"],
+        "max_concurrency_list": [12, 16, 24, 32, 64, 128, 200],
+        "server_environment_variables": {
+            "VLLM_RPC_TIMEOUT": 100000,
+	    "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
+	    "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
+	    "VLLM_CPU_SGL_KERNEL": 1,
+	    "VLLM_CPU_KVCACHE_SPACE": 40
+        },
+        "server_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "tensor_parallel_size": 1,
+	    "dtype": "bfloat16",
+	    "distributed_executor_backend": "mp",
+	    "block_size": 128,
+	    "trust_remote_code": "",
+            "disable_log_stats": "",
+	    "enforce_eager": "",
+	    "max_num_batched_tokens": 2048,
+	    "max_num_seqs": 256,
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "backend": "vllm",
+            "dataset_name": "sharegpt",
+            "dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
+            "num_prompts": 200
+        }
+    },
+    {
+        "test_name": "serving_llama8B_bf16_tp2_sharegpt",
+        "qps_list": ["inf"],
+        "max_concurrency_list": [12, 16, 24, 32, 64, 128, 200],
+        "server_environment_variables": {
+            "VLLM_RPC_TIMEOUT": 100000,
+	    "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
+	    "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
+	    "VLLM_CPU_SGL_KERNEL": 1,
+	    "VLLM_CPU_KVCACHE_SPACE": 40
+        },
+        "server_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "tensor_parallel_size": 2,
+	    "dtype": "bfloat16",
+	    "distributed_executor_backend": "mp",
+	    "block_size": 128,
+	    "trust_remote_code": "",
+            "disable_log_stats": "",
+	    "enforce_eager": "",
+	    "max_num_batched_tokens": 2048,
+	    "max_num_seqs": 256,
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "backend": "vllm",
+            "dataset_name": "sharegpt",
+            "dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
+            "num_prompts": 200
+        }
+    },
+    {
+        "test_name": "serving_llama8B_bf16_tp4_sharegpt",
+        "qps_list": ["inf"],
+        "max_concurrency_list": [12, 16, 24, 32, 64, 128, 200],
+        "server_environment_variables": {
+            "VLLM_RPC_TIMEOUT": 100000,
+	    "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
+	    "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
+	    "VLLM_CPU_SGL_KERNEL": 1,
+	    "VLLM_CPU_KVCACHE_SPACE": 40
+        },
+        "server_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "tensor_parallel_size": 4,
+	    "dtype": "bfloat16",
+	    "distributed_executor_backend": "mp",
+	    "block_size": 128,
+	    "trust_remote_code": "",
+            "disable_log_stats": "",
+	    "enforce_eager": "",
+	    "max_num_batched_tokens": 2048,
+	    "max_num_seqs": 256,
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "backend": "vllm",
+            "dataset_name": "sharegpt",
+            "dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
+            "num_prompts": 200
+        }
+    },
+    {
+        "test_name": "serving_llama8B_bf16_tp1_random_128_128",
+        "qps_list": ["inf"],
+        "max_concurrency_list": [12, 16, 24, 32, 64, 128, 200, 1000],
+        "server_environment_variables": {
+            "VLLM_RPC_TIMEOUT": 100000,
+	    "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
+	    "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
+	    "VLLM_CPU_SGL_KERNEL": 1,
+	    "VLLM_CPU_KVCACHE_SPACE": 40
+        },
+        "server_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "tensor_parallel_size": 1,
+	    "dtype": "bfloat16",
+	    "distributed_executor_backend": "mp",
+	    "block_size": 128,
+	    "trust_remote_code": "",
+	    "enable_chunked_prefill": "",
+            "disable_log_stats": "",
+	    "enforce_eager": "",
+	    "max_num_batched_tokens": 2048,
+	    "max_num_seqs": 256,
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "backend": "vllm",
+            "dataset_name": "random",
+	    "random-input-len": 128,
+	    "random-output-len": 128,
+	    "ignore-eos": "",
+            "num_prompts": 1000
+        }
+    },
+    {
+        "test_name": "serving_llama8B_bf16_tp2_random_128_128",
+        "qps_list": ["inf"],
+        "max_concurrency_list": [12, 16, 24, 32, 64, 128, 200, 1000],
+        "server_environment_variables": {
+            "VLLM_RPC_TIMEOUT": 100000,
+	    "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
+	    "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
+	    "VLLM_CPU_SGL_KERNEL": 1,
+	    "VLLM_CPU_KVCACHE_SPACE": 40
+        },
+        "server_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "tensor_parallel_size": 2,
+	    "dtype": "bfloat16",
+	    "distributed_executor_backend": "mp",
+	    "block_size": 128,
+	    "trust_remote_code": "",
+	    "enable_chunked_prefill": "",
+            "disable_log_stats": "",
+	    "enforce_eager": "",
+	    "max_num_batched_tokens": 2048,
+	    "max_num_seqs": 256,
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "backend": "vllm",
+            "dataset_name": "random",
+	    "random-input-len": 128,
+	    "random-output-len": 128,
+	    "ignore-eos": "",
+            "num_prompts": 1000
+        }
+    },
+    {
+        "test_name": "serving_llama8B_bf16_tp4_random_128_128",
+        "qps_list": ["inf"],
+        "max_concurrency_list": [12, 16, 24, 32, 64, 128, 200, 1000],
+        "server_environment_variables": {
+            "VLLM_RPC_TIMEOUT": 100000,
+	    "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
+	    "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
+	    "VLLM_CPU_SGL_KERNEL": 1,
+	    "VLLM_CPU_KVCACHE_SPACE": 40
+        },
+        "server_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "tensor_parallel_size": 4,
+	    "dtype": "bfloat16",
+	    "distributed_executor_backend": "mp",
+	    "block_size": 128,
+	    "trust_remote_code": "",
+	    "enable_chunked_prefill": "",
+            "disable_log_stats": "",
+	    "enforce_eager": "",
+	    "max_num_batched_tokens": 2048,
+	    "max_num_seqs": 256,
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "backend": "vllm",
+            "dataset_name": "random",
+	    "random-input-len": 128,
+	    "random-output-len": 128,
+            "num_prompts": 1000
+        }
+    },
+    {
+        "test_name": "serving_llama8B_int8_tp1_sharegpt",
+        "qps_list": ["inf"],
+        "max_concurrency_list": [12, 16, 24, 32, 64, 128, 200],
+        "server_environment_variables": {
+            "VLLM_RPC_TIMEOUT": 100000,
+	    "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
+	    "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
+	    "VLLM_CPU_SGL_KERNEL": 1,
+	    "VLLM_CPU_KVCACHE_SPACE": 40
+        },
+        "server_parameters": {
+            "model": "RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8",
+            "tensor_parallel_size": 1,
+	    "dtype": "bfloat16",
+	    "distributed_executor_backend": "mp",
+	    "block_size": 128,
+	    "trust_remote_code": "",
+            "disable_log_stats": "",
+	    "enforce_eager": "",
+	    "max_num_batched_tokens": 2048,
+	    "max_num_seqs": 256,
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8",
+            "backend": "vllm",
+            "dataset_name": "sharegpt",
+            "dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
+            "num_prompts": 200
+        }
+    },
+    {
+        "test_name": "serving_llama8B_int8_tp2_sharegpt",
+        "qps_list": ["inf"],
+        "max_concurrency_list": [12, 16, 24, 32, 64, 128, 200],
+        "server_environment_variables": {
+            "VLLM_RPC_TIMEOUT": 100000,
+	    "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
+	    "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
+	    "VLLM_CPU_SGL_KERNEL": 1,
+	    "VLLM_CPU_KVCACHE_SPACE": 40
+        },
+        "server_parameters": {
+            "model": "RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8",
+            "tensor_parallel_size": 2,
+	    "dtype": "bfloat16",
+	    "distributed_executor_backend": "mp",
+	    "block_size": 128,
+	    "trust_remote_code": "",
+            "disable_log_stats": "",
+	    "enforce_eager": "",
+	    "max_num_batched_tokens": 2048,
+	    "max_num_seqs": 256,
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8",
+            "backend": "vllm",
+            "dataset_name": "sharegpt",
+            "dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
+            "num_prompts": 200
+        }
+    },
+    {
+        "test_name": "serving_llama8B_int8_tp4_sharegpt",
+        "qps_list": ["inf"],
+        "max_concurrency_list": [12, 16, 24, 32, 64, 128, 200],
+        "server_environment_variables": {
+            "VLLM_RPC_TIMEOUT": 100000,
+	    "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
+	    "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
+	    "VLLM_CPU_SGL_KERNEL": 1,
+	    "VLLM_CPU_KVCACHE_SPACE": 40
+        },
+        "server_parameters": {
+            "model": "RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8",
+            "tensor_parallel_size": 4,
+	    "dtype": "bfloat16",
+	    "distributed_executor_backend": "mp",
+	    "block_size": 128,
+	    "trust_remote_code": "",
+            "disable_log_stats": "",
+	    "enforce_eager": "",
+	    "max_num_batched_tokens": 2048,
+	    "max_num_seqs": 256,
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8",
+            "backend": "vllm",
+            "dataset_name": "sharegpt",
+            "dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
+            "num_prompts": 200
+        }
+    },
+    {
+        "test_name": "serving_llama8B_int8_tp1_random_128_128",
+        "qps_list": ["inf"],
+        "max_concurrency_list": [12, 16, 24, 32, 64, 128, 200, 1000],
+        "server_environment_variables": {
+            "VLLM_RPC_TIMEOUT": 100000,
+	    "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
+	    "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
+	    "VLLM_CPU_SGL_KERNEL": 1,
+	    "VLLM_CPU_KVCACHE_SPACE": 40
+        },
+        "server_parameters": {
+            "model": "RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8",
+            "tensor_parallel_size": 1,
+	    "dtype": "bfloat16",
+	    "distributed_executor_backend": "mp",
+	    "block_size": 128,
+	    "trust_remote_code": "",
+	    "enable_chunked_prefill": "",
+            "disable_log_stats": "",
+	    "enforce_eager": "",
+	    "max_num_batched_tokens": 2048,
+	    "max_num_seqs": 256,
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8",
+            "backend": "vllm",
+            "dataset_name": "random",
+	    "random-input-len": 128,
+	    "random-output-len": 128,
+	    "ignore-eos": "",
+            "num_prompts": 1000
+        }
+    },
+    {
+        "test_name": "serving_llama8B_int8_tp2_random_128_128",
+        "qps_list": ["inf"],
+        "max_concurrency_list": [12, 16, 24, 32, 64, 128, 200, 1000],
+        "server_environment_variables": {
+            "VLLM_RPC_TIMEOUT": 100000,
+	    "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
+	    "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
+	    "VLLM_CPU_SGL_KERNEL": 1,
+	    "VLLM_CPU_KVCACHE_SPACE": 40
+        },
+        "server_parameters": {
+            "model": "RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8",
+            "tensor_parallel_size": 2,
+	    "dtype": "bfloat16",
+	    "distributed_executor_backend": "mp",
+	    "block_size": 128,
+	    "trust_remote_code": "",
+	    "enable_chunked_prefill": "",
+            "disable_log_stats": "",
+	    "enforce_eager": "",
+	    "max_num_batched_tokens": 2048,
+	    "max_num_seqs": 256,
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8",
+            "backend": "vllm",
+            "dataset_name": "random",
+	    "random-input-len": 128,
+	    "random-output-len": 128,
+	    "ignore-eos": "",
+            "num_prompts": 1000
+        }
+    },
+    {
+        "test_name": "serving_llama8B_int8_tp4_random_128_128",
+        "qps_list": ["inf"],
+        "max_concurrency_list": [12, 16, 24, 32, 64, 128, 200, 1000],
+        "server_environment_variables": {
+            "VLLM_RPC_TIMEOUT": 100000,
+	    "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
+	    "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
+	    "VLLM_CPU_SGL_KERNEL": 1,
+	    "VLLM_CPU_KVCACHE_SPACE": 40
+        },
+        "server_parameters": {
+            "model": "RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8",
+            "tensor_parallel_size": 4,
+	    "dtype": "bfloat16",
+	    "distributed_executor_backend": "mp",
+	    "block_size": 128,
+	    "trust_remote_code": "",
+	    "enable_chunked_prefill": "",
+            "disable_log_stats": "",
+	    "enforce_eager": "",
+	    "max_num_batched_tokens": 2048,
+	    "max_num_seqs": 256,
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8",
+            "backend": "vllm",
+            "dataset_name": "random",
+	    "random-input-len": 128,
+	    "random-output-len": 128,
+	    "ignore-eos": "",
+            "num_prompts": 1000
+        }
+    },
+    {
+        "test_name": "serving_llama8B_int4_tp1_sharegpt",
+        "qps_list": ["inf"],
+        "max_concurrency_list": [12, 16, 24, 32, 64, 128, 200],
+        "server_environment_variables": {
+            "VLLM_RPC_TIMEOUT": 100000,
+	    "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
+	    "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
+	    "VLLM_CPU_SGL_KERNEL": 1,
+	    "VLLM_CPU_KVCACHE_SPACE": 40
+        },
+        "server_parameters": {
+            "model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
+	    "quantization": "awq",
+            "tensor_parallel_size": 1,
+	    "dtype": "bfloat16",
+	    "distributed_executor_backend": "mp",
+	    "block_size": 128,
+	    "trust_remote_code": "",
+            "disable_log_stats": "",
+	    "enforce_eager": "",
+	    "max_num_batched_tokens": 2048,
+	    "max_num_seqs": 256,
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
+            "backend": "vllm",
+            "dataset_name": "sharegpt",
+            "dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
+            "num_prompts": 200
+        }
+    },
+    {
+        "test_name": "serving_llama8B_int4_tp2_sharegpt",
+        "qps_list": ["inf"],
+        "max_concurrency_list": [12, 16, 24, 32, 64, 128, 200],
+        "server_environment_variables": {
+            "VLLM_RPC_TIMEOUT": 100000,
+	    "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
+	    "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
+	    "VLLM_CPU_SGL_KERNEL": 1,
+	    "VLLM_CPU_KVCACHE_SPACE": 40
+        },
+        "server_parameters": {
+            "model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
+	    "quantization": "awq",
+            "tensor_parallel_size": 2,
+	    "dtype": "bfloat16",
+	    "distributed_executor_backend": "mp",
+	    "block_size": 128,
+	    "trust_remote_code": "",
+            "disable_log_stats": "",
+	    "enforce_eager": "",
+	    "max_num_batched_tokens": 2048,
+	    "max_num_seqs": 256,
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
+            "backend": "vllm",
+            "dataset_name": "sharegpt",
+            "dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
+            "num_prompts": 200
+        }
+    },
+    {
+        "test_name": "serving_llama8B_int4_tp4_sharegpt",
+        "qps_list": ["inf"],
+        "max_concurrency_list": [12, 16, 24, 32, 64, 128, 200],
+        "server_environment_variables": {
+            "VLLM_RPC_TIMEOUT": 100000,
+	    "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
+	    "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
+	    "VLLM_CPU_SGL_KERNEL": 1,
+	    "VLLM_CPU_KVCACHE_SPACE": 40
+        },
+        "server_parameters": {
+            "model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
+	    "quantization": "awq",
+            "tensor_parallel_size": 4,
+	    "dtype": "bfloat16",
+	    "distributed_executor_backend": "mp",
+	    "block_size": 128,
+	    "trust_remote_code": "",
+            "disable_log_stats": "",
+	    "enforce_eager": "",
+	    "max_num_batched_tokens": 2048,
+	    "max_num_seqs": 256,
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
+            "backend": "vllm",
+            "dataset_name": "sharegpt",
+            "dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
+            "num_prompts": 200
+        }
+    },
+    {
+        "test_name": "serving_llama8B_int4_tp1_random_128_128",
+        "qps_list": ["inf"],
+        "max_concurrency_list": [12, 16, 24, 32, 64, 128, 200, 1000],
+        "server_environment_variables": {
+            "VLLM_RPC_TIMEOUT": 100000,
+	    "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
+	    "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
+	    "VLLM_CPU_SGL_KERNEL": 1,
+	    "VLLM_CPU_KVCACHE_SPACE": 40
+        },
+        "server_parameters": {
+            "model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
+	    "quantization": "awq",
+            "tensor_parallel_size": 1,
+	    "dtype": "bfloat16",
+	    "distributed_executor_backend": "mp",
+	    "block_size": 128,
+	    "trust_remote_code": "",
+	    "enable_chunked_prefill": "",
+            "disable_log_stats": "",
+	    "enforce_eager": "",
+	    "max_num_batched_tokens": 2048,
+	    "max_num_seqs": 256,
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
+            "backend": "vllm",
+            "dataset_name": "random",
+	    "random-input-len": 128,
+	    "random-output-len": 128,
+	    "ignore-eos": "",
+            "num_prompts": 1000
+        }
+    },
+    {
+        "test_name": "serving_llama8B_int4_tp2_random_128_128",
+        "qps_list": ["inf"],
+        "max_concurrency_list": [12, 16, 24, 32, 64, 128, 200, 1000],
+        "server_environment_variables": {
+            "VLLM_RPC_TIMEOUT": 100000,
+	    "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
+	    "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
+	    "VLLM_CPU_SGL_KERNEL": 1,
+	    "VLLM_CPU_KVCACHE_SPACE": 40
+        },
+        "server_parameters": {
+            "model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
+	    "quantization": "awq",
+            "tensor_parallel_size": 2,
+	    "dtype": "bfloat16",
+	    "distributed_executor_backend": "mp",
+	    "block_size": 128,
+	    "trust_remote_code": "",
+	    "enable_chunked_prefill": "",
+            "disable_log_stats": "",
+	    "enforce_eager": "",
+	    "max_num_batched_tokens": 2048,
+	    "max_num_seqs": 256,
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
+            "backend": "vllm",
+            "dataset_name": "random",
+	    "random-input-len": 128,
+	    "random-output-len": 128,
+	    "ignore-eos": "",
+            "num_prompts": 1000
+        }
+    },
+    {
+        "test_name": "serving_llama8B_int4_tp4_random_128_128",
+        "qps_list": ["inf"],
+        "max_concurrency_list": [12, 16, 24, 32, 64, 128, 200, 1000],
+        "server_environment_variables": {
+            "VLLM_RPC_TIMEOUT": 100000,
+	    "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
+	    "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
+	    "VLLM_CPU_SGL_KERNEL": 1,
+	    "VLLM_CPU_KVCACHE_SPACE": 40
+        },
+        "server_parameters": {
+            "model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
+	    "quantization": "awq",
+            "tensor_parallel_size": 4,
+	    "dtype": "bfloat16",
+	    "distributed_executor_backend": "mp",
+	    "block_size": 128,
+	    "trust_remote_code": "",
+	    "enable_chunked_prefill": "",
+            "disable_log_stats": "",
+	    "enforce_eager": "",
+	    "max_num_batched_tokens": 2048,
+	    "max_num_seqs": 256,
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
+            "backend": "vllm",
+            "dataset_name": "random",
+	    "random-input-len": 128,
+	    "random-output-len": 128,
+	    "ignore-eos": "",
+            "num_prompts": 1000
+        }
+    }
+]
@@ -1,283 +1,276 @@
-{
-  "defaults": {
-    "qps_list": [
-      "inf"
-    ],
-    "max_concurrency_list": [12, 16, 24, 32, 64, 128, 200],
-    "server_environment_variables": {
-      "VLLM_RPC_TIMEOUT": 100000,
-      "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
-      "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
-      "VLLM_CPU_SGL_KERNEL": 1,
-      "VLLM_CPU_KVCACHE_SPACE": 40
+[
+    {
+        "test_name": "serving_llama8B_tp1_sharegpt",
+        "qps_list": [1, 4, 16, "inf"],
+        "max_concurrency_list": [32],
+        "server_environment_variables": {
+            "VLLM_RPC_TIMEOUT": 100000,
+	    "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
+	    "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
+	    "VLLM_CPU_SGL_KERNEL": 1,
+	    "VLLM_CPU_KVCACHE_SPACE": 40
+        },
+        "server_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "tensor_parallel_size": 1,
+	    "dtype": "bfloat16",
+	    "distributed_executor_backend": "mp",
+	    "block_size": 128,
+	    "trust_remote_code": "",
+            "disable_log_stats": "",
+	    "enforce_eager": "",
+	    "max_num_batched_tokens": 2048,
+	    "max_num_seqs": 256,
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "backend": "vllm",
+            "dataset_name": "sharegpt",
+            "dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
+            "num_prompts": 32
+        }
    },
-    "server_parameters": {
-      "model": "meta-llama/Llama-3.1-8B-Instruct",
-      "tensor_parallel_size": 1,
-      "dtype": "bfloat16",
-      "distributed_executor_backend": "mp",
-      "block_size": 128,
-      "trust_remote_code": "",
-      "disable_log_stats": "",
-      "max_num_batched_tokens": 2048,
-      "max_num_seqs": 256
+    {
+        "test_name": "serving_llama8B_tp2_sharegpt",
+        "qps_list": [1, 4, 16, "inf"],
+        "max_concurrency_list": [32],
+        "server_environment_variables": {
+            "VLLM_RPC_TIMEOUT": 100000,
+	    "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
+	    "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
+	    "VLLM_CPU_SGL_KERNEL": 1,
+	    "VLLM_CPU_KVCACHE_SPACE": 40
+        },
+        "server_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "tensor_parallel_size": 2,
+	    "dtype": "bfloat16",
+	    "distributed_executor_backend": "mp",
+	    "block_size": 128,
+	    "trust_remote_code": "",
+            "disable_log_stats": "",
+	    "enforce_eager": "",
+	    "max_num_batched_tokens": 2048,
+	    "max_num_seqs": 256,
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "backend": "vllm",
+            "dataset_name": "sharegpt",
+            "dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
+            "num_prompts": 32
+        }
    },
-    "client_parameters": {
-      "model": "meta-llama/Llama-3.1-8B-Instruct",
-      "backend": "vllm",
-      "ignore-eos": "",
-      "num_prompts": 200
+    {
+        "test_name": "serving_llama8B_tp1_random_128_128",
+        "qps_list": [1, 4, 16, "inf"],
+        "max_concurrency_list": [32],
+        "server_environment_variables": {
+            "VLLM_RPC_TIMEOUT": 100000,
+	    "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
+	    "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
+	    "VLLM_CPU_SGL_KERNEL": 1,
+	    "VLLM_CPU_KVCACHE_SPACE": 40
+        },
+        "server_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "tensor_parallel_size": 1,
+	    "dtype": "bfloat16",
+	    "distributed_executor_backend": "mp",
+	    "block_size": 128,
+	    "trust_remote_code": "",
+	    "enable_chunked_prefill": "",
+            "disable_log_stats": "",
+	    "enforce_eager": "",
+	    "max_num_batched_tokens": 2048,
+	    "max_num_seqs": 256,
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "backend": "vllm",
+            "dataset_name": "random",
+	    "random-input-len": 128,
+	    "random-output-len": 128,
+	    "ignore-eos": "",
+            "num_prompts": 32
+        }
+    },
+    {
+        "test_name": "serving_llama8B_tp2_random_128_128",
+        "qps_list": [1, 4, 16, "inf"],
+        "max_concurrency_list": [32],
+        "server_environment_variables": {
+            "VLLM_RPC_TIMEOUT": 100000,
+	    "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
+	    "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
+	    "VLLM_CPU_SGL_KERNEL": 1,
+	    "VLLM_CPU_KVCACHE_SPACE": 40
+        },
+        "server_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "tensor_parallel_size": 2,
+	    "dtype": "bfloat16",
+	    "distributed_executor_backend": "mp",
+	    "block_size": 128,
+	    "trust_remote_code": "",
+	    "enable_chunked_prefill": "",
+            "disable_log_stats": "",
+	    "enforce_eager": "",
+	    "max_num_batched_tokens": 2048,
+	    "max_num_seqs": 256,
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "backend": "vllm",
+            "dataset_name": "random",
+	    "random-input-len": 128,
+	    "random-output-len": 128,
+	    "ignore-eos": "",
+            "num_prompts": 32
+        }
+    },
+    {
+        "test_name": "serving_llama8B_tp1_random_128_2048",
+        "qps_list": [1, 4, 16, "inf"],
+        "max_concurrency_list": [32],
+        "server_environment_variables": {
+            "VLLM_RPC_TIMEOUT": 100000,
+	    "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
+	    "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
+	    "VLLM_CPU_SGL_KERNEL": 1,
+	    "VLLM_CPU_KVCACHE_SPACE": 40
+        },
+        "server_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "tensor_parallel_size": 1,
+	    "dtype": "bfloat16",
+	    "distributed_executor_backend": "mp",
+	    "block_size": 128,
+	    "trust_remote_code": "",
+	    "enable_chunked_prefill": "",
+            "disable_log_stats": "",
+	    "enforce_eager": "",
+	    "max_num_batched_tokens": 2048,
+	    "max_num_seqs": 256,
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "backend": "vllm",
+            "dataset_name": "random",
+	    "random-input-len": 128,
+	    "random-output-len": 2048,
+	    "ignore-eos": "",
+            "num_prompts": 32
+        }
+    },
+    {
+        "test_name": "serving_llama8B_tp2_random_128_2048",
+        "qps_list": [1, 4, 16, "inf"],
+        "max_concurrency_list": [32],
+        "server_environment_variables": {
+            "VLLM_RPC_TIMEOUT": 100000,
+	    "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
+	    "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
+	    "VLLM_CPU_SGL_KERNEL": 1,
+	    "VLLM_CPU_KVCACHE_SPACE": 40
+        },
+        "server_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "tensor_parallel_size": 2,
+	    "dtype": "bfloat16",
+	    "distributed_executor_backend": "mp",
+	    "block_size": 128,
+	    "trust_remote_code": "",
+	    "enable_chunked_prefill": "",
+            "disable_log_stats": "",
+	    "enforce_eager": "",
+	    "max_num_batched_tokens": 2048,
+	    "max_num_seqs": 256,
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "backend": "vllm",
+            "dataset_name": "random",
+	    "random-input-len": 128,
+	    "random-output-len": 2048,
+	    "ignore-eos": "",
+            "num_prompts": 32
+        }
+    },
+    {
+        "test_name": "serving_llama8B_tp1_random_2048_128",
+        "qps_list": [1, 4, 16, "inf"],
+        "max_concurrency_list": [32],
+        "server_environment_variables": {
+            "VLLM_RPC_TIMEOUT": 100000,
+	    "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
+	    "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
+	    "VLLM_CPU_SGL_KERNEL": 1,
+	    "VLLM_CPU_KVCACHE_SPACE": 40
+        },
+        "server_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "tensor_parallel_size": 1,
+	    "dtype": "bfloat16",
+	    "distributed_executor_backend": "mp",
+	    "block_size": 128,
+	    "trust_remote_code": "",
+	    "enable_chunked_prefill": "",
+            "disable_log_stats": "",
+	    "enforce_eager": "",
+	    "max_num_batched_tokens": 2048,
+	    "max_num_seqs": 256,
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "backend": "vllm",
+            "dataset_name": "random",
+	    "random-input-len": 2048,
+	    "random-output-len": 128,
+	    "ignore-eos": "",
+            "num_prompts": 32
+        }
+    },
+    {
+        "test_name": "serving_llama8B_tp2_random_2048_128",
+        "qps_list": [1, 4, 16, "inf"],
+        "max_concurrency_list": [32],
+        "server_environment_variables": {
+            "VLLM_RPC_TIMEOUT": 100000,
+	    "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
+	    "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
+	    "VLLM_CPU_SGL_KERNEL": 1,
+	    "VLLM_CPU_KVCACHE_SPACE": 40
+        },
+        "server_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "tensor_parallel_size": 2,
+	    "dtype": "bfloat16",
+	    "distributed_executor_backend": "mp",
+	    "block_size": 128,
+	    "trust_remote_code": "",
+	    "enable_chunked_prefill": "",
+            "disable_log_stats": "",
+	    "enforce_eager": "",
+	    "max_num_batched_tokens": 2048,
+	    "max_num_seqs": 256,
+            "load_format": "dummy"
+        },
+        "client_parameters": {
+            "model": "meta-llama/Llama-3.1-8B-Instruct",
+            "backend": "vllm",
+            "dataset_name": "random",
+	    "random-input-len": 2048,
+	    "random-output-len": 128,
+	    "ignore-eos": "",
+            "num_prompts": 32
+        }
    }
-  },
-  "tests": [
-    {
-      "test_name": "serving_llama8B_tp1_sharegpt",
-      "server_parameters": {
-        "tensor_parallel_size": 1
-      },
-      "client_parameters": {
-        "dataset_name": "sharegpt",
-        "dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json"
-      }
-    },
-    {
-      "test_name": "serving_llama8B_tp2_sharegpt",
-      "server_parameters": {
-        "tensor_parallel_size": 2
-      },
-      "client_parameters": {
-        "dataset_name": "sharegpt",
-        "dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json"
-      }
-    },
-    {
-      "test_name": "serving_llama8B_tp1_random_128_128",
-      "server_parameters": {
-        "tensor_parallel_size": 1
-      },
-      "client_parameters": {
-        "dataset_name": "random",
-        "random-input-len": 128,
-        "random-output-len": 128
-      }
-    },
-    {
-      "test_name": "serving_llama8B_tp2_random_128_128",
-      "server_parameters": {
-        "tensor_parallel_size": 2
-      },
-      "client_parameters": {
-        "dataset_name": "random",
-        "random-input-len": 128,
-        "random-output-len": 128
-      }
-    },
-    {
-      "test_name": "serving_llama8B_tp4_random_128_128",
-      "server_parameters": {
-        "tensor_parallel_size": 4
-      },
-      "client_parameters": {
-        "dataset_name": "random",
-        "random-input-len": 128,
-        "random-output-len": 128
-      }
-    },
-    {
-      "test_name": "serving_llama8B_tp1_random_128_2048",
-      "server_parameters": {
-        "tensor_parallel_size": 1
-      },
-      "client_parameters": {
-        "dataset_name": "random",
-        "random-input-len": 128,
-        "random-output-len": 2048
-      }
-    },
-    {
-      "test_name": "serving_llama8B_tp2_random_128_2048",
-      "server_parameters": {
-        "tensor_parallel_size": 2
-      },
-      "client_parameters": {
-        "dataset_name": "random",
-        "random-input-len": 128,
-        "random-output-len": 2048
-      }
-    },
-    {
-      "test_name": "serving_llama8B_tp4_random_128_2048",
-      "server_parameters": {
-        "tensor_parallel_size": 4
-      },
-      "client_parameters": {
-        "dataset_name": "random",
-        "random-input-len": 128,
-        "random-output-len": 2048
-      }
-    },
-    {
-      "test_name": "serving_llama8B_tp1_random_2048_128",
-      "server_parameters": {
-        "tensor_parallel_size": 1
-      },
-      "client_parameters": {
-        "dataset_name": "random",
-        "random-input-len": 2048,
-        "random-output-len": 128
-      }
-    },
-    {
-      "test_name": "serving_llama8B_tp2_random_2048_128",
-      "server_parameters": {
-        "tensor_parallel_size": 2
-      },
-      "client_parameters": {
-        "dataset_name": "random",
-        "random-input-len": 2048,
-        "random-output-len": 128
-      }
-    },
-    {
-      "test_name": "serving_llama8B_tp4_random_2048_128",
-      "server_parameters": {
-        "tensor_parallel_size": 4
-      },
-      "client_parameters": {
-        "dataset_name": "random",
-        "random-input-len": 2048,
-        "random-output-len": 128
-      }
-    },
-    {
-      "test_name": "serving_llama8B_int4_tp1_random_128_128",
-      "server_parameters": {
-        "model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
-        "tensor_parallel_size": 1
-      },
-      "client_parameters": {
-        "model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
-        "dataset_name": "random",
-        "random-input-len": 128,
-        "random-output-len": 128
-      }
-    },
-    {
-      "test_name": "serving_llama8B_int4_tp2_random_128_128",
-      "server_parameters": {
-        "model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
-        "tensor_parallel_size": 2
-      },
-      "client_parameters": {
-        "model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
-        "dataset_name": "random",
-        "random-input-len": 128,
-        "random-output-len": 128
-      }
-    },
-    {
-      "test_name": "serving_llama8B_int4_tp4_random_128_128",
-      "server_parameters": {
-        "model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
-        "tensor_parallel_size": 4
-      },
-      "client_parameters": {
-        "model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
-        "dataset_name": "random",
-        "random-input-len": 128,
-        "random-output-len": 128
-      }
-    },
-    {
-      "test_name": "serving_llama3B_tp1_random_128_128",
-      "server_parameters": {
-        "model": "meta-llama/Llama-3.2-3B-Instruct",
-        "tensor_parallel_size": 1
-      },
-      "client_parameters": {
-        "model": "meta-llama/Llama-3.2-3B-Instruct",
-        "dataset_name": "random",
-        "random-input-len": 128,
-        "random-output-len": 128
-      }
-    },
-    {
-      "test_name": "serving_granite2B_tp1_random_128_128",
-      "server_parameters": {
-        "model": "ibm-granite/granite-3.2-2b-instruct",
-        "tensor_parallel_size": 1
-      },
-      "client_parameters": {
-        "model": "ibm-granite/granite-3.2-2b-instruct",
-        "dataset_name": "random",
-        "random-input-len": 128,
-        "random-output-len": 128
-      }
-    },
-    {
-      "test_name": "serving_qwen1.7B_tp1_random_128_128",
-      "server_parameters": {
-        "model": "Qwen/Qwen3-1.7B",
-        "tensor_parallel_size": 1
-      },
-      "client_parameters": {
-        "model": "Qwen/Qwen3-1.7B",
-        "dataset_name": "random",
-        "random-input-len": 128,
-        "random-output-len": 128
-      }
-    },
-    {
-      "test_name": "serving_qwen4B_tp1_random_128_128",
-      "server_parameters": {
-        "model": "Qwen/Qwen3-4B",
-        "tensor_parallel_size": 1
-      },
-      "client_parameters": {
-        "model": "Qwen/Qwen3-4B",
-        "dataset_name": "random",
-        "random-input-len": 128,
-        "random-output-len": 128
-      }
-    },
-    {
-      "test_name": "serving_qwen8B_tp1_random_128_128",
-      "server_parameters": {
-        "model": "Qwen/Qwen3-8B",
-        "tensor_parallel_size": 1
-      },
-      "client_parameters": {
-        "model": "Qwen/Qwen3-8B",
-        "dataset_name": "random",
-        "random-input-len": 128,
-        "random-output-len": 128
-      }
-    },
-    {
-      "test_name": "serving_glm9B_tp1_random_128_128",
-      "server_parameters": {
-        "model": "zai-org/glm-4-9b-hf",
-        "tensor_parallel_size": 1
-      },
-      "client_parameters": {
-        "model": "zai-org/glm-4-9b-hf",
-        "dataset_name": "random",
-        "random-input-len": 128,
-        "random-output-len": 128
-      }
-    },
-    {
-      "test_name": "serving_gemma7B_tp1_random_128_128",
-      "server_parameters": {
-        "model": "google/gemma-7b",
-        "tensor_parallel_size": 1
-      },
-      "client_parameters": {
-        "model": "google/gemma-7b",
-        "dataset_name": "random",
-        "random-input-len": 128,
-        "random-output-len": 128
-      }
-    }
-  ]
-}
+]
@@ -1,27 +0,0 @@
-[
-    {
-        "test_name": "throughput_llama8B_tp1",
-        "environment_variables": {
-            "VLLM_RPC_TIMEOUT": 100000,
-            "VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
-            "VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
-            "VLLM_CPU_KVCACHE_SPACE": 40
-        },
-        "parameters": {
-            "model": "meta-llama/Llama-3.1-8B-Instruct",
-            "tensor_parallel_size": 1,
-            "load_format": "dummy",
-            "dtype": "bfloat16",
-            "distributed_executor_backend": "mp",
-            "block_size": 128,
-            "trust_remote_code": "",
-            "disable_log_stats": "",
-            "enforce_eager": "",
-            "max_num_batched_tokens": 2048,
-            "max_num_seqs": 256,
-            "dataset": "./ShareGPT_V3_unfiltered_cleaned_split.json",
-            "num_prompts": 200,
-            "backend": "vllm"
-        }
-    }
-]
@@ -8,28 +8,13 @@ steps:
    commands:
      # #NOTE: torch_cuda_arch_list is derived from upstream PyTorch build files here:
      # https://github.com/pytorch/pytorch/blob/main/.ci/aarch64_linux/aarch64_ci_build.sh#L7
-      - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --build-arg CUDA_VERSION=12.9.1 --build-arg torch_cuda_arch_list='8.7 8.9 9.0 10.0+PTX 12.0' --tag vllm-ci:build-image --target build --progress plain -f docker/Dockerfile ."
+      - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --build-arg CUDA_VERSION=12.9.1 --build-arg VLLM_MAIN_CUDA_VERSION=12.9 --build-arg torch_cuda_arch_list='8.7 8.9 9.0 10.0+PTX 12.0' --tag vllm-ci:build-image --target build --progress plain -f docker/Dockerfile ."
      - "mkdir artifacts"
      - "docker run --rm -v $(pwd)/artifacts:/artifacts_host vllm-ci:build-image bash -c 'cp -r dist /artifacts_host && chmod -R a+rw /artifacts_host'"
      - "bash .buildkite/scripts/upload-wheels.sh"
    env:
      DOCKER_BUILDKIT: "1"

-  - label: "Build arm64 wheel - CUDA 13.0"
-    depends_on: ~
-    id: build-wheel-arm64-cuda-13-0
-    agents:
-      queue: arm64_cpu_queue_postmerge
-    commands:
-      # #NOTE: torch_cuda_arch_list is derived from upstream PyTorch build files here:
-      # https://github.com/pytorch/pytorch/blob/main/.ci/aarch64_linux/aarch64_ci_build.sh#L7
-      - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --build-arg CUDA_VERSION=13.0.1 --build-arg torch_cuda_arch_list='8.7 8.9 9.0 10.0+PTX 12.0' --build-arg BUILD_BASE_IMAGE=nvidia/cuda:13.0.1-devel-ubuntu22.04  --tag vllm-ci:build-image --target build --progress plain -f docker/Dockerfile ."
-      - "mkdir artifacts"
-      - "docker run --rm -v $(pwd)/artifacts:/artifacts_host vllm-ci:build-image bash -c 'cp -r dist /artifacts_host && chmod -R a+rw /artifacts_host'"
-      - "bash .buildkite/scripts/upload-wheels.sh manylinux_2_35"
-    env:
-      DOCKER_BUILDKIT: "1"
-
  # aarch64 build
  - label: "Build arm64 CPU wheel"
    depends_on: ~
@@ -40,11 +25,24 @@ steps:
      - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg GIT_REPO_CHECK=1 --build-arg VLLM_BUILD_ACL=ON --tag vllm-ci:build-image --target vllm-build --progress plain -f docker/Dockerfile.cpu ."
      - "mkdir artifacts"
      - "docker run --rm -v $(pwd)/artifacts:/artifacts_host vllm-ci:build-image bash -c 'cp -r dist /artifacts_host && chmod -R a+rw /artifacts_host'"
-      - "bash .buildkite/scripts/upload-wheels.sh manylinux_2_35"
+      - "bash .buildkite/scripts/upload-wheels.sh"
    env:
      DOCKER_BUILDKIT: "1"

  # x86 + CUDA builds
+  - label: "Build wheel - CUDA 12.8"
+    depends_on: ~
+    id: build-wheel-cuda-12-8
+    agents:
+      queue: cpu_queue_postmerge
+    commands:
+      - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --build-arg CUDA_VERSION=12.8.1 --tag vllm-ci:build-image --target build --progress plain -f docker/Dockerfile ."
+      - "mkdir artifacts"
+      - "docker run --rm -v $(pwd)/artifacts:/artifacts_host vllm-ci:build-image bash -c 'cp -r dist /artifacts_host && chmod -R a+rw /artifacts_host'"
+      - "bash .buildkite/scripts/upload-wheels.sh"
+    env:
+      DOCKER_BUILDKIT: "1"
+
  - label: "Build wheel - CUDA 12.9"
    depends_on: ~
    id: build-wheel-cuda-12-9
@@ -54,7 +52,7 @@ steps:
      - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --build-arg CUDA_VERSION=12.9.1 --tag vllm-ci:build-image --target build --progress plain -f docker/Dockerfile ."
      - "mkdir artifacts"
      - "docker run --rm -v $(pwd)/artifacts:/artifacts_host vllm-ci:build-image bash -c 'cp -r dist /artifacts_host && chmod -R a+rw /artifacts_host'"
-      - "bash .buildkite/scripts/upload-wheels.sh manylinux_2_31"
+      - "bash .buildkite/scripts/upload-wheels.sh"
    env:
      DOCKER_BUILDKIT: "1"

@@ -67,21 +65,7 @@ steps:
      - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --build-arg CUDA_VERSION=13.0.1 --build-arg BUILD_BASE_IMAGE=nvidia/cuda:13.0.1-devel-ubuntu22.04 --tag vllm-ci:build-image --target build --progress plain -f docker/Dockerfile ."
      - "mkdir artifacts"
      - "docker run --rm -v $(pwd)/artifacts:/artifacts_host vllm-ci:build-image bash -c 'cp -r dist /artifacts_host && chmod -R a+rw /artifacts_host'"
-      - "bash .buildkite/scripts/upload-wheels.sh manylinux_2_35"
-    env:
-      DOCKER_BUILDKIT: "1"
-
-  # x86 CPU wheel build
-  - label: "Build x86 CPU wheel"
-    depends_on: ~
-    id: build-wheel-x86-cpu
-    agents:
-      queue: cpu_queue_postmerge
-    commands:
-      - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg GIT_REPO_CHECK=1 --build-arg VLLM_CPU_AVX512BF16=true --build-arg VLLM_CPU_AVX512VNNI=true --build-arg VLLM_CPU_AMXBF16=true --tag vllm-ci:build-image --target vllm-build --progress plain -f docker/Dockerfile.cpu ."
-      - "mkdir artifacts"
-      - "docker run --rm -v $(pwd)/artifacts:/artifacts_host vllm-ci:build-image bash -c 'cp -r dist /artifacts_host && chmod -R a+rw /artifacts_host'"
-      - "bash .buildkite/scripts/upload-wheels.sh manylinux_2_35"
+      - "bash .buildkite/scripts/upload-wheels.sh"
    env:
      DOCKER_BUILDKIT: "1"

@@ -125,6 +109,7 @@ steps:
  - label: "Annotate release workflow"
    depends_on:
      - create-multi-arch-manifest
+      - build-wheel-cuda-12-8
    id: annotate-release-workflow
    agents:
      queue: cpu_queue_postmerge
@@ -147,7 +132,7 @@ steps:
      queue: cpu_queue_postmerge
    commands:
      - "aws ecr-public get-login-password --region us-east-1 | docker login --username AWS --password-stdin public.ecr.aws/q9t5s3a7"
-      - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg GIT_REPO_CHECK=1 --build-arg VLLM_CPU_AVX512BF16=true --build-arg VLLM_CPU_AVX512VNNI=true --build-arg VLLM_CPU_AMXBF16=true --tag public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:$(buildkite-agent meta-data get release-version) --tag public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:latest --progress plain --target vllm-openai -f docker/Dockerfile.cpu ."
+      - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg GIT_REPO_CHECK=1 --build-arg VLLM_CPU_AVX512BF16=true --build-arg VLLM_CPU_AVX512VNNI=true --tag public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:$(buildkite-agent meta-data get release-version) --tag public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:latest --progress plain --target vllm-openai -f docker/Dockerfile.cpu ."
      - "docker push public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:latest"
      - "docker push public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:$(buildkite-agent meta-data get release-version)"
    env:
@@ -2,29 +2,22 @@

 set -ex

-# Get release version, default to 1.0.0.dev for nightly/per-commit builds
-RELEASE_VERSION=$(buildkite-agent meta-data get release-version 2>/dev/null | sed 's/^v//')
-if [ -z "${RELEASE_VERSION}" ]; then
-  RELEASE_VERSION="1.0.0.dev"
+# Get release version and strip leading 'v' if present
+RELEASE_VERSION=$(buildkite-agent meta-data get release-version | sed 's/^v//')
+
+if [ -z "$RELEASE_VERSION" ]; then
+  echo "Error: RELEASE_VERSION is empty. 'release-version' metadata might not be set or is invalid."
+  exit 1
 fi

 buildkite-agent annotate --style 'info' --context 'release-workflow' << EOF
-To download the wheel (by commit):
-\`\`\`
-aws s3 cp s3://vllm-wheels/${BUILDKITE_COMMIT}/vllm-${RELEASE_VERSION}-cp38-abi3-manylinux1_x86_64.whl .
-aws s3 cp s3://vllm-wheels/${BUILDKITE_COMMIT}/vllm-${RELEASE_VERSION}-cp38-abi3-manylinux2014_aarch64.whl .
-
-aws s3 cp s3://vllm-wheels/${BUILDKITE_COMMIT}/vllm-${RELEASE_VERSION}+cu129-cp38-abi3-manylinux1_x86_64.whl .
-aws s3 cp s3://vllm-wheels/${BUILDKITE_COMMIT}/vllm-${RELEASE_VERSION}+cu129-cp38-abi3-manylinux1_x86_64.whl .
-\`\`\`
-
-To download the wheel (by version):
+To download the wheel:
 \`\`\`
 aws s3 cp s3://vllm-wheels/${RELEASE_VERSION}/vllm-${RELEASE_VERSION}-cp38-abi3-manylinux1_x86_64.whl .
 aws s3 cp s3://vllm-wheels/${RELEASE_VERSION}/vllm-${RELEASE_VERSION}-cp38-abi3-manylinux2014_aarch64.whl .

+aws s3 cp s3://vllm-wheels/${RELEASE_VERSION}+cu126/vllm-${RELEASE_VERSION}+cu126-cp38-abi3-manylinux1_x86_64.whl .
 aws s3 cp s3://vllm-wheels/${RELEASE_VERSION}+cu129/vllm-${RELEASE_VERSION}+cu129-cp38-abi3-manylinux1_x86_64.whl .
-aws s3 cp s3://vllm-wheels/${RELEASE_VERSION}+cu130/vllm-${RELEASE_VERSION}+cu130-cp38-abi3-manylinux1_x86_64.whl .
 \`\`\`

 To download and upload the image:
@@ -45,10 +38,9 @@ docker tag vllm/vllm-openai:aarch64 vllm/vllm-openai:v${RELEASE_VERSION}-aarch64
 docker push vllm/vllm-openai:latest-aarch64
 docker push vllm/vllm-openai:v${RELEASE_VERSION}-aarch64

-docker manifest rm vllm/vllm-openai:latest
-docker manifest create vllm/vllm-openai:latest vllm/vllm-openai:latest-x86_64 vllm/vllm-openai:latest-aarch64
-docker manifest create vllm/vllm-openai:v${RELEASE_VERSION} vllm/vllm-openai:v${RELEASE_VERSION}-x86_64 vllm/vllm-openai:v${RELEASE_VERSION}-aarch64
+docker manifest create vllm/vllm-openai:latest vllm/vllm-openai:latest-x86_64 vllm/vllm-openai:latest-aarch64 --amend
+docker manifest create vllm/vllm-openai:v${RELEASE_VERSION} vllm/vllm-openai:v${RELEASE_VERSION}-x86_64 vllm/vllm-openai:v${RELEASE_VERSION}-aarch64 --amend
 docker manifest push vllm/vllm-openai:latest
 docker manifest push vllm/vllm-openai:v${RELEASE_VERSION}
 \`\`\`
-EOF 
+EOF 
@@ -1,408 +0,0 @@
-#!/usr/bin/env python3
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-
-# do not complain about line length (for docstring)
-# ruff: noqa: E501
-
-import argparse
-import json
-import sys
-from dataclasses import asdict, dataclass
-from datetime import datetime
-from pathlib import Path
-from typing import Any
-from urllib.parse import quote
-
-import regex as re
-
-if not sys.version_info >= (3, 12):
-    raise RuntimeError("This script requires Python 3.12 or higher.")
-
-INDEX_HTML_TEMPLATE = """<!DOCTYPE html>
-<html>
-  <!-- {comment} -->
-  <meta name="pypi:repository-version" content="1.0">
-  <body>
-{items}
-  </body>
-</html>
-"""
-
-
-@dataclass
-class WheelFileInfo:
-    package_name: str
-    version: str
-    build_tag: str | None
-    python_tag: str
-    abi_tag: str
-    platform_tag: str
-    variant: str | None
-    filename: str
-
-
-def parse_from_filename(file: str) -> WheelFileInfo:
-    """
-    Parse wheel file name to extract metadata.
-
-    The format of wheel names:
-        {package_name}-{version}(-{build_tag})?-{python_tag}-{abi_tag}-{platform_tag}.whl
-    All versions could contain a variant like '+cu129' or '.cpu' or `.rocm` (or not).
-    Example:
-        vllm-0.11.0-cp38-abi3-manylinux1_x86_64.whl
-        vllm-0.10.2rc2+cu129-cp38-abi3-manylinux2014_aarch64.whl
-        vllm-0.11.1rc8.dev14+gaa384b3c0-cp38-abi3-manylinux2014_aarch64.whl
-        vllm-0.11.1rc8.dev14+gaa384b3c0.cu130-cp38-abi3-manylinux1_x86_64.whl
-    """
-    wheel_file_re = re.compile(
-        r"^(?P<package_name>.+)-(?P<version>[^-]+?)(-(?P<build_tag>[^-]+))?-(?P<python_tag>[^-]+)-(?P<abi_tag>[^-]+)-(?P<platform_tag>[^-]+)\.whl$"
-    )
-    match = wheel_file_re.match(file)
-    if not match:
-        raise ValueError(f"Invalid wheel file name: {file}")
-
-    package_name = match.group("package_name")
-    version = match.group("version")
-    build_tag = match.group("build_tag")
-    python_tag = match.group("python_tag")
-    abi_tag = match.group("abi_tag")
-    platform_tag = match.group("platform_tag")
-
-    # extract variant from version
-    variant = None
-    if "dev" in version:
-        ver_after_dev = version.split("dev")[-1]
-        if "." in ver_after_dev:
-            variant = ver_after_dev.split(".")[-1]
-            version = version.removesuffix("." + variant)
-    else:
-        if "+" in version:
-            version, variant = version.split("+")
-
-    return WheelFileInfo(
-        package_name=package_name,
-        version=version,
-        build_tag=build_tag,
-        python_tag=python_tag,
-        abi_tag=abi_tag,
-        platform_tag=platform_tag,
-        variant=variant,
-        filename=file,
-    )
-
-
-def generate_project_list(subdir_names: list[str], comment: str = "") -> str:
-    """
-    Generate project list HTML content linking to each project & variant sub-directory.
-    """
-    href_tags = []
-    for name in sorted(subdir_names):
-        name = name.strip("/").strip(".")
-        href_tags.append(f'    <a href="{name}/">{name}/</a><br/>')
-    return INDEX_HTML_TEMPLATE.format(items="\n".join(href_tags), comment=comment)
-
-
-def generate_package_index_and_metadata(
-    wheel_files: list[WheelFileInfo],
-    wheel_base_dir: Path,
-    index_base_dir: Path,
-    comment: str = "",
-) -> tuple[str, str]:
-    """
-    Generate package index HTML content for a specific package, linking to actual wheel files.
-    """
-    href_tags = []
-    metadata = []
-    for file in sorted(wheel_files, key=lambda x: x.filename):
-        relative_path = (
-            wheel_base_dir.relative_to(index_base_dir, walk_up=True) / file.filename
-        )
-        # handle with '+' in URL, and avoid double-encoding '/' and already-encoded '%2B'
-        # NOTE: this is AWS S3 specific behavior!
-        file_path_quoted = quote(relative_path.as_posix(), safe=":%/")
-        href_tags.append(f'    <a href="{file_path_quoted}">{file.filename}</a><br/>')
-        file_meta = asdict(file)
-        file_meta["path"] = file_path_quoted
-        metadata.append(file_meta)
-    index_str = INDEX_HTML_TEMPLATE.format(items="\n".join(href_tags), comment=comment)
-    metadata_str = json.dumps(metadata, indent=2)
-    return index_str, metadata_str
-
-
-def generate_index_and_metadata(
-    whl_files: list[str],
-    wheel_base_dir: Path,
-    index_base_dir: Path,
-    default_variant: str | None = None,
-    alias_to_default: str | None = None,
-    comment: str = "",
-):
-    """
-    Generate index for all wheel files.
-
-    Args:
-        whl_files (list[str]): List of wheel files (must be directly under `wheel_base_dir`).
-        wheel_base_dir (Path): Base directory for wheel files.
-        index_base_dir (Path): Base directory to store index files.
-        default_variant (str | None): The default variant name, if any.
-        alias_to_default (str | None): Alias variant name for the default variant, if any.
-        comment (str | None): Optional comment to include in the generated HTML files.
-
-    First, parse all wheel files to extract metadata.
-    We need to collect all wheel files for each variant, and generate an index for it (in a sub-directory).
-    The index for the default variant (if any) is generated in the root index directory.
-
-    If `default_variant` is provided, all wheels must have variant suffixes, and the default variant index
-    is purely a copy of the corresponding variant index, with only the links adjusted.
-    Otherwise, all wheels without variant suffixes are treated as the default variant.
-
-    If `alias_to_default` is provided, an additional alias sub-directory is created, it has the same content
-    as the default variant index, but the links are adjusted accordingly.
-
-    Index directory structure:
-        index_base_dir/ (hosted at wheels.vllm.ai/{nightly,$commit,$version}/)
-            index.html  # project list, linking to "vllm/" and other packages, and all variant sub-directories
-            vllm/
-                index.html # package index, pointing to actual files in wheel_base_dir (relative path)
-                metadata.json # machine-readable metadata for all wheels in this package
-            cpu/ # cpu variant sub-directory
-                index.html
-                vllm/
-                    index.html
-                    metadata.json
-            cu129/ # cu129 is actually the alias to default variant
-                index.html
-                vllm/
-                    index.html
-                    metadata.json
-            cu130/ # cu130 variant sub-directory
-                index.html
-                vllm/
-                    index.html
-                    metadata.json
-            ...
-
-    metadata.json stores a dump of all wheel files' metadata in a machine-readable format:
-        [
-            {
-                "package_name": "vllm",
-                "version": "0.10.2rc2",
-                "build_tag": null,
-                "python_tag": "cp38",
-                "abi_tag": "abi3",
-                "platform_tag": "manylinux2014_aarch64",
-                "variant": "cu129",
-                "filename": "vllm-0.10.2rc2+cu129-cp38-abi3-manylinux2014_aarch64.whl",
-                "path": "../vllm-0.10.2rc2%2Bcu129-cp38-abi3-manylinux2014_aarch64.whl" # to be concatenated with the directory URL and URL-encoded
-            },
-            ...
-        ]
-    """
-
-    parsed_files = [parse_from_filename(f) for f in whl_files]
-
-    if not parsed_files:
-        print("No wheel files found, skipping index generation.")
-        return
-
-    # Group by variant
-    variant_to_files: dict[str, list[WheelFileInfo]] = {}
-    for file in parsed_files:
-        variant = file.variant or "default"
-        if variant not in variant_to_files:
-            variant_to_files[variant] = []
-        variant_to_files[variant].append(file)
-
-    print(f"Found variants: {list(variant_to_files.keys())}")
-
-    # sanity check for default variant
-    if default_variant:
-        if "default" in variant_to_files:
-            raise ValueError(
-                "All wheel files must have variant suffixes when `default_variant` is specified."
-            )
-        if default_variant not in variant_to_files:
-            raise ValueError(
-                f"Default variant '{default_variant}' not found among wheel files."
-            )
-
-    if alias_to_default:
-        if "default" not in variant_to_files:
-            # e.g. only some wheels are uploaded to S3 currently
-            print(
-                "[WARN] Alias to default variant specified, but no default variant found."
-            )
-        elif alias_to_default in variant_to_files:
-            raise ValueError(
-                f"Alias variant name '{alias_to_default}' already exists among wheel files."
-            )
-        else:
-            variant_to_files[alias_to_default] = variant_to_files["default"].copy()
-            print(f"Alias variant '{alias_to_default}' created for default variant.")
-
-    # Generate comment in HTML header
-    comment_str = f" ({comment})" if comment else ""
-    comment_tmpl = f"Generated on {datetime.now().isoformat()}{comment_str}"
-
-    # Generate index for each variant
-    subdir_names = set()
-    for variant, files in variant_to_files.items():
-        if variant == "default":
-            variant_dir = index_base_dir
-        else:
-            variant_dir = index_base_dir / variant
-            subdir_names.add(variant)
-
-        variant_dir.mkdir(parents=True, exist_ok=True)
-
-        # gather all package names in this variant
-        packages = set(f.package_name for f in files)
-        if variant == "default":
-            # these packages should also appear in the "project list"
-            # generate after all variants are processed
-            subdir_names = subdir_names.union(packages)
-        else:
-            # generate project list for this variant directly
-            project_list_str = generate_project_list(sorted(packages), comment_tmpl)
-            with open(variant_dir / "index.html", "w") as f:
-                f.write(project_list_str)
-
-        for package in packages:
-            # filter files belonging to this package only
-            package_files = [f for f in files if f.package_name == package]
-            package_dir = variant_dir / package
-            package_dir.mkdir(parents=True, exist_ok=True)
-            index_str, metadata_str = generate_package_index_and_metadata(
-                package_files, wheel_base_dir, package_dir, comment
-            )
-            with open(package_dir / "index.html", "w") as f:
-                f.write(index_str)
-            with open(package_dir / "metadata.json", "w") as f:
-                f.write(metadata_str)
-
-    # Generate top-level project list index
-    project_list_str = generate_project_list(sorted(subdir_names), comment_tmpl)
-    with open(index_base_dir / "index.html", "w") as f:
-        f.write(project_list_str)
-
-
-if __name__ == "__main__":
-    """
-    Arguments:
-        --version <version> : version string for the current build (e.g., commit hash)
-        --wheel-dir <wheel_directory> : directory containing wheel files (default to be same as `version`)
-        --current-objects <path_to_json> : path to JSON file containing current S3 objects listing in this version directory
-        --output-dir <output_directory> : directory to store generated index files
-        --alias-to-default <alias_variant_name> : (optional) alias variant name for the default variant
-        --comment <comment_string> : (optional) comment string to include in generated HTML files
-    """
-
-    parser = argparse.ArgumentParser(
-        description="Process nightly build wheel files to generate indices."
-    )
-    parser.add_argument(
-        "--version",
-        type=str,
-        required=True,
-        help="Version string for the current build (e.g., commit hash)",
-    )
-    parser.add_argument(
-        "--current-objects",
-        type=str,
-        required=True,
-        help="Path to JSON file containing current S3 objects listing in this version directory",
-    )
-    parser.add_argument(
-        "--output-dir",
-        type=str,
-        required=True,
-        help="Directory to store generated index files",
-    )
-    parser.add_argument(
-        "--wheel-dir",
-        type=str,
-        default=None,
-        help="Directory containing wheel files (default to be same as `version`)",
-    )
-    parser.add_argument(
-        "--alias-to-default",
-        type=str,
-        default=None,
-        help="Alias variant name for the default variant",
-    )
-    parser.add_argument(
-        "--comment",
-        type=str,
-        default="",
-        help="Optional comment string to include in generated HTML files",
-    )
-
-    args = parser.parse_args()
-
-    version = args.version
-    if "/" in version or "\\" in version:
-        raise ValueError("Version string must not contain slashes.")
-    current_objects_path = Path(args.current_objects)
-    output_dir = Path(args.output_dir)
-    if not output_dir.exists():
-        output_dir.mkdir(parents=True, exist_ok=True)
-
-    # Read current objects JSON
-    with open(current_objects_path) as f:
-        current_objects: dict[str, list[dict[str, Any]]] = json.load(f)
-
-    # current_objects looks like from list_objects_v2 S3 API:
-    """
-    "Contents": [
-        {
-            "Key": "e2f56c309d2a28899c68975a7e104502d56deb8f/vllm-0.11.2.dev363+ge2f56c309-cp38-abi3-manylinux1_x86_64.whl",
-            "LastModified": "2025-11-28T14:00:32+00:00",
-            "ETag": "\"37a38339c7cdb61ca737021b968075df-52\"",
-            "ChecksumAlgorithm": [
-                "CRC64NVME"
-            ],
-            "ChecksumType": "FULL_OBJECT",
-            "Size": 435649349,
-            "StorageClass": "STANDARD"
-        },
-        ...
-    ]
-    """
-
-    # Extract wheel file keys
-    wheel_files = []
-    for item in current_objects.get("Contents", []):
-        key: str = item["Key"]
-        if key.endswith(".whl"):
-            wheel_files.append(key.split("/")[-1])  # only the filename is used
-
-    print(f"Found {len(wheel_files)} wheel files for version {version}: {wheel_files}")
-
-    # keep only "official" files for a non-nightly version (specified by cli args)
-    PY_VERSION_RE = re.compile(r"^\d+\.\d+\.\d+([a-zA-Z0-9.+-]*)?$")
-    if PY_VERSION_RE.match(version):
-        # upload-wheels.sh ensures no "dev" is in args.version
-        wheel_files = list(
-            filter(lambda x: version in x and "dev" not in x, wheel_files)
-        )
-        print(f"Non-nightly version detected, wheel files used: {wheel_files}")
-    else:
-        print("Nightly version detected, keeping all wheel files.")
-
-    # Generate index and metadata, assuming wheels and indices are stored as:
-    # s3://vllm-wheels/{wheel_dir}/<wheel files>
-    # s3://vllm-wheels/<anything>/<index files>
-    wheel_dir = args.wheel_dir or version
-    wheel_base_dir = Path(output_dir).parent / wheel_dir.strip().rstrip("/")
-    index_base_dir = Path(output_dir)
-
-    generate_index_and_metadata(
-        whl_files=wheel_files,
-        wheel_base_dir=wheel_base_dir,
-        index_base_dir=index_base_dir,
-        default_variant=None,
-        alias_to_default=args.alias_to_default,
-        comment=args.comment.strip(),
-    )
-    print(f"Successfully generated index and metadata in {output_dir}")
@@ -59,7 +59,7 @@ while true; do
        fi
 done

-echo "--- Pulling container"
+echo "--- Pulling container" 
 image_name="rocm/vllm-ci:${BUILDKITE_COMMIT}"
 container_name="rocm_${BUILDKITE_COMMIT}_$(tr -dc A-Za-z0-9 < /dev/urandom | head -c 10; echo)"
 docker pull "${image_name}"
@@ -78,13 +78,17 @@ HF_MOUNT="/root/.cache/huggingface"
 commands=$@
 echo "Commands:$commands"

-commands=${commands//"pytest -v -s basic_correctness/test_basic_correctness.py"/"pytest -v -s basic_correctness/test_basic_correctness.py"}
+if [[ $commands == *"pytest -v -s basic_correctness/test_basic_correctness.py"* ]]; then
+  commands=${commands//"pytest -v -s basic_correctness/test_basic_correctness.py"/"VLLM_USE_TRITON_FLASH_ATTN=0 pytest -v -s basic_correctness/test_basic_correctness.py"}
+fi

 if [[ $commands == *"pytest -v -s models/test_registry.py"* ]]; then
  commands=${commands//"pytest -v -s models/test_registry.py"/"pytest -v -s models/test_registry.py -k 'not BambaForCausalLM and not GritLM and not Mamba2ForCausalLM and not Zamba2ForCausalLM'"}
 fi

-commands=${commands//"pytest -v -s compile/test_basic_correctness.py"/"pytest -v -s compile/test_basic_correctness.py"}
+if [[ $commands == *"pytest -v -s compile/test_basic_correctness.py"* ]]; then
+  commands=${commands//"pytest -v -s compile/test_basic_correctness.py"/"VLLM_USE_TRITON_FLASH_ATTN=0 pytest -v -s compile/test_basic_correctness.py"}
+fi

 if [[ $commands == *"pytest -v -s lora"* ]]; then
  commands=${commands//"pytest -v -s lora"/"VLLM_ROCM_CUSTOM_PAGED_ATTN=0 pytest -v -s lora"}
@@ -141,6 +145,7 @@ if [[ $commands == *" entrypoints/openai "* ]]; then
  --ignore=entrypoints/openai/test_audio.py \
  --ignore=entrypoints/openai/test_shutdown.py \
  --ignore=entrypoints/openai/test_completion.py \
+  --ignore=entrypoints/openai/test_sleep.py \
  --ignore=entrypoints/openai/test_models.py \
  --ignore=entrypoints/openai/test_lora_adapters.py \
  --ignore=entrypoints/openai/test_return_tokens_as_ids.py \
@@ -168,28 +173,19 @@ fi
 PARALLEL_JOB_COUNT=8
 MYPYTHONPATH=".."

-# Test that we're launching on the machine that has
-# proper access to GPUs
-render_gid=$(getent group render | cut -d: -f3)
-if [[ -z "$render_gid" ]]; then
-  echo "Error: 'render' group not found. This is required for GPU access." >&2
-  exit 1
-fi
-
-# check if the command contains shard flag, we will run all shards in parallel because the host have 8 GPUs.
+# check if the command contains shard flag, we will run all shards in parallel because the host have 8 GPUs. 
 if [[ $commands == *"--shard-id="* ]]; then
-  # assign job count as the number of shards used
-  commands=$(echo "$commands" | sed -E "s/--num-shards[[:blank:]]*=[[:blank:]]*[0-9]*/--num-shards=${PARALLEL_JOB_COUNT} /g" | sed 's/ \\ / /g')
+  # assign job count as the number of shards used   
+  commands=${commands//"--num-shards= "/"--num-shards=${PARALLEL_JOB_COUNT} "}
  for GPU in $(seq 0 $(($PARALLEL_JOB_COUNT-1))); do
    # assign shard-id for each shard
-    commands_gpu=$(echo "$commands" | sed -E "s/--shard-id[[:blank:]]*=[[:blank:]]*[0-9]*/--shard-id=${GPU} /g" | sed 's/ \\ / /g')
+    commands_gpu=${commands//"--shard-id= "/"--shard-id=${GPU} "}
    echo "Shard ${GPU} commands:$commands_gpu"
    echo "Render devices: $BUILDKITE_AGENT_META_DATA_RENDER_DEVICES"
    docker run \
        --device /dev/kfd $BUILDKITE_AGENT_META_DATA_RENDER_DEVICES \
        --network=host \
        --shm-size=16gb \
-        --group-add "$render_gid" \
        --rm \
        -e HIP_VISIBLE_DEVICES="${GPU}" \
        -e HF_TOKEN \
@@ -221,8 +217,8 @@ else
          --device /dev/kfd $BUILDKITE_AGENT_META_DATA_RENDER_DEVICES \
          --network=host \
          --shm-size=16gb \
-          --group-add "$render_gid" \
          --rm \
+          -e HIP_VISIBLE_DEVICES=0 \
          -e HF_TOKEN \
          -e AWS_ACCESS_KEY_ID \
          -e AWS_SECRET_ACCESS_KEY \
@@ -1,68 +0,0 @@
-#!/bin/bash
-
-# This script build the CPU docker image and run the offline inference inside the container.
-# It serves a sanity check for compilation and basic model usage.
-set -ex
-
-# allow to bind to different cores
-CORE_RANGE=${CORE_RANGE:-0-16}
-OMP_CORE_RANGE=${OMP_CORE_RANGE:-0-16}
-
-export CMAKE_BUILD_PARALLEL_LEVEL=16
-
-# Setup cleanup
-remove_docker_container() {
-    set -e;
-    docker rm -f cpu-test || true;
-}
-trap remove_docker_container EXIT
-remove_docker_container
-
-# Try building the docker image
-docker build --tag cpu-test --target vllm-test -f docker/Dockerfile.cpu .
-
-# Run the image
-docker run -itd --cpuset-cpus="$CORE_RANGE" --entrypoint /bin/bash -v ~/.cache/huggingface:/root/.cache/huggingface -e HF_TOKEN --env VLLM_CPU_KVCACHE_SPACE=16 --env VLLM_CPU_CI_ENV=1 -e E2E_OMP_THREADS="$OMP_CORE_RANGE" --shm-size=4g --name cpu-test cpu-test
-
-function cpu_tests() {
-  set -e
-
-  docker exec cpu-test bash -c "
-    set -e
-    pip list"
-
-  # offline inference
-  docker exec cpu-test bash -c "
-    set -e
-    python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m"
-
-  # Run model tests
-  docker exec cpu-test bash -c "
-    set -e
-    pytest -x -v -s tests/models/multimodal/generation/test_whisper.py -m cpu_model"
-
-  # Run kernel tests
-  docker exec cpu-test bash -c "
-    set -e
-    pytest -x -v -s tests/kernels/test_onednn.py
-    pytest -x -v -s tests/kernels/attention/test_cpu_attn.py
-    pytest -x -v -s tests/kernels/moe/test_moe.py -k test_cpu_fused_moe_basic"
-
-  # basic online serving
-  docker exec cpu-test bash -c '
-    set -e
-    VLLM_CPU_OMP_THREADS_BIND=$E2E_OMP_THREADS vllm serve Qwen/Qwen3-0.6B --max-model-len 2048 &
-    server_pid=$!
-    timeout 600 bash -c "until curl localhost:8000/v1/models; do sleep 1; done" || exit 1
-    vllm bench serve \
-      --backend vllm \
-      --dataset-name random \
-      --model Qwen/Qwen3-0.6B \
-      --num-prompts 20 \
-      --endpoint /v1/completions
-    kill -s SIGTERM $server_pid &'
-}
-
-# All of CPU tests are expected to be finished less than 40 mins.
-export -f cpu_tests
-timeout 2h bash -c cpu_tests
@@ -25,22 +25,20 @@ function cpu_tests() {

  # offline inference
  podman exec -it "$container_id" bash -c "
-    export TORCH_COMPILE_DISABLE=1
    set -xve
    python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m" >> $HOME/test_basic.log

  # Run basic model test
  podman exec -it "$container_id" bash -c "
-    export TORCH_COMPILE_DISABLE=1
    set -evx
    pip install pytest pytest-asyncio einops peft Pillow soundfile transformers_stream_generator matplotlib
-    pip install sentence-transformers datamodel_code_generator tblib 
+    pip install sentence-transformers datamodel_code_generator

    # Note: disable Bart until supports V1
    # pytest -v -s tests/models/language/generation/test_bart.py -m cpu_model
-    pytest -v -s tests/models/language/generation/test_common.py::test_models[False-False-5-32-openai-community/gpt2]
-    pytest -v -s tests/models/language/generation/test_common.py::test_models[False-False-5-32-facebook/opt-125m]
-    pytest -v -s tests/models/language/generation/test_common.py::test_models[False-False-5-32-google/gemma-1.1-2b-it]
+    pytest -v -s tests/models/language/generation/test_common.py::test_models[False-5-32-openai-community/gpt2]
+    pytest -v -s tests/models/language/generation/test_common.py::test_models[False-5-32-facebook/opt-125m]
+    pytest -v -s tests/models/language/generation/test_common.py::test_models[False-5-32-google/gemma-1.1-2b-it]
    pytest -v -s tests/models/language/pooling/test_classification.py::test_models[float-jason9693/Qwen2.5-1.5B-apeach]
    # TODO: Below test case tests/models/language/pooling/test_embedding.py::test_models[True-ssmits/Qwen2-7B-Instruct-embed-base] fails on ppc64le. Disabling it for time being.
    # pytest -v -s tests/models/language/pooling/test_embedding.py -m cpu_model" >> $HOME/test_rest.log
@@ -21,8 +21,8 @@ trap remove_docker_container EXIT
 remove_docker_container

 # Try building the docker image
-numactl -C "$CORE_RANGE" -N "$NUMA_NODE" docker build --progress plain --tag cpu-test-"$NUMA_NODE" --target vllm-test -f docker/Dockerfile.cpu .
-numactl -C "$CORE_RANGE" -N "$NUMA_NODE" docker build --progress plain --build-arg VLLM_CPU_DISABLE_AVX512="true" --tag cpu-test-"$NUMA_NODE"-avx2 --target vllm-test -f docker/Dockerfile.cpu .
+numactl -C "$CORE_RANGE" -N "$NUMA_NODE" docker build --tag cpu-test-"$NUMA_NODE" --target vllm-test -f docker/Dockerfile.cpu .
+numactl -C "$CORE_RANGE" -N "$NUMA_NODE" docker build --build-arg VLLM_CPU_DISABLE_AVX512="true" --tag cpu-test-"$NUMA_NODE"-avx2 --target vllm-test -f docker/Dockerfile.cpu .

 # Run the image, setting --shm-size=4g for tensor parallel.
 docker run -itd --cpuset-cpus="$CORE_RANGE" --cpuset-mems="$NUMA_NODE" --entrypoint /bin/bash -v ~/.cache/huggingface:/root/.cache/huggingface --privileged=true -e HF_TOKEN --env VLLM_CPU_KVCACHE_SPACE=16 --env VLLM_CPU_CI_ENV=1 -e E2E_OMP_THREADS="$OMP_CORE_RANGE" --shm-size=4g --name cpu-test-"$NUMA_NODE" cpu-test-"$NUMA_NODE"
@@ -49,8 +49,6 @@ function cpu_tests() {
  # Run kernel tests
  docker exec cpu-test-"$NUMA_NODE" bash -c "
    set -e
-    pytest -x -v -s tests/kernels/attention/test_cpu_attn.py
-    pytest -x -v -s tests/kernels/moe/test_cpu_fused_moe.py
    pytest -x -v -s tests/kernels/test_onednn.py"

  # Run basic model test
@@ -74,17 +72,18 @@ function cpu_tests() {
    pytest -x -s -v \
    tests/quantization/test_compressed_tensors.py::test_compressed_tensors_w8a8_logprobs"

-  # Run AWQ/GPTQ test
-  docker exec cpu-test-"$NUMA_NODE" bash -c "
-    set -e
-    pytest -x -s -v \
-    tests/quantization/test_cpu_wna16.py"
+  # Note: disable it until supports V1
+  # Run AWQ test
+  # docker exec cpu-test-"$NUMA_NODE" bash -c "
+  #   set -e
+  #   VLLM_USE_V1=0 pytest -x -s -v \
+  #   tests/quantization/test_ipex_quant.py"

  # Run multi-lora tests
  docker exec cpu-test-"$NUMA_NODE" bash -c "
    set -e
    pytest -x -s -v \
-    tests/lora/test_qwenvl.py"
+    tests/lora/test_qwen2vl.py"

  # online serving: tp+pp
  docker exec cpu-test-"$NUMA_NODE" bash -c '
@@ -117,4 +116,4 @@ function cpu_tests() {

 # All of CPU tests are expected to be finished less than 40 mins.
 export -f cpu_tests
-timeout 2.5h bash -c "cpu_tests $CORE_RANGE $NUMA_NODE"
+timeout 2h bash -c "cpu_tests $CORE_RANGE $NUMA_NODE"
@@ -74,7 +74,6 @@ FROM ${BASE_IMAGE_NAME}

 # Define environments
 ENV DEBIAN_FRONTEND=noninteractive
-ENV SOC_VERSION="ascend910b1"

 RUN pip config set global.index-url http://cache-service-vllm.nginx-pypi-cache.svc.cluster.local:${PYPI_CACHE_PORT}/pypi/simple && \
    pip config set global.trusted-host cache-service-vllm.nginx-pypi-cache.svc.cluster.local && \
@@ -61,7 +61,7 @@ echo "Results will be stored in: $RESULTS_DIR"
 echo "--- Installing Python dependencies ---"
 python3 -m pip install --progress-bar off git+https://github.com/thuml/depyf.git \
    && python3 -m pip install --progress-bar off pytest pytest-asyncio tpu-info \
-    && python3 -m pip install --progress-bar off "lm-eval[api]>=0.4.9.2" \
+    && python3 -m pip install --progress-bar off "lm-eval @ git+https://github.com/EleutherAI/lm-evaluation-harness.git@206b7722158f58c35b7ffcd53b035fdbdda5126d" \
    && python3 -m pip install --progress-bar off hf-transfer tblib==3.1.0
 echo "--- Python dependencies installed ---"

@@ -61,7 +61,7 @@ echo "Results will be stored in: $RESULTS_DIR"
 echo "--- Installing Python dependencies ---"
 python3 -m pip install --progress-bar off git+https://github.com/thuml/depyf.git \
    && python3 -m pip install --progress-bar off pytest pytest-asyncio tpu-info \
-    && python3 -m pip install --progress-bar off "lm-eval[api]>=0.4.9.2" \
+    && python3 -m pip install --progress-bar off "lm-eval @ git+https://github.com/EleutherAI/lm-evaluation-harness.git@206b7722158f58c35b7ffcd53b035fdbdda5126d" \
    && python3 -m pip install --progress-bar off hf-transfer tblib==3.1.0
 echo "--- Python dependencies installed ---"

@@ -35,11 +35,10 @@ docker run \
    echo $ZE_AFFINITY_MASK
    pip install tblib==3.1.0
    python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m --block-size 64 --enforce-eager
-    python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m --block-size 64 -O3 -cc.cudagraph_mode=NONE
+    python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m --block-size 64 -O3 -O.cudagraph_mode=NONE
    python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m --block-size 64 --enforce-eager -tp 2 --distributed-executor-backend ray
    python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m --block-size 64 --enforce-eager -tp 2 --distributed-executor-backend mp
-    python3 examples/offline_inference/basic/generate.py --model Intel/Qwen2.5-0.5B-W4A16-G128-AutoRound-LLMC-TEST-ONLY --enforce-eager
-    python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m --block-size 64 --enforce-eager --attention-backend=TRITON_ATTN
+    VLLM_ATTENTION_BACKEND=TRITON_ATTN python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m --block-size 64 --enforce-eager
    cd tests
    pytest -v -s v1/core
    pytest -v -s v1/engine
@@ -47,6 +46,6 @@ docker run \
    pytest -v -s v1/worker --ignore=v1/worker/test_gpu_model_runner.py
    pytest -v -s v1/structured_output
    pytest -v -s v1/spec_decode --ignore=v1/spec_decode/test_max_len.py --ignore=v1/spec_decode/test_tree_attention.py --ignore=v1/spec_decode/test_speculators_eagle3.py
-    pytest -v -s v1/kv_connector/unit --ignore=v1/kv_connector/unit/test_multi_connector.py --ignore=v1/kv_connector/unit/test_nixl_connector.py --ignore=v1/kv_connector/unit/test_example_connector.py --ignore=v1/kv_connector/unit/test_lmcache_integration.py
+    pytest -v -s v1/kv_connector/unit --ignore=v1/kv_connector/unit/test_multi_connector.py --ignore=v1/kv_connector/unit/test_nixl_connector.py --ignore=v1/kv_connector/unit/test_shared_storage_connector.py
    pytest -v -s v1/test_serial_utils.py
 '
@@ -12,11 +12,6 @@ REPO_ROOT="$(cd "${SCRIPT_DIR}/../.." && pwd)"
 PRIME_RL_REPO="https://github.com/PrimeIntellect-ai/prime-rl.git"
 PRIME_RL_DIR="${REPO_ROOT}/prime-rl"

-if command -v rocm-smi &> /dev/null || command -v rocminfo &> /dev/null; then
-    echo "AMD GPU detected. Prime-RL currently only supports NVIDIA. Skipping..."
-    exit 0
-fi
-
 echo "Setting up Prime-RL integration test environment..."

 # Clean up any existing Prime-RL directory
@@ -17,17 +17,7 @@ wait_for_server() {
 }

 MODEL="deepseek-ai/DeepSeek-V2-lite"
-
-# Set BACKENDS based on platform
-if command -v rocm-smi &> /dev/null || [[ -d /opt/rocm ]] || [[ -n "${ROCM_PATH:-}" ]]; then
-  # ROCm platform
-  BACKENDS=("allgather_reducescatter")
-  # Disable MOE padding for ROCm since it is causing eplb to fail
-  export VLLM_ROCM_MOE_PADDING=0
-else
-  # Non-ROCm platform (CUDA/other)
-  BACKENDS=("deepep_high_throughput" "deepep_low_latency")
-fi
+BACKENDS=("deepep_high_throughput" "deepep_low_latency")

 cleanup() {
  if [[ -n "${SERVER_PID:-}" ]] && kill -0 "${SERVER_PID}" 2>/dev/null; then
@@ -2,9 +2,9 @@
 set -euxo pipefail

 # args: [THRESHOLD] [NUM_QUESTIONS] [START_PORT]
-THRESHOLD=${1:-0.25}
+THRESHOLD=${1:-0.8}
 NUM_Q=${2:-1319}
-PORT=${3:-8040}
+PORT=${3:-8020}
 OUT_DIR=${OUT_DIR:-/tmp/vllm-scheduled}
 mkdir -p "${OUT_DIR}"

@@ -16,18 +16,8 @@ wait_for_server() {
    done'
 }

-MODEL="Qwen/Qwen3-Next-80B-A3B-Instruct"
-
-# Set BACKENDS based on platform
-if command -v rocm-smi &> /dev/null || [[ -d /opt/rocm ]] || [[ -n "${ROCM_PATH:-}" ]]; then
-  # ROCm platform
-  BACKENDS=("allgather_reducescatter")
-  # Disable MOE padding for ROCm since it is causing eplb to fail
-  export VLLM_ROCM_MOE_PADDING=0
-else
-  # Non-ROCm platform (CUDA/other)
-  BACKENDS=("deepep_high_throughput" "deepep_low_latency")
-fi
+MODEL="QWen/Qwen3-30B-A3B-FP8"
+BACKENDS=("deepep_high_throughput" "deepep_low_latency")

 cleanup() {
  if [[ -n "${SERVER_PID:-}" ]] && kill -0 "${SERVER_PID}" 2>/dev/null; then
@@ -43,17 +33,14 @@ trap cleanup EXIT

 for BACK in "${BACKENDS[@]}"; do
  VLLM_DEEP_GEMM_WARMUP=skip \
+  VLLM_ALL2ALL_BACKEND=$BACK \
  vllm serve "$MODEL" \
    --enforce-eager \
-    --tensor-parallel-size 4 \
+    --tensor-parallel-size 2 \
+    --data-parallel-size 2 \
    --enable-expert-parallel \
-    --enable-eplb \
-    --all2all-backend $BACK \
-    --eplb-config '{"window_size":200,"step_interval":600,"use_async":true}' \
-    --speculative-config '{"method":"qwen3_next_mtp","num_speculative_tokens":1}' \
    --trust-remote-code \
    --max-model-len 2048 \
-    --gpu-memory-utilization 0.9 \
    --port $PORT &
  SERVER_PID=$!
  wait_for_server $PORT
@@ -1,74 +0,0 @@
-#!/usr/bin/env bash
-set -euxo pipefail
-
-# args: [THRESHOLD] [NUM_QUESTIONS] [START_PORT] [DATA_PARALLEL_SIZE] [TENSOR_PARALLEL_SIZE]
-THRESHOLD=${1:-0.8}
-NUM_Q=${2:-1319}
-PORT=${3:-8020}
-DATA_PARALLEL_SIZE=${4:-2}
-TENSOR_PARALLEL_SIZE=${5:-2}
-OUT_DIR=${OUT_DIR:-/tmp/vllm-scheduled}
-mkdir -p "${OUT_DIR}"
-
-wait_for_server() {
-  local port=$1
-  timeout 600 bash -c '
-    until curl -sf "http://127.0.0.1:'"$port"'/health" > /dev/null; do
-      sleep 1
-    done'
-}
-
-MODEL="QWen/Qwen3-30B-A3B-FP8"
-# Set BACKENDS based on platform
-if command -v rocm-smi &> /dev/null || [[ -d /opt/rocm ]] || [[ -n "${ROCM_PATH:-}" ]]; then
-  # ROCm platform
-  BACKENDS=("allgather_reducescatter")
-  # Disable MOE padding for ROCm since it is causing eplb to fail
-  export VLLM_ROCM_MOE_PADDING=0
-else
-  # Non-ROCm platform (CUDA/other)
-  BACKENDS=("deepep_high_throughput" "deepep_low_latency")
-fi
-
-cleanup() {
-  if [[ -n "${SERVER_PID:-}" ]] && kill -0 "${SERVER_PID}" 2>/dev/null; then
-    kill "${SERVER_PID}" 2>/dev/null || true
-    for _ in {1..20}; do
-      kill -0 "${SERVER_PID}" 2>/dev/null || break
-      sleep 0.5
-    done
-    kill -9 "${SERVER_PID}" 2>/dev/null || true
-  fi
-}
-trap cleanup EXIT
-
-for BACK in "${BACKENDS[@]}"; do
-  VLLM_DEEP_GEMM_WARMUP=skip \
-  vllm serve "$MODEL" \
-    --enforce-eager \
-    --enable-eplb \
-    --all2all-backend $BACK \
-    --eplb-config '{"window_size":10, "step_interval":100, "num_redundant_experts":0, "log_balancedness":true}' \
-    --tensor-parallel-size ${TENSOR_PARALLEL_SIZE} \
-    --data-parallel-size ${DATA_PARALLEL_SIZE} \
-    --enable-expert-parallel \
-    --trust-remote-code \
-    --max-model-len 2048 \
-    --port $PORT &
-  SERVER_PID=$!
-  wait_for_server $PORT
-
-  TAG=$(echo "$MODEL" | tr '/: \\n' '_____')
-  OUT="${OUT_DIR}/${TAG}_${BACK}.json"
-  python3 tests/evals/gsm8k/gsm8k_eval.py --host http://127.0.0.1 --port $PORT --num-questions ${NUM_Q} --save-results ${OUT}
-  python3 - <<PY
-import json; acc=json.load(open('${OUT}'))['accuracy']
-print(f"${MODEL} ${BACK}: accuracy {acc:.3f}")
-assert acc >= ${THRESHOLD}, f"${MODEL} ${BACK} accuracy {acc}"
-PY
-
-  cleanup
-  SERVER_PID=
-  sleep 1
-  PORT=$((PORT+1))
-done
@@ -2,28 +2,6 @@

 set -ex

-# ======== part 0: setup ========
-
-BUCKET="vllm-wheels"
-INDICES_OUTPUT_DIR="indices"
-DEFAULT_VARIANT_ALIAS="cu129" # align with vLLM_MAIN_CUDA_VERSION in vllm/envs.py
-PYTHON=${PYTHON_PROG:=python3} # try to read from env var, otherwise use python3
-SUBPATH=$BUILDKITE_COMMIT
-S3_COMMIT_PREFIX="s3://$BUCKET/$SUBPATH/"
-
-# detect if python3.10+ is available
-has_new_python=$($PYTHON -c "print(1 if __import__('sys').version_info >= (3,12) else 0)")
-if [[ "$has_new_python" -eq 0 ]]; then
-    # use new python from docker
-    docker pull python:3-slim
-    PYTHON="docker run --rm -v $(pwd):/app -w /app python:3-slim python3"
-fi
-
-echo "Using python interpreter: $PYTHON"
-echo "Python version: $($PYTHON --version)"
-
-# ========= part 1: collect, rename & upload the wheel ==========
-
 # Assume wheels are in artifacts/dist/*.whl
 wheel_files=(artifacts/dist/*.whl)

@@ -32,77 +10,74 @@ if [[ ${#wheel_files[@]} -ne 1 ]]; then
  echo "Error: Expected exactly one wheel file in artifacts/dist/, but found ${#wheel_files[@]}"
  exit 1
 fi
+
+# Get the single wheel file
 wheel="${wheel_files[0]}"

-# default build image uses ubuntu 20.04, which corresponds to manylinux_2_31
-# we also accept params as manylinux tag
-# refer to https://github.com/mayeut/pep600_compliance?tab=readme-ov-file#acceptable-distros-to-build-wheels
-manylinux_version="${1:-manylinux_2_31}"
+# Detect architecture and rename 'linux' to appropriate manylinux version
+arch=$(uname -m)
+if [[ $arch == "x86_64" ]]; then
+    manylinux_version="manylinux1"
+elif [[ $arch == "aarch64" ]]; then
+    manylinux_version="manylinux2014"
+else
+    echo "Warning: Unknown architecture $arch, using manylinux1 as default"
+    manylinux_version="manylinux1"
+fi

 # Rename 'linux' to the appropriate manylinux version in the wheel filename
-if [[ "$wheel" != *"linux"* ]]; then
-  echo "Error: Wheel filename does not contain 'linux': $wheel"
-  exit 1
-fi
 new_wheel="${wheel/linux/$manylinux_version}"
 mv -- "$wheel" "$new_wheel"
 wheel="$new_wheel"
-echo "Renamed wheel to: $wheel"

 # Extract the version from the wheel
 version=$(unzip -p "$wheel" '**/METADATA' | grep '^Version: ' | cut -d' ' -f2)
-echo "Version in wheel: $version"
-pure_version="${version%%+*}"
-echo "Pure version (without variant): $pure_version"
+echo "Version: $version"

-# copy wheel to its own bucket
-aws s3 cp "$wheel" "$S3_COMMIT_PREFIX"
+normal_wheel="$wheel" # Save the original wheel filename

-# ========= part 2: generate and upload indices ==========
-# generate indices for all existing wheels in the commit directory
-# this script might be run multiple times if there are multiple variants being built
-# so we need to guarantee there is little chance for "TOCTOU" issues
-# i.e., one process is generating indices while another is uploading a new wheel
-# so we need to ensure no time-consuming operations happen below
+# If the version contains "dev", rename it to v1.0.0.dev for consistency
+if [[ $version == *dev* ]]; then
+    suffix="${version##*.}"
+    if [[ $suffix == cu* ]]; then
+        new_version="1.0.0.dev+${suffix}"
+    else
+        new_version="1.0.0.dev"
+    fi
+    new_wheel="${wheel/$version/$new_version}"
+    # use cp to keep both files in the artifacts directory
+    cp -- "$wheel" "$new_wheel"
+    wheel="$new_wheel"
+    version="$new_version"
+fi

-# list all wheels in the commit directory
-echo "Existing wheels on S3:"
-aws s3 ls "$S3_COMMIT_PREFIX"
-obj_json="objects.json"
-aws s3api list-objects-v2 --bucket "$BUCKET" --prefix "$SUBPATH/" --delimiter / --output json > "$obj_json"
-mkdir -p "$INDICES_OUTPUT_DIR"
+# Upload the wheel to S3
+python3 .buildkite/generate_index.py --wheel "$normal_wheel"

-# call script to generate indicies for all existing wheels
-# this indices have relative paths that could work as long as it is next to the wheel directory in s3
-# i.e., the wheels are always in s3://vllm-wheels/<commit>/
-# and indices can be placed in /<commit>/, or /nightly/, or /<version>/
-if [[ ! -z "$DEFAULT_VARIANT_ALIAS" ]]; then
-    alias_arg="--alias-to-default $DEFAULT_VARIANT_ALIAS"
+# generate index for this commit
+aws s3 cp "$wheel" "s3://vllm-wheels/$BUILDKITE_COMMIT/"
+aws s3 cp "$normal_wheel" "s3://vllm-wheels/$BUILDKITE_COMMIT/"
+
+if [[ $normal_wheel == *"cu129"* ]]; then
+    # only upload index.html for cu129 wheels (default wheels) as it
+    # is available on both x86 and arm64
+    aws s3 cp index.html "s3://vllm-wheels/$BUILDKITE_COMMIT/vllm/index.html"
+    aws s3 cp "s3://vllm-wheels/nightly/index.html" "s3://vllm-wheels/$BUILDKITE_COMMIT/index.html"
 else
-    alias_arg=""
+    echo "Skipping index files for non-cu129 wheels"
 fi

-# HACK: we do not need regex module here, but it is required by pre-commit hook
-# To avoid any external dependency, we simply replace it back to the stdlib re module
-sed -i 's/import regex as re/import re/g' .buildkite/scripts/generate-nightly-index.py
-$PYTHON .buildkite/scripts/generate-nightly-index.py --version "$SUBPATH" --current-objects "$obj_json" --output-dir "$INDICES_OUTPUT_DIR" --comment "commit $BUILDKITE_COMMIT" $alias_arg
+# generate index for nightly
+aws s3 cp "$wheel" "s3://vllm-wheels/nightly/"
+aws s3 cp "$normal_wheel" "s3://vllm-wheels/nightly/"

-# copy indices to /<commit>/ unconditionally
-echo "Uploading indices to $S3_COMMIT_PREFIX"
-aws s3 cp --recursive "$INDICES_OUTPUT_DIR/" "$S3_COMMIT_PREFIX"
-
-# copy to /nightly/ only if it is on the main branch and not a PR 
-if [[ "$BUILDKITE_BRANCH" == "main" && "$BUILDKITE_PULL_REQUEST" == "false" ]]; then
-    echo "Uploading indices to overwrite /nightly/"
-    aws s3 cp --recursive "$INDICES_OUTPUT_DIR/" "s3://$BUCKET/nightly/"
+if [[ $normal_wheel == *"cu129"* ]]; then
+    # only upload index.html for cu129 wheels (default wheels) as it
+    # is available on both x86 and arm64
+    aws s3 cp index.html "s3://vllm-wheels/nightly/vllm/index.html"
+else
+    echo "Skipping index files for non-cu129 wheels"
 fi

-# re-generate and copy to /<pure_version>/ only if it does not have "dev" in the version
-if [[ "$version" != *"dev"* ]]; then
-    echo "Re-generating indices for /$pure_version/"
-    rm -rf "$INDICES_OUTPUT_DIR/*"
-    mkdir -p "$INDICES_OUTPUT_DIR"
-    # wheel-dir is overridden to be the commit directory, so that the indices point to the correct wheel path
-    $PYTHON .buildkite/scripts/generate-nightly-index.py --version "$pure_version" --wheel-dir "$SUBPATH" --current-objects "$obj_json" --output-dir "$INDICES_OUTPUT_DIR" --comment "version $pure_version" $alias_arg
-    aws s3 cp --recursive "$INDICES_OUTPUT_DIR/" "s3://$BUCKET/$pure_version/"
-fi
+aws s3 cp "$wheel" "s3://vllm-wheels/$version/"
+aws s3 cp index.html "s3://vllm-wheels/$version/vllm/index.html"
@@ -25,7 +25,6 @@
 #     and $$BUILDKITE_PARALLEL_JOB_COUNT environment variables.
 # working_dir(str): specify the place where the command should execute, default to /vllm-workspace/tests
 # source_file_dependencies(list): the list of prefixes to opt-in the test for, if empty, the test will always run.
-# autorun_on_main (bool): default to false, if true, the test will run automatically when commit is pushed to main branch.

 # When adding a test
 # - If the test belongs to an existing group, add it there
@@ -57,28 +56,22 @@ steps:
  - pytest -v -s -m 'not cpu_test' multimodal
  - pytest -v -s utils_

- label: Async Engine, Inputs, Utils, Worker, Config Test (CPU) # 20min
-  timeout_in_minutes: 30
+- label: Async Engine, Inputs, Utils, Worker Test (CPU) # 4 mins
+  timeout_in_minutes: 10
  source_file_dependencies:
  - vllm/
  - tests/test_inputs.py
  - tests/test_outputs.py
  - tests/multimodal
  - tests/standalone_tests/lazy_imports.py
-  - tests/tokenizers_
-  - tests/tool_parsers
  - tests/transformers_utils
-  - tests/config
  no_gpu: true
  commands:
  - python3 standalone_tests/lazy_imports.py
  - pytest -v -s test_inputs.py
  - pytest -v -s test_outputs.py
  - pytest -v -s -m 'cpu_test' multimodal
-  - pytest -v -s tokenizers_
-  - pytest -v -s tool_parsers
  - pytest -v -s transformers_utils
-  - pytest -v -s config

 - label: Python-only Installation Test # 10min
  timeout_in_minutes: 20
@@ -114,7 +107,7 @@ steps:
  - tests/entrypoints/
  commands:
  - pytest -v -s entrypoints/openai/tool_parsers
-  - pytest -v -s entrypoints/ --ignore=entrypoints/llm --ignore=entrypoints/rpc --ignore=entrypoints/sleep --ignore=entrypoints/instrumentator --ignore=entrypoints/openai --ignore=entrypoints/offline_mode --ignore=entrypoints/test_chat_utils.py  --ignore=entrypoints/pooling
+  - pytest -v -s entrypoints/ --ignore=entrypoints/llm --ignore=entrypoints/openai --ignore=entrypoints/offline_mode --ignore=entrypoints/test_chat_utils.py  --ignore=entrypoints/pooling

 - label: Entrypoints Integration Test (LLM) # 30min
  timeout_in_minutes: 40
@@ -132,7 +125,7 @@ steps:
  - pytest -v -s entrypoints/llm/test_generate.py # it needs a clean process
  - pytest -v -s entrypoints/offline_mode # Needs to avoid interference with other tests

- label: Entrypoints Integration Test (API Server 1) # 100min
+- label: Entrypoints Integration Test (API Server) # 100min
  timeout_in_minutes: 130
  mirror_hardwares: [amdexperimental]
  working_dir: "/vllm-workspace/tests"
@@ -144,26 +137,10 @@ steps:
  - tests/entrypoints/test_chat_utils
  commands:
  - export VLLM_WORKER_MULTIPROC_METHOD=spawn
-  - pytest -v -s entrypoints/openai --ignore=entrypoints/openai/test_chat_with_tool_reasoning.py --ignore=entrypoints/openai/test_oot_registration.py --ignore=entrypoints/openai/test_tensorizer_entrypoint.py --ignore=entrypoints/openai/correctness/  --ignore=entrypoints/openai/tool_parsers/
+  - PYTHONPATH=/vllm-workspace pytest -v -s entrypoints/openai/test_collective_rpc.py # PYTHONPATH is needed to import custom Worker extension
+  - pytest -v -s entrypoints/openai --ignore=entrypoints/openai/test_chat_with_tool_reasoning.py --ignore=entrypoints/openai/test_oot_registration.py --ignore=entrypoints/openai/test_tensorizer_entrypoint.py --ignore=entrypoints/openai/correctness/ --ignore=entrypoints/openai/test_collective_rpc.py --ignore=entrypoints/openai/tool_parsers/
  - pytest -v -s entrypoints/test_chat_utils.py

- label: Entrypoints Integration Test (API Server 2)
-  timeout_in_minutes: 50
-  mirror_hardwares: [amdexperimental]
-  working_dir: "/vllm-workspace/tests"
-  fast_check: true
-  torch_nightly: true
-  source_file_dependencies:
-  - vllm/
-  - tests/entrypoints/sleep
-  - tests/entrypoints/rpc
-  - tests/tool_use
-  commands:
-  - export VLLM_WORKER_MULTIPROC_METHOD=spawn
-  - pytest -v -s entrypoints/sleep
-  - PYTHONPATH=/vllm-workspace  pytest -v -s entrypoints/rpc
-  - pytest -v -s tool_use
-
 - label: Entrypoints Integration Test (Pooling)
  timeout_in_minutes: 50
  mirror_hardwares: [amdexperimental]
@@ -187,7 +164,7 @@ steps:
  - tests/distributed/test_utils
  - tests/distributed/test_pynccl
  - tests/distributed/test_events
-  - tests/compile/fullgraph/test_basic_correctness.py
+  - tests/compile/test_basic_correctness
  - examples/offline_inference/rlhf.py
  - examples/offline_inference/rlhf_colocate.py
  - tests/examples/offline_inference/data_parallel.py
@@ -212,13 +189,12 @@ steps:
  # test with internal dp
  - python3 ../examples/offline_inference/data_parallel.py --enforce-eager
  - TP_SIZE=2 DP_SIZE=2 pytest -v -s v1/distributed/test_async_llm_dp.py
-  - TP_SIZE=2 DP_SIZE=2 pytest -v -s v1/distributed/test_eagle_dp.py
  - TP_SIZE=2 DP_SIZE=2 pytest -v -s v1/distributed/test_external_lb_dp.py
  - TP_SIZE=1 DP_SIZE=4 pytest -v -s v1/distributed/test_internal_lb_dp.py
  - TP_SIZE=1 DP_SIZE=4 pytest -v -s v1/distributed/test_hybrid_lb_dp.py
  - pytest -v -s v1/engine/test_engine_core_client.py::test_kv_cache_events_dp
  - pytest -v -s distributed/test_utils.py
-  - pytest -v -s compile/fullgraph/test_basic_correctness.py
+  - pytest -v -s compile/test_basic_correctness.py
  - pytest -v -s distributed/test_pynccl.py
  - pytest -v -s distributed/test_events.py
  - pytest -v -s distributed/test_symm_mem_allreduce.py
@@ -256,8 +232,8 @@ steps:
  commands:
  - pytest -v -s distributed/test_eplb_algo.py

- label: EPLB Execution Test # 10min
-  timeout_in_minutes: 20
+- label: EPLB Execution Test # 5min
+  timeout_in_minutes: 15
  working_dir: "/vllm-workspace/tests"
  num_gpus: 4
  source_file_dependencies:
@@ -265,7 +241,6 @@ steps:
  - tests/distributed/test_eplb_execute.py
  commands:
  - pytest -v -s distributed/test_eplb_execute.py
-  - pytest -v -s distributed/test_eplb_spec_decode.py

 - label: Metrics, Tracing Test # 12min
  timeout_in_minutes: 20
@@ -296,18 +271,21 @@ steps:
  - pytest -v -s test_regression.py
  working_dir: "/vllm-workspace/tests" # optional

- label: Engine Test # 9min
-  timeout_in_minutes: 15
+- label: Engine Test # 25min
+  timeout_in_minutes: 40
  mirror_hardwares: [amdexperimental]
  source_file_dependencies:
  - vllm/
  - tests/engine
+  - tests/tokenization
  - tests/test_sequence
  - tests/test_config
  - tests/test_logger
  - tests/test_vllm_port
  commands:
  - pytest -v -s engine test_sequence.py test_config.py test_logger.py test_vllm_port.py
+  # OOM in the CI unless we run this separately
+  - pytest -v -s tokenization

 - label: V1 Test e2e + engine # 30min
  timeout_in_minutes: 45
@@ -319,10 +297,7 @@ steps:
    # TODO: accuracy does not match, whether setting
    # VLLM_USE_FLASHINFER_SAMPLER or not on H100.
    - pytest -v -s v1/e2e
-    # Run this test standalone for now;
-    # need to untangle use (implicit) use of spawn/fork across the tests.
-    - pytest -v -s v1/engine/test_preprocess_error_handling.py
-    - pytest -v -s v1/engine --ignore v1/engine/test_preprocess_error_handling.py
+    - pytest -v -s v1/engine

 - label: V1 Test entrypoints # 35min
  timeout_in_minutes: 50
@@ -340,7 +315,6 @@ steps:
    - vllm/
    - tests/v1
  commands:
-    - uv pip install --system -r /vllm-workspace/requirements/kv_connectors.txt
    # split the test to avoid interference
    - pytest -v -s -m 'not cpu_test' v1/core
    - pytest -v -s v1/executor
@@ -353,7 +327,6 @@ steps:
    - pytest -v -s -m 'not cpu_test' v1/metrics
    - pytest -v -s v1/test_oracle.py
    - pytest -v -s v1/test_request.py
-    - pytest -v -s v1/test_outputs.py
    # Integration test for streaming correctness (requires special branch).
    - pip install -U git+https://github.com/robertgshaw2-redhat/lm-evaluation-harness.git@streaming-api
    - pytest -v -s entrypoints/openai/correctness/test_lmeval.py::test_lm_eval_accuracy_v1_engine
@@ -367,23 +340,9 @@ steps:
  commands:
    - pytest -v -s v1/attention

- label: Batch Invariance Tests (H100) # 10min
-  timeout_in_minutes: 25
-  gpu: h100
-  source_file_dependencies:
-    - vllm/v1/attention
-    - vllm/model_executor/layers
-    - tests/v1/determinism/
-  commands:
-    - export VLLM_WORKER_MULTIPROC_METHOD=spawn
-    - pip install pytest-timeout pytest-forked
-    - pytest -v -s v1/determinism/test_batch_invariance.py
-    - pytest -v -s v1/determinism/test_rms_norm_batch_invariant.py
-
 - label: V1 Test attention (B200) # 10min
  timeout_in_minutes: 30
  gpu: b200
-  optional: true # TODO(mgoin): disable optional once runner is fixed
  source_file_dependencies:
    - vllm/v1/attention
    - tests/v1/attention
@@ -410,28 +369,23 @@ steps:
  working_dir: "/vllm-workspace/examples"
  source_file_dependencies:
  - vllm/entrypoints
-  - vllm/multimodal
  - examples/
  commands:
    - pip install tensorizer # for tensorizer test
-    # for basic
-    - python3 offline_inference/basic/chat.py
    - python3 offline_inference/basic/generate.py --model facebook/opt-125m
    - python3 offline_inference/basic/generate.py --model meta-llama/Llama-2-13b-chat-hf --cpu-offload-gb 10
+    - python3 offline_inference/basic/chat.py
+    - python3 offline_inference/prefix_caching.py
+    - python3 offline_inference/llm_engine_example.py
+    - python3 offline_inference/audio_language.py --seed 0
+    - python3 offline_inference/vision_language.py --seed 0
+    - python3 offline_inference/vision_language_pooling.py --seed 0
+    - python3 offline_inference/vision_language_multi_image.py --seed 0
+    - python3 others/tensorize_vllm_model.py --model facebook/opt-125m serialize --serialized-directory /tmp/ --suffix v1 && python3 others/tensorize_vllm_model.py --model facebook/opt-125m deserialize --path-to-tensors /tmp/vllm/facebook/opt-125m/v1/model.tensors
+    - python3 offline_inference/encoder_decoder_multimodal.py --model-type whisper --seed 0
    - python3 offline_inference/basic/classify.py
    - python3 offline_inference/basic/embed.py
    - python3 offline_inference/basic/score.py
-    # for multi-modal models
-    - python3 offline_inference/audio_language.py --seed 0
-    - python3 offline_inference/vision_language.py --seed 0
-    - python3 offline_inference/vision_language_multi_image.py --seed 0
-    - python3 offline_inference/encoder_decoder_multimodal.py --model-type whisper --seed 0
-    # for pooling models
-    - python3 pooling/pooling/vision_language_pooling.py --seed 0
-    # for features demo
-    - python3 offline_inference/prefix_caching.py
-    - python3 offline_inference/llm_engine_example.py
-    - python3 others/tensorize_vllm_model.py --model facebook/opt-125m serialize --serialized-directory /tmp/ --suffix v1 && python3 others/tensorize_vllm_model.py --model facebook/opt-125m deserialize --path-to-tensors /tmp/vllm/facebook/opt-125m/v1/model.tensors
    - python3 offline_inference/spec_decode.py --test --method eagle --num_spec_tokens 3 --dataset-name hf --dataset-path philschmid/mt-bench --num-prompts 80 --temp 0 --top-p 1.0 --top-k -1 --tp 1 --enable-chunked-prefill --max-model-len 2048
    # https://github.com/vllm-project/vllm/pull/26682 uses slightly more memory in PyTorch 2.9+ causing this test to OOM in 1xL4 GPU
    - python3 offline_inference/spec_decode.py --test --method eagle3 --num_spec_tokens 3 --dataset-name hf --dataset-path philschmid/mt-bench --num-prompts 80 --temp 0 --top-p 1.0 --top-k -1 --tp 1 --enable-chunked-prefill --max-model-len 1536
@@ -485,14 +439,15 @@ steps:
    - vllm/
    - tests/compile
  commands:
-  # Run unit tests defined directly under compile/,
-  # not including subdirectories, which are usually heavier
-  # tests covered elsewhere.
-  # Use `find` to launch multiple instances of pytest so that
-  # they do not suffer from https://github.com/vllm-project/vllm/issues/28965
-  # However, find does not normally propagate error codes, so we combine it with xargs
-  # (using -0 for proper path handling)
-  - "find compile/ -maxdepth 1 -name 'test_*.py' -print0 | xargs -0 -n1 -I{} pytest -s -v '{}'"
+    - pytest -v -s compile/test_pass_manager.py
+    - pytest -v -s compile/test_fusion.py
+    - pytest -v -s compile/test_fusion_attn.py
+    - pytest -v -s compile/test_functionalization.py
+    - pytest -v -s compile/test_silu_mul_quant_fusion.py
+    - pytest -v -s compile/test_fusion_all_reduce.py
+    - pytest -v -s compile/test_decorator.py
+    - pytest -v -s compile/test_noop_elimination.py
+    - pytest -v -s compile/test_aot_compile.py

 - label: PyTorch Fullgraph Smoke Test # 15min
  timeout_in_minutes: 30
@@ -502,27 +457,19 @@ steps:
  - vllm/
  - tests/compile
  commands:
-  # Run smoke tests under fullgraph directory, except test_full_graph.py
-  # as it is a heavy test that is covered in other steps.
-  # Use `find` to launch multiple instances of pytest so that
-  # they do not suffer from https://github.com/vllm-project/vllm/issues/28965
-  # However, find does not normally propagate error codes, so we combine it with xargs
-  # (using -0 for proper path handling)
-  - "find compile/fullgraph -maxdepth 1 -name 'test_*.py' -not -name 'test_full_graph.py' -print0 | xargs -0 -n1 -I{} pytest -s -v '{}'"
+  - pytest -v -s compile/test_basic_correctness.py
+  - pytest -v -s compile/piecewise/

- label: PyTorch Fullgraph Test # 27min
-  timeout_in_minutes: 40
+- label: PyTorch Fullgraph Test # 22min
+  timeout_in_minutes: 35
  mirror_hardwares: [amdexperimental]
  torch_nightly: true
  source_file_dependencies:
  - vllm/
  - tests/compile
  commands:
-    # fp8 kv scales not supported on sm89, tested on Blackwell instead
-  - pytest -v -s compile/fullgraph/test_full_graph.py -k 'not test_fp8_kv_scale_compile'
-    # Limit to no custom ops to reduce running time
-    # Wrap with quotes to escape yaml and avoid starting -k string with a -
-  - "pytest -v -s compile/distributed/test_fusions_e2e.py -k 'TRITON and not +quant_fp8 and not Llama-4'"
+  - pytest -v -s compile/test_full_graph.py
+  - pytest -v -s compile/test_fusions_e2e.py

 - label: Cudagraph test
  timeout_in_minutes: 20
@@ -594,32 +541,10 @@ steps:
  commands:
    - pytest -v -s kernels/mamba

- label: Kernels DeepGEMM Test (H100)
-  timeout_in_minutes: 45
-  gpu: h100
-  num_gpus: 1
-  source_file_dependencies:
-  - tools/install_deepgemm.sh
-  - vllm/utils/deep_gemm.py
-  - vllm/model_executor/layers/fused_moe
-  - vllm/model_executor/layers/quantization
-  - tests/kernels/quantization/test_block_fp8.py
-  - tests/kernels/moe/test_deepgemm.py
-  - tests/kernels/moe/test_batched_deepgemm.py
-  - tests/kernels/attention/test_deepgemm_attention.py
-  commands:
-    - pytest -v -s kernels/quantization/test_block_fp8.py -k deep_gemm
-    - pytest -v -s kernels/moe/test_deepgemm.py
-    - pytest -v -s kernels/moe/test_batched_deepgemm.py
-    - pytest -v -s kernels/attention/test_deepgemm_attention.py
-
 - label: Model Executor Test # 23min
  timeout_in_minutes: 35
-  torch_nightly: true
  mirror_hardwares: [amdexperimental]
  source_file_dependencies:
-  - vllm/engine/arg_utils.py
-  - vllm/config/model.py
  - vllm/model_executor
  - tests/model_executor
  - tests/entrypoints/openai/test_tensorizer_entrypoint.py
@@ -662,8 +587,7 @@ steps:
  # https://github.com/pytorch/ao/issues/2919, we'll have to skip new torchao tests for now
  # we can only upgrade after this is resolved
  # TODO(jerryzh168): resolve the above comment
-  - uv pip install --system torchao==0.14.1 --index-url https://download.pytorch.org/whl/cu129
-  - uv pip install --system conch-triton-kernels
+  - uv pip install --system torchao==0.13.0 --index-url https://download.pytorch.org/whl/cu129
  - VLLM_TEST_FORCE_LOAD_FORMAT=auto pytest -v -s quantization/ --ignore quantization/test_blackwell_moe.py

 - label: LM Eval Small Models # 53min
@@ -672,9 +596,8 @@ steps:
  source_file_dependencies:
  - csrc/
  - vllm/model_executor/layers/quantization
-  autorun_on_main: true
  commands:
-  - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-small.txt
+  - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-small.txt --tp-size=1

 - label: OpenAI API correctness # 22min
  timeout_in_minutes: 30
@@ -686,6 +609,25 @@ steps:
  commands: # LMEval+Transcription WER check
  - pytest -s entrypoints/openai/correctness/

+- label: OpenAI-Compatible Tool Use # 23 min
+  timeout_in_minutes: 35
+  mirror_hardwares: [amdexperimental]
+  fast_check: false
+  source_file_dependencies:
+    - vllm/
+    - tests/tool_use
+  commands:
+    - pytest -v -s -m 'not cpu_test' tool_use
+
+- label: OpenAI-Compatible Tool Use (CPU) # 5 mins
+  timeout_in_minutes: 10
+  source_file_dependencies:
+    - vllm/
+    - tests/tool_use
+  no_gpu: true
+  commands:
+    - pytest -v -s -m 'cpu_test' tool_use
+
 #####  models test  #####

 - label: Basic Models Tests (Initialization)
@@ -695,7 +637,6 @@ steps:
  source_file_dependencies:
  - vllm/
  - tests/models/test_initialization.py
-  - tests/models/registry.py
  commands:
    # Run a subset of model initialization tests
    - pytest -v -s models/test_initialization.py::test_can_initialize_small_subset
@@ -706,9 +647,7 @@ steps:
  torch_nightly: true
  source_file_dependencies:
  - vllm/model_executor/models/
-  - vllm/transformers_utils/
  - tests/models/test_initialization.py
-  - tests/models/registry.py
  commands:
    # Only when vLLM model source is modified - test initialization of a large
    # subset of supported models (the complement of the small subset in the above
@@ -834,24 +773,14 @@ steps:
  commands:
    - pytest -v -s models/language/pooling_mteb_test

- label: Multi-Modal Processor Test (CPU)
-  timeout_in_minutes: 60
-  source_file_dependencies:
-  - vllm/
-  - tests/models/multimodal
-  no_gpu: true
-  commands:
-    - "pip install git+https://github.com/TIGER-AI-Lab/Mantis.git || echo 'Mantis installation skipped (decord not available on CPU-only environment)'"
-    - pytest -v -s models/multimodal/processing --ignore models/multimodal/processing/test_tensor_schema.py
-
- label: Multi-Modal Processor Test
+- label: Multi-Modal Processor Test # 44min
  timeout_in_minutes: 60
  source_file_dependencies:
  - vllm/
  - tests/models/multimodal
  commands:
    - pip install git+https://github.com/TIGER-AI-Lab/Mantis.git
-    - pytest -v -s models/multimodal/processing/test_tensor_schema.py
+    - pytest -v -s models/multimodal/processing

 - label: Multi-Modal Models Test (Standard) # 60min
  timeout_in_minutes: 80
@@ -928,7 +857,6 @@ steps:
 - label: Transformers Nightly Models Test
  working_dir: "/vllm-workspace/"
  optional: true
-  soft_fail: true
  commands:
    - pip install --upgrade git+https://github.com/huggingface/transformers
    - pytest -v -s tests/models/test_initialization.py
@@ -944,7 +872,7 @@ steps:
  timeout_in_minutes: 30
  working_dir: "/vllm-workspace/"
  gpu: b200
-  optional: true # TODO(mgoin): disable optional once runner is fixed
+  # optional: true
  source_file_dependencies:
  - csrc/quantization/fp4/
  - csrc/attention/mla/
@@ -954,16 +882,11 @@ steps:
  - vllm/model_executor/layers/fused_moe/flashinfer_cutlass_prepare_finalize.py
  - vllm/model_executor/layers/quantization/utils/flashinfer_utils.py
  - vllm/v1/attention/backends/flashinfer.py
-  - vllm/v1/attention/backends/mla/cutlass_mla.py
-  - vllm/v1/attention/backends/mla/flashinfer_mla.py
-  - vllm/platforms/cuda.py
-  - vllm/attention/selector.py
  commands:
    - nvidia-smi
    - python3 examples/offline_inference/basic/chat.py
    # Attention
    # num_heads2 broken by https://github.com/flashinfer-ai/flashinfer/issues/1353
-    - pytest -v -s tests/kernels/attention/test_attention_selector.py
    - pytest -v -s tests/kernels/attention/test_flashinfer.py -k 'not num_heads2'
    - pytest -v -s tests/kernels/attention/test_flashinfer_trtllm_attention.py
    - pytest -v -s tests/kernels/attention/test_cutlass_mla_decode.py
@@ -980,61 +903,27 @@ steps:
    - pytest -v -s tests/kernels/moe/test_nvfp4_moe.py
    - pytest -v -s tests/kernels/moe/test_ocp_mx_moe.py
    - pytest -v -s tests/kernels/moe/test_flashinfer.py
-    - pytest -v -s tests/kernels/moe/test_cutedsl_moe.py

- label: Blackwell Fusion and Compile Tests # 30 min
+- label: Blackwell Fusion Tests # 30 min
  timeout_in_minutes: 40
  working_dir: "/vllm-workspace/"
  gpu: b200
-  optional: true # TODO(mgoin): disable optional once runner is fixed
  source_file_dependencies:
  - csrc/quantization/fp4/
  - vllm/model_executor/layers/quantization/utils/flashinfer_utils.py
  - vllm/v1/attention/backends/flashinfer.py
-  - vllm/v1/worker/
-  - vllm/v1/cudagraph_dispatcher.py
  - vllm/compilation/
  # can affect pattern matching
  - vllm/model_executor/layers/layernorm.py
  - vllm/model_executor/layers/activation.py
  - vllm/model_executor/layers/quantization/input_quant_fp8.py
-  - tests/compile/test_fusion_attn.py
-  - tests/compile/test_silu_mul_quant_fusion.py
-  - tests/compile/distributed/test_fusion_all_reduce.py
-  - tests/compile/distributed/test_fusions_e2e.py
-  - tests/compile/fullgraph/test_full_graph.py
  commands:
    - nvidia-smi
    - pytest -v -s tests/compile/test_fusion_attn.py
    - pytest -v -s tests/compile/test_silu_mul_quant_fusion.py
    # this runner has 2 GPUs available even though num_gpus=2 is not set
-    - pytest -v -s tests/compile/distributed/test_fusion_all_reduce.py
-    # Limit to Inductor partition, no custom ops, and allreduce & attn fusion to reduce running time
-    # Wrap with quotes to escape yaml
-    - "pytest -v -s tests/compile/distributed/test_fusions_e2e.py::test_tp2_attn_quant_allreduce_rmsnorm -k 'True and not +quant_fp8 and not +rms_norm'"
-    # test_fp8_kv_scale_compile requires FlashAttention (not supported on default L4/L40)
-    - pytest -v -s tests/compile/fullgraph/test_full_graph.py::test_fp8_kv_scale_compile
-
- label: Blackwell Fusion E2E Tests # 30 min
-  timeout_in_minutes: 40
-  working_dir: "/vllm-workspace/"
-  gpu: b200
-  optional: true
-  num_gpus: 2
-  source_file_dependencies:
-  - csrc/quantization/fp4/
-  - vllm/model_executor/layers/quantization/utils/flashinfer_utils.py
-  - vllm/v1/attention/backends/flashinfer.py
-  - vllm/compilation/
-  # can affect pattern matching
-  - vllm/model_executor/layers/layernorm.py
-  - vllm/model_executor/layers/activation.py
-  - vllm/model_executor/layers/quantization/input_quant_fp8.py
-  - tests/compile/distributed/test_fusions_e2e.py
-  commands:
-    - nvidia-smi
-    # Run all e2e fusion tests
-    - pytest -v -s tests/compile/distributed/test_fusions_e2e.py
+    - pytest -v -s tests/compile/test_fusion_all_reduce.py
+    - pytest -v -s tests/compile/test_fusions_e2e.py

 - label: Blackwell GPT-OSS Eval
  timeout_in_minutes: 60
@@ -1054,7 +943,6 @@ steps:
  timeout_in_minutes: 60
  working_dir: "/vllm-workspace/"
  gpu: b200
-  optional: true # TODO(mgoin): disable optional once runner is fixed
  source_file_dependencies:
  - tests/quantization/test_blackwell_moe.py
  - vllm/model_executor/models/deepseek_v2.py
@@ -1076,7 +964,7 @@ steps:
  - csrc/
  - vllm/model_executor/layers/quantization
  commands:
-  - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-blackwell.txt
+  - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-blackwell.txt --tp-size=1

 #####  1 GPU test  #####
 #####  multi gpus test  #####
@@ -1112,13 +1000,13 @@ steps:
  - # the following commands are for the first node, with ip 192.168.10.10 (ray environment already set up)
    - VLLM_TEST_SAME_HOST=0 torchrun --nnodes 2 --nproc-per-node=2 --rdzv_backend=c10d --rdzv_endpoint=192.168.10.10 distributed/test_same_node.py | grep 'Same node test passed'
    - NUM_NODES=2 torchrun --nnodes 2 --nproc-per-node=2 --rdzv_backend=c10d --rdzv_endpoint=192.168.10.10 distributed/test_node_count.py | grep 'Node count test passed'
-    - python3 ../examples/offline_inference/data_parallel.py -dp=2 -tp=1 --dp-num-nodes=2 --dp-node-rank=0 --dp-master-addr=192.168.10.10 --dp-master-port=12345 --enforce-eager --trust-remote-code
+    - python3 ../examples/offline_inference/data_parallel.py --dp-size=2 --tp-size=1 --node-size=2 --node-rank=0 --master-addr=192.168.10.10 --master-port=12345 --enforce-eager --trust-remote-code
    - VLLM_MULTI_NODE=1 pytest -v -s distributed/test_multi_node_assignment.py
    - VLLM_MULTI_NODE=1 pytest -v -s distributed/test_pipeline_parallel.py
  - # the following commands are for the second node, with ip 192.168.10.11 (ray environment already set up)
    - VLLM_TEST_SAME_HOST=0 torchrun --nnodes 2 --nproc-per-node=2 --rdzv_backend=c10d --rdzv_endpoint=192.168.10.10 distributed/test_same_node.py | grep 'Same node test passed'
    - NUM_NODES=2 torchrun --nnodes 2 --nproc-per-node=2 --rdzv_backend=c10d --rdzv_endpoint=192.168.10.10 distributed/test_node_count.py | grep 'Node count test passed'
-    - python3 ../examples/offline_inference/data_parallel.py -dp=2 -tp=1 --dp-num-nodes=2 --dp-node-rank=1 --dp-master-addr=192.168.10.10 --dp-master-port=12345 --enforce-eager --trust-remote-code
+    - python3 ../examples/offline_inference/data_parallel.py --dp-size=2 --tp-size=1 --node-size=2 --node-rank=1 --master-addr=192.168.10.10 --master-port=12345 --enforce-eager --trust-remote-code

 - label: Distributed Tests (2 GPUs) # 68min
  timeout_in_minutes: 90
@@ -1133,7 +1021,7 @@ steps:
  - vllm/worker/worker_base.py
  - vllm/v1/engine/
  - vllm/v1/worker/
-  - tests/compile/fullgraph/test_basic_correctness.py
+  - tests/compile/test_basic_correctness.py
  - tests/compile/test_wrapper.py
  - tests/distributed/
  - tests/entrypoints/llm/test_collective_rpc.py
@@ -1145,11 +1033,10 @@ steps:
  # https://github.com/NVIDIA/nccl/issues/1838
  - export NCCL_CUMEM_HOST_ENABLE=0
  - TP_SIZE=1 DP_SIZE=2 pytest -v -s v1/distributed/test_async_llm_dp.py
-  - TP_SIZE=1 DP_SIZE=2 pytest -v -s v1/distributed/test_eagle_dp.py
  - TP_SIZE=1 DP_SIZE=2 pytest -v -s v1/distributed/test_external_lb_dp.py
  - DP_SIZE=2 pytest -v -s v1/entrypoints/openai/test_multi_api_servers.py
  - pytest -v -s entrypoints/llm/test_collective_rpc.py
-  - pytest -v -s ./compile/fullgraph/test_basic_correctness.py
+  - pytest -v -s ./compile/test_basic_correctness.py
  - pytest -v -s ./compile/test_wrapper.py
  - VLLM_TEST_SAME_HOST=1 torchrun --nproc-per-node=4 distributed/test_same_node.py | grep 'Same node test passed'
  - VLLM_TEST_SAME_HOST=1 VLLM_TEST_WITH_DEFAULT_DEVICE_SET=1 torchrun --nproc-per-node=4 distributed/test_same_node.py | grep 'Same node test passed'
@@ -1235,8 +1122,6 @@ steps:
    # FIXIT: find out which code initialize cuda before running the test
    # before the fix, we need to use spawn to test it
    - export VLLM_WORKER_MULTIPROC_METHOD=spawn
-    # Alot of these tests are on the edge of OOMing
-    - export PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True
    # There is some Tensor Parallelism related processing logic in LoRA that
    # requires multi-GPU testing for validation.
    - pytest -v -s -x lora/test_chatglm3_tp.py
@@ -1279,18 +1164,7 @@ steps:
    - tests/v1/kv_connector/nixl_integration/
  commands:
    - uv pip install --system -r /vllm-workspace/requirements/kv_connectors.txt
-    - bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh
-
- label: DP EP NixlConnector PD accuracy tests (Distributed)
-  timeout_in_minutes: 30
-  working_dir: "/vllm-workspace/tests"
-  num_gpus: 4
-  source_file_dependencies:
-    - vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py
-    - tests/v1/kv_connector/nixl_integration/
-  commands:
-    - uv pip install --system -r /vllm-workspace/requirements/kv_connectors.txt
-    - DP_EP=1 bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh
+    - bash v1/kv_connector/nixl_integration/tp_config_sweep_accuracy_test.sh


 ##### multi gpus test #####
@@ -1342,14 +1216,13 @@ steps:
  working_dir: "/vllm-workspace/"
  num_gpus: 2
  commands:
-    - VLLM_TEST_CLEAN_GPU_MEMORY=1 pytest -v -s tests/compile/distributed/test_async_tp.py
-    - pytest -v -s tests/compile/distributed/test_sequence_parallelism.py
-    - pytest -v -s tests/compile/distributed/test_fusion_all_reduce.py
-    - "VLLM_TEST_CLEAN_GPU_MEMORY=1 pytest -v -s tests/compile/distributed/test_fusions_e2e.py -k 'not Llama-4'"
-    - VLLM_TEST_CLEAN_GPU_MEMORY=1 pytest -v -s tests/distributed/test_sequence_parallel.py
+    - pytest -v -s tests/compile/test_async_tp.py
+    - pytest -v -s tests/compile/test_sequence_parallelism.py
+    - pytest -v -s tests/compile/test_fusion_all_reduce.py
+    - pytest -v -s tests/compile/test_fusions_e2e.py::test_tp2_attn_quant_allreduce_rmsnorm
    - pytest -v -s tests/distributed/test_context_parallel.py
-    - CUDA_VISIBLE_DEVICES=1,2 VLLM_USE_DEEP_GEMM=1 VLLM_LOGGING_LEVEL=DEBUG python3 examples/offline_inference/data_parallel.py --model=Qwen/Qwen1.5-MoE-A2.7B -tp=1 -dp=2 --max-model-len=2048 --all2all-backend=deepep_high_throughput
-    - pytest -v -s tests/v1/distributed/test_dbo.py
+    - CUDA_VISIBLE_DEVICES=1,2 VLLM_ALL2ALL_BACKEND=deepep_high_throughput VLLM_USE_DEEP_GEMM=1 VLLM_LOGGING_LEVEL=DEBUG python3 examples/offline_inference/data_parallel.py --model Qwen/Qwen1.5-MoE-A2.7B --tp-size=1  --dp-size=2 --max-model-len 2048
+    - pytest -v -s tests/v1/distributed/test_dbo.py  

 ##### B200 test #####
 - label: Distributed Tests (B200) # optional
@@ -1366,14 +1239,12 @@ steps:
 - label: Prime-RL Integration Test # 15min
  timeout_in_minutes: 30
  optional: true
-  soft_fail: true
  num_gpus: 2
  working_dir: "/vllm-workspace"
  source_file_dependencies:
  - vllm/
  - .buildkite/scripts/run-prime-rl-test.sh
  commands:
-    - nvidia-smi
    - bash .buildkite/scripts/run-prime-rl-test.sh

 - label: DeepSeek V2-Lite Accuracy
@@ -1385,20 +1256,11 @@ steps:
  commands:
  - bash .buildkite/scripts/scheduled_integration_test/deepseek_v2_lite_ep_eplb.sh 0.25 200 8010

- label: Qwen3-30B-A3B-FP8-block Accuracy (H100)
+- label: Qwen3-30B-A3B-FP8-block Accuracy
  timeout_in_minutes: 60
  gpu: h100
  optional: true
  num_gpus: 4
  working_dir: "/vllm-workspace"
  commands:
-  - bash .buildkite/scripts/scheduled_integration_test/qwen30b_a3b_fp8_block_ep_eplb.sh 0.8 200 8020
-
- label: Qwen3-30B-A3B-FP8-block Accuracy (B200)
-  timeout_in_minutes: 60
-  gpu: b200
-  optional: true
-  num_gpus: 2
-  working_dir: "/vllm-workspace"
-  commands:
-  - bash .buildkite/scripts/scheduled_integration_test/qwen30b_a3b_fp8_block_ep_eplb.sh 0.8 200 8020 2 1
+  - bash .buildkite/scripts/scheduled_integration_test/qwen30b_a3b_fp8_block_ep.sh 0.8 200 8020
@@ -1,21 +0,0 @@
-group: Attention
-depends_on: 
-  - image-build
-steps:
- label: V1 attention (H100)
-  timeout_in_minutes: 30
-  gpu: h100
-  source_file_dependencies:
-    - vllm/v1/attention
-    - tests/v1/attention
-  commands:
-    - pytest -v -s v1/attention
-
- label: V1 attention (B200)
-  timeout_in_minutes: 30
-  gpu: b200
-  source_file_dependencies:
-    - vllm/v1/attention
-    - tests/v1/attention
-  commands:
-    - VLLM_DISABLE_FLASHINFER_PREFILL=1 pytest -v -s v1/attention # TODO: FI prefill is bugged and causes incorrectness, fix this
@@ -1,16 +0,0 @@
-group: Basic Correctness
-depends_on: 
-  - image-build
-steps:
- label: Basic Correctness
-  timeout_in_minutes: 30
-  source_file_dependencies:
-  - vllm/
-  - tests/basic_correctness/test_basic_correctness
-  - tests/basic_correctness/test_cpu_offload
-  - tests/basic_correctness/test_cumem.py
-  commands:
-  - export VLLM_WORKER_MULTIPROC_METHOD=spawn
-  - pytest -v -s basic_correctness/test_cumem.py
-  - pytest -v -s basic_correctness/test_basic_correctness.py
-  - pytest -v -s basic_correctness/test_cpu_offload.py
@@ -1,19 +0,0 @@
-group: Benchmarks
-depends_on: 
-  - image-build
-steps:
- label: Benchmarks
-  timeout_in_minutes: 20
-  working_dir: "/vllm-workspace/.buildkite"
-  source_file_dependencies:
-  - benchmarks/
-  commands:
-  - bash scripts/run-benchmarks.sh
-
- label: Benchmarks CLI Test
-  timeout_in_minutes: 20
-  source_file_dependencies:
-  - vllm/
-  - tests/benchmarks/
-  commands:
-  - pytest -v -s benchmarks/
@@ -1,57 +0,0 @@
-group: Compile
-depends_on: 
-  - image-build
-steps:
- label: Fusion and Compile Tests (B200)
-  timeout_in_minutes: 40
-  working_dir: "/vllm-workspace/"
-  gpu: b200
-  source_file_dependencies:
-  - csrc/quantization/fp4/
-  - vllm/model_executor/layers/quantization/utils/flashinfer_utils.py
-  - vllm/v1/attention/backends/flashinfer.py
-  - vllm/v1/worker/
-  - vllm/v1/cudagraph_dispatcher.py
-  - vllm/compilation/
-  # can affect pattern matching
-  - vllm/model_executor/layers/layernorm.py
-  - vllm/model_executor/layers/activation.py
-  - vllm/model_executor/layers/quantization/input_quant_fp8.py
-  - tests/compile/test_fusion_attn.py
-  - tests/compile/test_silu_mul_quant_fusion.py
-  - tests/compile/distributed/test_fusion_all_reduce.py
-  - tests/compile/distributed/test_fusions_e2e.py
-  - tests/compile/fullgraph/test_full_graph.py
-  commands:
-    - nvidia-smi
-    - pytest -v -s tests/compile/test_fusion_attn.py
-    - pytest -v -s tests/compile/test_silu_mul_quant_fusion.py
-    # this runner has 2 GPUs available even though num_gpus=2 is not set
-    - pytest -v -s tests/compile/distributed/test_fusion_all_reduce.py
-    # Limit to Inductor partition, no custom ops, and allreduce & attn fusion to reduce running time
-    # Wrap with quotes to escape yaml
-    - "pytest -v -s tests/compile/distributed/test_fusions_e2e.py::test_tp2_attn_quant_allreduce_rmsnorm -k 'True and not +quant_fp8 and not +rms_norm'"
-    # test_fp8_kv_scale_compile requires FlashAttention (not supported on default L4/L40)
-    - pytest -v -s tests/compile/fullgraph/test_full_graph.py::test_fp8_kv_scale_compile
-
- label: Fusion E2E (2 GPUs)(B200)
-  timeout_in_minutes: 40
-  working_dir: "/vllm-workspace/"
-  gpu: b200
-  optional: true
-  num_gpus: 2
-  source_file_dependencies:
-  - csrc/quantization/fp4/
-  - vllm/model_executor/layers/quantization/utils/flashinfer_utils.py
-  - vllm/v1/attention/backends/flashinfer.py
-  - vllm/compilation/
-  # can affect pattern matching
-  - vllm/model_executor/layers/layernorm.py
-  - vllm/model_executor/layers/activation.py
-  - vllm/model_executor/layers/quantization/input_quant_fp8.py
-  - tests/compile/distributed/test_fusions_e2e.py
-  commands:
-    - nvidia-smi
-    # Run all e2e fusion tests
-    - pytest -v -s tests/compile/distributed/test_fusions_e2e.py
-
@@ -1,22 +0,0 @@
-group: CUDA
-depends_on: 
-  - image-build
-steps:
- label: Platform Tests (CUDA)
-  timeout_in_minutes: 15
-  source_file_dependencies:
-  - vllm/
-  - tests/cuda
-  commands:
-    - pytest -v -s cuda/test_cuda_context.py
-
- label: Cudagraph
-  timeout_in_minutes: 20
-  source_file_dependencies:
-  - tests/v1/cudagraph
-  - vllm/v1/cudagraph_dispatcher.py
-  - vllm/config/compilation.py
-  - vllm/compilation
-  commands:
-    - pytest -v -s v1/cudagraph/test_cudagraph_dispatch.py
-    - pytest -v -s v1/cudagraph/test_cudagraph_mode.py
@@ -1,199 +0,0 @@
-group: Distributed
-depends_on: 
-  - image-build
-steps:
- label: Distributed Comm Ops
-  timeout_in_minutes: 20
-  working_dir: "/vllm-workspace/tests"
-  num_gpus: 2
-  source_file_dependencies:
-  - vllm/distributed
-  - tests/distributed
-  commands:
-  - pytest -v -s distributed/test_comm_ops.py
-  - pytest -v -s distributed/test_shm_broadcast.py
-  - pytest -v -s distributed/test_shm_buffer.py
-  - pytest -v -s distributed/test_shm_storage.py
-
- label: Distributed (2 GPUs)
-  timeout_in_minutes: 90
-  working_dir: "/vllm-workspace/tests"
-  num_gpus: 2
-  source_file_dependencies:
-  - vllm/compilation/
-  - vllm/distributed/
-  - vllm/engine/
-  - vllm/executor/
-  - vllm/worker/worker_base.py
-  - vllm/v1/engine/
-  - vllm/v1/worker/
-  - tests/compile/fullgraph/test_basic_correctness.py
-  - tests/compile/test_wrapper.py
-  - tests/distributed/
-  - tests/entrypoints/llm/test_collective_rpc.py
-  - tests/v1/distributed
-  - tests/v1/entrypoints/openai/test_multi_api_servers.py
-  - tests/v1/shutdown
-  - tests/v1/worker/test_worker_memory_snapshot.py
-  commands:
-  # https://github.com/NVIDIA/nccl/issues/1838
-  - export NCCL_CUMEM_HOST_ENABLE=0
-  - TP_SIZE=1 DP_SIZE=2 pytest -v -s v1/distributed/test_async_llm_dp.py
-  - TP_SIZE=1 DP_SIZE=2 pytest -v -s v1/distributed/test_eagle_dp.py
-  - TP_SIZE=1 DP_SIZE=2 pytest -v -s v1/distributed/test_external_lb_dp.py
-  - DP_SIZE=2 pytest -v -s v1/entrypoints/openai/test_multi_api_servers.py
-  - pytest -v -s entrypoints/llm/test_collective_rpc.py
-  - pytest -v -s ./compile/fullgraph/test_basic_correctness.py
-  - pytest -v -s ./compile/test_wrapper.py
-  - VLLM_TEST_SAME_HOST=1 torchrun --nproc-per-node=4 distributed/test_same_node.py | grep 'Same node test passed'
-  - VLLM_TEST_SAME_HOST=1 VLLM_TEST_WITH_DEFAULT_DEVICE_SET=1 torchrun --nproc-per-node=4 distributed/test_same_node.py | grep 'Same node test passed'
-  - pytest -v -s distributed/test_sequence_parallel.py
-  - CUDA_VISIBLE_DEVICES=0,1 pytest -v -s v1/shutdown
-  - pytest -v -s v1/worker/test_worker_memory_snapshot.py
-
- label: Distributed Tests (4 GPUs)
-  timeout_in_minutes: 50
-  working_dir: "/vllm-workspace/tests"
-  num_gpus: 4
-  source_file_dependencies:
-  - vllm/distributed/
-  - tests/distributed/test_utils
-  - tests/distributed/test_pynccl
-  - tests/distributed/test_events
-  - tests/compile/fullgraph/test_basic_correctness.py
-  - examples/offline_inference/rlhf.py
-  - examples/offline_inference/rlhf_colocate.py
-  - tests/examples/offline_inference/data_parallel.py
-  - tests/v1/distributed
-  - tests/v1/engine/test_engine_core_client.py
-  - tests/distributed/test_symm_mem_allreduce.py
-  commands:
-  # https://github.com/NVIDIA/nccl/issues/1838
-  - export NCCL_CUMEM_HOST_ENABLE=0
-  # test with torchrun tp=2 and external_dp=2
-  - torchrun --nproc-per-node=4 distributed/test_torchrun_example.py
-  # test with torchrun tp=2 and pp=2
-  - PP_SIZE=2 torchrun --nproc-per-node=4 distributed/test_torchrun_example.py
-  # test with torchrun tp=4 and dp=1
-  - TP_SIZE=4 torchrun --nproc-per-node=4 distributed/test_torchrun_example_moe.py
-  # test with torchrun tp=2, pp=2 and dp=1
-  - PP_SIZE=2 TP_SIZE=2 torchrun --nproc-per-node=4 distributed/test_torchrun_example_moe.py
-  # test with torchrun tp=1 and dp=4 with ep
-  - DP_SIZE=4 ENABLE_EP=1 torchrun --nproc-per-node=4 distributed/test_torchrun_example_moe.py
-  # test with torchrun tp=2 and dp=2 with ep
-  - TP_SIZE=2 DP_SIZE=2 ENABLE_EP=1 torchrun --nproc-per-node=4 distributed/test_torchrun_example_moe.py
-  # test with internal dp
-  - python3 ../examples/offline_inference/data_parallel.py --enforce-eager
-  - TP_SIZE=2 DP_SIZE=2 pytest -v -s v1/distributed/test_async_llm_dp.py
-  - TP_SIZE=2 DP_SIZE=2 pytest -v -s v1/distributed/test_eagle_dp.py
-  - TP_SIZE=2 DP_SIZE=2 pytest -v -s v1/distributed/test_external_lb_dp.py
-  - TP_SIZE=1 DP_SIZE=4 pytest -v -s v1/distributed/test_internal_lb_dp.py
-  - TP_SIZE=1 DP_SIZE=4 pytest -v -s v1/distributed/test_hybrid_lb_dp.py
-  - pytest -v -s v1/engine/test_engine_core_client.py::test_kv_cache_events_dp
-  - pytest -v -s distributed/test_utils.py
-  - pytest -v -s compile/fullgraph/test_basic_correctness.py
-  - pytest -v -s distributed/test_pynccl.py
-  - pytest -v -s distributed/test_events.py
-  - pytest -v -s distributed/test_symm_mem_allreduce.py
-  # TODO: create a dedicated test section for multi-GPU example tests
-  # when we have multiple distributed example tests
-  - cd ../examples/offline_inference
-  - VLLM_ALLOW_INSECURE_SERIALIZATION=1 python3 rlhf.py
-  - VLLM_ALLOW_INSECURE_SERIALIZATION=1 RAY_DEDUP_LOGS=0 python3 rlhf_colocate.py
-
- label: Distributed Tests (8 GPUs)(H100)
-  timeout_in_minutes: 10
-  gpu: h100
-  num_gpus: 8
-  working_dir: "/vllm-workspace/tests"
-  source_file_dependencies:
-  - examples/offline_inference/torchrun_dp_example.py
-  - vllm/config/parallel.py
-  - vllm/distributed/
-  - vllm/v1/engine/llm_engine.py
-  - vllm/v1/executor/uniproc_executor.py
-  - vllm/v1/worker/gpu_worker.py
-  commands:
-  # https://github.com/NVIDIA/nccl/issues/1838
-  - export NCCL_CUMEM_HOST_ENABLE=0
-  # test with torchrun tp=2 and dp=4 with ep
-  - torchrun --nproc-per-node=8 ../examples/offline_inference/torchrun_dp_example.py --tp-size=2 --pp-size=1 --dp-size=4 --enable-ep
-
- label: Distributed Tests (4 GPUs)(A100)
-  gpu: a100
-  optional: true
-  num_gpus: 4
-  source_file_dependencies:
-  - vllm/
-  commands:
-  # NOTE: don't test llama model here, it seems hf implementation is buggy
-  # see https://github.com/vllm-project/vllm/pull/5689 for details
-  - pytest -v -s distributed/test_custom_all_reduce.py
-  - torchrun --nproc_per_node=2 distributed/test_ca_buffer_sharing.py
-  - TARGET_TEST_SUITE=A100 pytest basic_correctness/ -v -s -m 'distributed(num_gpus=2)'
-  - pytest -v -s -x lora/test_mixtral.py
-
- label: Distributed Tests (2 GPUs)(H200)
-  gpu: h200
-  optional: true
-  working_dir: "/vllm-workspace/"
-  num_gpus: 2
-  commands:
-    - VLLM_TEST_CLEAN_GPU_MEMORY=1 pytest -v -s tests/compile/distributed/test_async_tp.py
-    - pytest -v -s tests/compile/distributed/test_sequence_parallelism.py
-    - pytest -v -s tests/compile/distributed/test_fusion_all_reduce.py
-    - VLLM_TEST_CLEAN_GPU_MEMORY=1 pytest -v -s tests/compile/distributed/test_fusions_e2e.py -k 'not Llama-4'
-    - VLLM_TEST_CLEAN_GPU_MEMORY=1 pytest -v -s tests/distributed/test_sequence_parallel.py
-    - pytest -v -s tests/distributed/test_context_parallel.py
-    - CUDA_VISIBLE_DEVICES=1,2 VLLM_USE_DEEP_GEMM=1 VLLM_LOGGING_LEVEL=DEBUG python3 examples/offline_inference/data_parallel.py --model=Qwen/Qwen1.5-MoE-A2.7B -tp=1 -dp=2 --max-model-len=2048 --all2all-backend=deepep_high_throughput
-    - pytest -v -s tests/v1/distributed/test_dbo.py
-
- label: Distributed Tests (2 GPUs)(B200)
-  gpu: b200
-  optional: true
-  working_dir: "/vllm-workspace/"
-  num_gpus: 2
-  commands:
-    - pytest -v -s tests/distributed/test_context_parallel.py
-    - pytest -v -s tests/distributed/test_nccl_symm_mem_allreduce.py
-    - pytest -v -s tests/v1/distributed/test_dbo.py
-
- label: 2 Node Test (4 GPUs)
-  timeout_in_minutes: 30
-  working_dir: "/vllm-workspace/tests"
-  num_gpus: 2
-  num_nodes: 2
-  source_file_dependencies:
-  - vllm/distributed/
-  - vllm/engine/
-  - vllm/executor/
-  - vllm/model_executor/models/
-  - tests/distributed/
-  - tests/examples/offline_inference/data_parallel.py
-  commands:
-    - ./.buildkite/scripts/run-multi-node-test.sh /vllm-workspace/tests 2 2 public.ecr.aws/q9t5s3a7/vllm-ci-postmerge-repo:0bec63fa317e1fbd62e19b0fc31c43c81bf89077 "VLLM_TEST_SAME_HOST=0 torchrun --nnodes 2 --nproc-per-node=2 --rdzv_backend=c10d --rdzv_endpoint=192.168.10.10 distributed/test_same_node.py | grep 'Same node test passed' && NUM_NODES=2 torchrun --nnodes 2 --nproc-per-node=2 --rdzv_backend=c10d --rdzv_endpoint=192.168.10.10 distributed/test_node_count.py | grep 'Node count test passed' && python3 ../examples/offline_inference/data_parallel.py -dp=2 -tp=1 --dp-num-nodes=2 --dp-node-rank=0 --dp-master-addr=192.168.10.10 --dp-master-port=12345 --enforce-eager --trust-remote-code && VLLM_MULTI_NODE=1 pytest -v -s distributed/test_multi_node_assignment.py && VLLM_MULTI_NODE=1 pytest -v -s distributed/test_pipeline_parallel.py" "VLLM_TEST_SAME_HOST=0 torchrun --nnodes 2 --nproc-per-node=2 --rdzv_backend=c10d --rdzv_endpoint=192.168.10.10 distributed/test_same_node.py | grep 'Same node test passed' && NUM_NODES=2 torchrun --nnodes 2 --nproc-per-node=2 --rdzv_backend=c10d --rdzv_endpoint=192.168.10.10 distributed/test_node_count.py | grep 'Node count test passed' && python3 ../examples/offline_inference/data_parallel.py -dp=2 -tp=1 --dp-num-nodes=2 --dp-node-rank=1 --dp-master-addr=192.168.10.10 --dp-master-port=12345 --enforce-eager --trust-remote-code"
-
- label: Distributed NixlConnector PD accuracy (4 GPUs)
-  timeout_in_minutes: 30
-  working_dir: "/vllm-workspace/tests"
-  num_gpus: 4
-  source_file_dependencies:
-    - vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py
-    - tests/v1/kv_connector/nixl_integration/
-  commands:
-    - uv pip install --system -r /vllm-workspace/requirements/kv_connectors.txt
-    - bash v1/kv_connector/nixl_integration/tp_config_sweep_accuracy_test.sh
-
- label: Pipeline + Context Parallelism (4 GPUs))
-  timeout_in_minutes: 60
-  working_dir: "/vllm-workspace/tests"
-  num_gpus: 4
-  source_file_dependencies:
-  - vllm/distributed/
-  - vllm/engine/
-  - vllm/executor/
-  - vllm/model_executor/models/
-  - tests/distributed/
-  commands:
-  - pytest -v -s distributed/test_pp_cudagraph.py
-  - pytest -v -s distributed/test_pipeline_parallel.py
@@ -1,42 +0,0 @@
-group: E2E Integration
-depends_on: 
-  - image-build
-steps:
- label: DeepSeek V2-Lite Accuracy
-  timeout_in_minutes: 60
-  gpu: h100
-  optional: true
-  num_gpus: 4
-  working_dir: "/vllm-workspace"
-  commands:
-  - bash .buildkite/scripts/scheduled_integration_test/deepseek_v2_lite_ep_eplb.sh 0.25 200 8010
-
- label: Qwen3-30B-A3B-FP8-block Accuracy
-  timeout_in_minutes: 60
-  gpu: h100
-  optional: true
-  num_gpus: 4
-  working_dir: "/vllm-workspace"
-  commands:
-  - bash .buildkite/scripts/scheduled_integration_test/qwen30b_a3b_fp8_block_ep_eplb.sh 0.8 200 8020
-
- label: Qwen3-30B-A3B-FP8-block Accuracy (B200)
-  timeout_in_minutes: 60
-  gpu: b200
-  optional: true
-  num_gpus: 2
-  working_dir: "/vllm-workspace"
-  commands:
-  - bash .buildkite/scripts/scheduled_integration_test/qwen30b_a3b_fp8_block_ep_eplb.sh 0.8 200 8020 2 1
-
- label: Prime-RL Integration (2 GPUs)
-  timeout_in_minutes: 30
-  optional: true
-  soft_fail: true
-  num_gpus: 2
-  working_dir: "/vllm-workspace"
-  source_file_dependencies:
-  - vllm/
-  - .buildkite/scripts/run-prime-rl-test.sh
-  commands:
-    - bash .buildkite/scripts/run-prime-rl-test.sh
@@ -1,26 +0,0 @@
-group: Engine
-depends_on: 
-  - image-build
-steps:
- label: Engine
-  timeout_in_minutes: 15
-  source_file_dependencies:
-  - vllm/
-  - tests/engine
-  - tests/test_sequence
-  - tests/test_config
-  - tests/test_logger
-  - tests/test_vllm_port
-  commands:
-  - pytest -v -s engine test_sequence.py test_config.py test_logger.py test_vllm_port.py
-
- label: V1 e2e + engine
-  timeout_in_minutes: 45
-  source_file_dependencies:
-    - vllm/
-    - tests/v1
-  commands:
-    # TODO: accuracy does not match, whether setting
-    # VLLM_USE_FLASHINFER_SAMPLER or not on H100.
-    - pytest -v -s v1/e2e
-    - pytest -v -s v1/engine
@@ -1,83 +0,0 @@
-group: Entrypoints
-depends_on: 
-  - image-build
-steps:
- label: Entrypoints Unit Tests  
-  timeout_in_minutes: 10
-  working_dir: "/vllm-workspace/tests"
-  source_file_dependencies:
-  - vllm/entrypoints
-  - tests/entrypoints/
-  commands:
-  - pytest -v -s entrypoints/openai/tool_parsers
-  - pytest -v -s entrypoints/ --ignore=entrypoints/llm --ignore=entrypoints/rpc --ignore=entrypoints/sleep --ignore=entrypoints/instrumentator --ignore=entrypoints/openai --ignore=entrypoints/offline_mode --ignore=entrypoints/test_chat_utils.py  --ignore=entrypoints/pooling
-
- label: Entrypoints Integration (LLM)
-  timeout_in_minutes: 40
-  working_dir: "/vllm-workspace/tests"
-  source_file_dependencies:
-  - vllm/
-  - tests/entrypoints/llm
-  - tests/entrypoints/offline_mode
-  commands:
-  - export VLLM_WORKER_MULTIPROC_METHOD=spawn
-  - pytest -v -s entrypoints/llm --ignore=entrypoints/llm/test_generate.py --ignore=entrypoints/llm/test_collective_rpc.py
-  - pytest -v -s entrypoints/llm/test_generate.py # it needs a clean process
-  - pytest -v -s entrypoints/offline_mode # Needs to avoid interference with other tests
-
- label: Entrypoints Integration (API Server 1)
-  timeout_in_minutes: 130
-  working_dir: "/vllm-workspace/tests"
-  source_file_dependencies:
-  - vllm/
-  - tests/entrypoints/openai
-  - tests/entrypoints/test_chat_utils
-  commands:
-  - export VLLM_WORKER_MULTIPROC_METHOD=spawn
-  - pytest -v -s entrypoints/openai --ignore=entrypoints/openai/test_chat_with_tool_reasoning.py --ignore=entrypoints/openai/test_oot_registration.py --ignore=entrypoints/openai/test_tensorizer_entrypoint.py --ignore=entrypoints/openai/correctness/  --ignore=entrypoints/openai/tool_parsers/
-  - pytest -v -s entrypoints/test_chat_utils.py
-
-
- label: Entrypoints Integration (API Server 2)
-  timeout_in_minutes: 130
-  working_dir: "/vllm-workspace/tests"
-  source_file_dependencies:
-  - vllm/
-  - tests/tool_use
-  - tests/entrypoints/sleep
-  - tests/entrypoints/instrumentator
-  - tests/entrypoints/rpc
-  commands:
-  - export VLLM_WORKER_MULTIPROC_METHOD=spawn
-  - PYTHONPATH=/vllm-workspace pytest -v -s entrypoints/rpc
-  - pytest -v -s entrypoints/instrumentator
-  - pytest -v -s entrypoints/sleep
-  - pytest -v -s tool_use
-
- label: Entrypoints Integration (Pooling)
-  timeout_in_minutes: 50
-  working_dir: "/vllm-workspace/tests"
-  source_file_dependencies:
-  - vllm/
-  - tests/entrypoints/pooling
-  commands:
-  - export VLLM_WORKER_MULTIPROC_METHOD=spawn
-  - pytest -v -s entrypoints/pooling
-
-
- label: Entrypoints V1
-  timeout_in_minutes: 50
-  source_file_dependencies:
-    - vllm/
-    - tests/v1
-  commands:
-    - pytest -v -s v1/entrypoints
-
- label: OpenAI API Correctness
-  timeout_in_minutes: 30
-  source_file_dependencies:
-  - csrc/
-  - vllm/entrypoints/openai/
-  - vllm/model_executor/models/whisper.py
-  commands: # LMEval+Transcription WER check
-  - pytest -s entrypoints/openai/correctness/
@@ -1,23 +0,0 @@
-group: Expert Parallelism
-depends_on: 
-  - image-build
-steps:
- label: EPLB Algorithm
-  timeout_in_minutes: 15
-  working_dir: "/vllm-workspace/tests"
-  source_file_dependencies:
-  - vllm/distributed/eplb
-  - tests/distributed/test_eplb_algo.py
-  commands:
-  - pytest -v -s distributed/test_eplb_algo.py
-
- label: EPLB Execution
-  timeout_in_minutes: 20
-  working_dir: "/vllm-workspace/tests"
-  num_gpus: 4
-  source_file_dependencies:
-  - vllm/distributed/eplb
-  - tests/distributed/test_eplb_execute.py
-  commands:
-  - pytest -v -s distributed/test_eplb_execute.py
-  - pytest -v -s distributed/test_eplb_spec_decode.py
@@ -1,117 +0,0 @@
-group: Kernels
-depends_on: 
-  - image-build
-steps:
- label: Kernels Core Operation Test
-  timeout_in_minutes: 75
-  source_file_dependencies:
-  - csrc/
-  - tests/kernels/core
-  - tests/kernels/test_top_k_per_row.py
-  commands:
-    - pytest -v -s kernels/core kernels/test_top_k_per_row.py
-
- label: Kernels Attention Test %N
-  timeout_in_minutes: 35
-  source_file_dependencies:
-  - csrc/attention/
-  - vllm/attention
-  - vllm/v1/attention
-  - tests/kernels/attention
-  commands:
-    - pytest -v -s kernels/attention --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
-  parallelism: 2
-
- label: Kernels Quantization Test %N
-  timeout_in_minutes: 90
-  source_file_dependencies:
-  - csrc/quantization/
-  - vllm/model_executor/layers/quantization
-  - tests/kernels/quantization
-  commands:
-    - pytest -v -s kernels/quantization --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
-  parallelism: 2
-
- label: Kernels MoE Test %N
-  timeout_in_minutes: 60
-  source_file_dependencies:
-  - csrc/quantization/cutlass_w8a8/moe/
-  - csrc/moe/
-  - tests/kernels/moe
-  - vllm/model_executor/layers/fused_moe/
-  - vllm/distributed/device_communicators/
-  - vllm/envs.py
-  - vllm/config
-  commands:
-    - pytest -v -s kernels/moe --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
-  parallelism: 2
-
- label: Kernels Mamba Test
-  timeout_in_minutes: 45
-  source_file_dependencies:
-  - csrc/mamba/
-  - tests/kernels/mamba
-  - vllm/model_executor/layers/mamba/ops
-  commands:
-    - pytest -v -s kernels/mamba
-
- label: Kernels DeepGEMM Test (H100)
-  timeout_in_minutes: 45
-  gpu: h100
-  num_gpus: 1
-  source_file_dependencies:
-  - tools/install_deepgemm.sh
-  - vllm/utils/deep_gemm.py
-  - vllm/model_executor/layers/fused_moe
-  - vllm/model_executor/layers/quantization
-  - tests/kernels/quantization/test_block_fp8.py
-  - tests/kernels/moe/test_deepgemm.py
-  - tests/kernels/moe/test_batched_deepgemm.py
-  - tests/kernels/attention/test_deepgemm_attention.py
-  commands:
-    - pytest -v -s kernels/quantization/test_block_fp8.py -k deep_gemm
-    - pytest -v -s kernels/moe/test_deepgemm.py
-    - pytest -v -s kernels/moe/test_batched_deepgemm.py
-    - pytest -v -s kernels/attention/test_deepgemm_attention.py
-
- label: Kernels (B200)
-  timeout_in_minutes: 30
-  working_dir: "/vllm-workspace/"
-  gpu: b200
-  # optional: true
-  source_file_dependencies:
-  - csrc/quantization/fp4/
-  - csrc/attention/mla/
-  - csrc/quantization/cutlass_w8a8/moe/
-  - vllm/model_executor/layers/fused_moe/cutlass_moe.py
-  - vllm/model_executor/layers/fused_moe/flashinfer_cutlass_moe.py
-  - vllm/model_executor/layers/fused_moe/flashinfer_cutlass_prepare_finalize.py
-  - vllm/model_executor/layers/quantization/utils/flashinfer_utils.py
-  - vllm/v1/attention/backends/flashinfer.py
-  - vllm/v1/attention/backends/mla/cutlass_mla.py
-  - vllm/v1/attention/backends/mla/flashinfer_mla.py
-  - vllm/platforms/cuda.py
-  - vllm/attention/selector.py
-  commands:
-    - nvidia-smi
-    - python3 examples/offline_inference/basic/chat.py
-    # Attention
-    # num_heads2 broken by https://github.com/flashinfer-ai/flashinfer/issues/1353
-    - pytest -v -s tests/kernels/attention/test_attention_selector.py
-    - pytest -v -s tests/kernels/attention/test_flashinfer.py -k 'not num_heads2'
-    - pytest -v -s tests/kernels/attention/test_flashinfer_trtllm_attention.py
-    - pytest -v -s tests/kernels/attention/test_cutlass_mla_decode.py
-    - pytest -v -s tests/kernels/attention/test_flashinfer_mla_decode.py
-    # Quantization
-    - pytest -v -s tests/kernels/quantization/test_cutlass_scaled_mm.py -k 'fp8'
-    - pytest -v -s tests/kernels/quantization/test_nvfp4_quant.py
-    - pytest -v -s tests/kernels/quantization/test_silu_mul_nvfp4_quant.py
-    - pytest -v -s tests/kernels/quantization/test_nvfp4_scaled_mm.py
-    - pytest -v -s tests/kernels/quantization/test_flashinfer_scaled_mm.py
-    - pytest -v -s tests/kernels/quantization/test_flashinfer_nvfp4_scaled_mm.py
-    - pytest -v -s tests/kernels/quantization/test_nvfp4_qutlass.py
-    - pytest -v -s tests/kernels/quantization/test_mxfp4_qutlass.py
-    - pytest -v -s tests/kernels/moe/test_nvfp4_moe.py
-    - pytest -v -s tests/kernels/moe/test_ocp_mx_moe.py
-    - pytest -v -s tests/kernels/moe/test_flashinfer.py
-    - pytest -v -s tests/kernels/moe/test_cutedsl_moe.py
@@ -1,46 +0,0 @@
-group: LM Eval
-depends_on: 
-  - image-build
-steps:
- label: LM Eval Small Models
-  timeout_in_minutes: 75
-  source_file_dependencies:
-  - csrc/
-  - vllm/model_executor/layers/quantization
-  autorun_on_main: true
-  commands:
-  - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-small.txt
-
- label: LM Eval Large Models (4 GPUs)(A100)
-  gpu: a100
-  optional: true
-  num_gpus: 4
-  working_dir: "/vllm-workspace/.buildkite/lm-eval-harness"
-  source_file_dependencies:
-  - csrc/
-  - vllm/model_executor/layers/quantization
-  commands:
-  - export VLLM_WORKER_MULTIPROC_METHOD=spawn
-  - pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-large.txt --tp-size=4
-
- label: LM Eval Large Models (4 GPUs)(H100)
-  gpu: h100
-  optional: true
-  num_gpus: 4
-  working_dir: "/vllm-workspace/.buildkite/lm-eval-harness"
-  source_file_dependencies:
-  - csrc/
-  - vllm/model_executor/layers/quantization
-  commands:
-    - export VLLM_USE_DEEP_GEMM=0  # We found Triton is faster than DeepGEMM for H100
-    - pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-large-hopper.txt --tp-size=4
-
- label: LM Eval Small Models (B200)
-  timeout_in_minutes: 120
-  gpu: b200
-  optional: true
-  source_file_dependencies:
-  - csrc/
-  - vllm/model_executor/layers/quantization
-  commands:
-  - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-blackwell.txt
@@ -1,33 +0,0 @@
-group: LoRA
-depends_on: 
-  - image-build
-steps:
- label: LoRA %N
-  timeout_in_minutes: 30
-  source_file_dependencies:
-  - vllm/lora
-  - tests/lora
-  commands:
-    - pytest -v -s lora --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --ignore=lora/test_chatglm3_tp.py --ignore=lora/test_llama_tp.py --ignore=lora/test_llm_with_multi_loras.py --ignore=lora/test_olmoe_tp.py --ignore=lora/test_deepseekv2_tp.py --ignore=lora/test_gptoss_tp.py --ignore=lora/test_qwen3moe_tp.py
-  parallelism: 4
-
-
- label: LoRA TP (Distributed)
-  timeout_in_minutes: 30
-  num_gpus: 4
-  source_file_dependencies:
-  - vllm/lora
-  - tests/lora
-  commands:
-    # FIXIT: find out which code initialize cuda before running the test
-    # before the fix, we need to use spawn to test it
-    - export VLLM_WORKER_MULTIPROC_METHOD=spawn
-    # Alot of these tests are on the edge of OOMing
-    - export PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True
-    # There is some Tensor Parallelism related processing logic in LoRA that
-    # requires multi-GPU testing for validation.
-    - pytest -v -s -x lora/test_chatglm3_tp.py
-    - pytest -v -s -x lora/test_llama_tp.py
-    - pytest -v -s -x lora/test_llm_with_multi_loras.py
-    - pytest -v -s -x lora/test_olmoe_tp.py
-    - pytest -v -s -x lora/test_gptoss_tp.py
@@ -1,165 +0,0 @@
-group: Miscellaneous
-depends_on: 
-  - image-build
-steps:
- label: V1 Others
-  timeout_in_minutes: 60
-  source_file_dependencies:
-    - vllm/
-    - tests/v1
-  commands:
-    - uv pip install --system -r /vllm-workspace/requirements/kv_connectors.txt
-    # split the test to avoid interference
-    - pytest -v -s -m 'not cpu_test' v1/core
-    - pytest -v -s v1/executor
-    - pytest -v -s v1/kv_offload
-    - pytest -v -s v1/sample
-    - pytest -v -s v1/logits_processors
-    - pytest -v -s v1/worker
-    - pytest -v -s v1/spec_decode
-    - pytest -v -s -m 'not cpu_test' v1/kv_connector/unit
-    - pytest -v -s -m 'not cpu_test' v1/metrics
-    - pytest -v -s v1/test_oracle.py
-    - pytest -v -s v1/test_request.py
-    - pytest -v -s v1/test_outputs.py
-    # Integration test for streaming correctness (requires special branch).
-    - pip install -U git+https://github.com/robertgshaw2-redhat/lm-evaluation-harness.git@streaming-api
-    - pytest -v -s entrypoints/openai/correctness/test_lmeval.py::test_lm_eval_accuracy_v1_engine
-
- label: V1 Others (CPU)
-  depends_on: ~
-  source_file_dependencies:
-    - vllm/
-    - tests/v1
-  no_gpu: true
-  commands:
-    # split the test to avoid interference
-    - pytest -v -s -m 'cpu_test' v1/core
-    - pytest -v -s v1/structured_output
-    - pytest -v -s v1/test_serial_utils.py
-    - pytest -v -s -m 'cpu_test' v1/kv_connector/unit
-    - pytest -v -s -m 'cpu_test' v1/metrics
-
- label: Regression
-  timeout_in_minutes: 20
-  source_file_dependencies:
-  - vllm/
-  - tests/test_regression
-  commands:
-  - pip install modelscope
-  - pytest -v -s test_regression.py
-  working_dir: "/vllm-workspace/tests" # optional
-
- label: Examples
-  timeout_in_minutes: 45
-  working_dir: "/vllm-workspace/examples"
-  source_file_dependencies:
-  - vllm/entrypoints
-  - vllm/multimodal
-  - examples/
-  commands:
-    - pip install tensorizer # for tensorizer test
-    - python3 offline_inference/basic/chat.py # for basic
-    - python3 offline_inference/basic/generate.py --model facebook/opt-125m
-    - python3 offline_inference/basic/generate.py --model meta-llama/Llama-2-13b-chat-hf --cpu-offload-gb 10
-    - python3 offline_inference/basic/classify.py
-    - python3 offline_inference/basic/embed.py
-    - python3 offline_inference/basic/score.py
-    # for multi-modal models
-    - python3 offline_inference/audio_language.py --seed 0
-    - python3 offline_inference/vision_language.py --seed 0
-    - python3 offline_inference/vision_language_multi_image.py --seed 0
-    - python3 offline_inference/encoder_decoder_multimodal.py --model-type whisper --seed 0
-     # for pooling models
-    - python3 pooling/pooling/vision_language_pooling.py --seed 0
-    # for features demo
-    - python3 offline_inference/prefix_caching.py
-    - python3 offline_inference/llm_engine_example.py
-    - python3 others/tensorize_vllm_model.py --model facebook/opt-125m serialize --serialized-directory /tmp/ --suffix v1 && python3 others/tensorize_vllm_model.py --model facebook/opt-125m deserialize --path-to-tensors /tmp/vllm/facebook/opt-125m/v1/model.tensors
-    - python3 offline_inference/spec_decode.py --test --method eagle --num_spec_tokens 3 --dataset-name hf --dataset-path philschmid/mt-bench --num-prompts 80 --temp 0 --top-p 1.0 --top-k -1 --tp 1 --enable-chunked-prefill --max-model-len 2048
-    # https://github.com/vllm-project/vllm/pull/26682 uses slightly more memory in PyTorch 2.9+ causing this test to OOM in 1xL4 GPU
-    - python3 offline_inference/spec_decode.py --test --method eagle3 --num_spec_tokens 3 --dataset-name hf --dataset-path philschmid/mt-bench --num-prompts 80 --temp 0 --top-p 1.0 --top-k -1 --tp 1 --enable-chunked-prefill --max-model-len 1536
-
- label: Metrics, Tracing (2 GPUs)
-  timeout_in_minutes: 20
-  num_gpus: 2
-  source_file_dependencies:
-  - vllm/
-  - tests/v1/tracing
-  commands:
-  - "pip install \
-      'opentelemetry-sdk>=1.26.0' \
-      'opentelemetry-api>=1.26.0' \
-      'opentelemetry-exporter-otlp>=1.26.0' \
-      'opentelemetry-semantic-conventions-ai>=0.4.1'"
-  - pytest -v -s v1/tracing
-
- label: Python-only Installation
-  depends_on: ~
-  timeout_in_minutes: 20
-  source_file_dependencies:
-  - tests/standalone_tests/python_only_compile.sh
-  - setup.py
-  commands:
-  - bash standalone_tests/python_only_compile.sh
-
- label: Async Engine, Inputs, Utils, Worker
-  timeout_in_minutes: 50
-  source_file_dependencies:
-  - vllm/
-  - tests/multimodal
-  - tests/utils_
-  commands:
-  - pytest -v -s -m 'not cpu_test' multimodal
-  - pytest -v -s utils_
-
- label: Async Engine, Inputs, Utils, Worker, Config (CPU)
-  depends_on: ~
-  timeout_in_minutes: 30
-  source_file_dependencies:
-  - vllm/
-  - tests/test_inputs.py
-  - tests/test_outputs.py
-  - tests/multimodal
-  - tests/standalone_tests/lazy_imports.py
-  - tests/tokenizers_
-  - tests/tool_parsers
-  - tests/transformers_utils
-  - tests/config
-  no_gpu: true
-  commands:
-  - python3 standalone_tests/lazy_imports.py
-  - pytest -v -s test_inputs.py
-  - pytest -v -s test_outputs.py
-  - pytest -v -s -m 'cpu_test' multimodal
-  - pytest -v -s tokenizers_
-  - pytest -v -s tool_parsers
-  - pytest -v -s transformers_utils
-  - pytest -v -s config
-
- label: GPT-OSS Eval (B200)
-  timeout_in_minutes: 60
-  working_dir: "/vllm-workspace/"
-  gpu: b200
-  optional: true
-  source_file_dependencies:
-  - tests/evals/gpt_oss
-  - vllm/model_executor/models/gpt_oss.py
-  - vllm/model_executor/layers/quantization/mxfp4.py
-  - vllm/v1/attention/backends/flashinfer.py
-  commands:
-    - uv pip install --system 'gpt-oss[eval]==0.0.5'
-    - pytest -s -v tests/evals/gpt_oss/test_gpqa_correctness.py --model openai/gpt-oss-20b --metric 0.58
-
- label: Batch Invariance (H100)
-  timeout_in_minutes: 25
-  gpu: h100
-  source_file_dependencies:
-    - vllm/v1/attention
-    - vllm/model_executor/layers
-    - tests/v1/determinism/
-  commands:
-    - export VLLM_WORKER_MULTIPROC_METHOD=spawn
-    - pip install pytest-timeout pytest-forked
-    - pytest -v -s v1/determinism/test_batch_invariance.py
-    - pytest -v -s v1/determinism/test_rms_norm_batch_invariant.py
@@ -1,17 +0,0 @@
-group: Model Executor
-depends_on: 
-  - image-build
-steps:
- label: Model Executor
-  timeout_in_minutes: 35
-  source_file_dependencies:
-  - vllm/engine/arg_utils.py
-  - vllm/config/model.py
-  - vllm/model_executor
-  - tests/model_executor
-  - tests/entrypoints/openai/test_tensorizer_entrypoint.py
-  commands:
-    - apt-get update && apt-get install -y curl libsodium23
-    - export VLLM_WORKER_MULTIPROC_METHOD=spawn
-    - pytest -v -s model_executor
-    - pytest -v -s entrypoints/openai/test_tensorizer_entrypoint.py
@@ -1,64 +0,0 @@
-group: Models - Basic
-depends_on: 
-  - image-build
-steps:
- label: Basic Models Tests (Initialization)
-  timeout_in_minutes: 45
-  mirror_hardwares: [amdexperimental]
-  torch_nightly: true
-  source_file_dependencies:
-  - vllm/
-  - tests/models/test_initialization.py
-  - tests/models/registry.py
-  commands:
-    # Run a subset of model initialization tests
-    - pytest -v -s models/test_initialization.py::test_can_initialize_small_subset
-
- label: Basic Models Tests (Extra Initialization) %N
-  timeout_in_minutes: 45
-  mirror_hardwares: [amdexperimental]
-  torch_nightly: true
-  source_file_dependencies:
-  - vllm/model_executor/models/
-  - tests/models/test_initialization.py
-  - tests/models/registry.py
-  commands:
-    # Only when vLLM model source is modified - test initialization of a large
-    # subset of supported models (the complement of the small subset in the above
-    # test.) Also run if model initialization test file is modified
-    - pytest -v -s models/test_initialization.py -k 'not test_can_initialize_small_subset' --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
-  parallelism: 2
-
- label: Basic Models Tests (Other)
-  timeout_in_minutes: 45
-  source_file_dependencies:
-  - vllm/
-  - tests/models/test_transformers.py
-  - tests/models/test_registry.py
-  commands:
-    - pytest -v -s models/test_transformers.py models/test_registry.py
-
- label: Basic Models Test (Other CPU) # 5min
-  timeout_in_minutes: 10
-  source_file_dependencies:
-  - vllm/
-  - tests/models/test_utils.py
-  - tests/models/test_vision.py
-  no_gpu: true
-  commands:
-    - pytest -v -s models/test_utils.py models/test_vision.py
-
- label: Transformers Nightly Models
-  working_dir: "/vllm-workspace/"
-  optional: true
-  soft_fail: true
-  commands:
-    - pip install --upgrade git+https://github.com/huggingface/transformers
-    - pytest -v -s tests/models/test_initialization.py
-    - pytest -v -s tests/models/test_transformers.py
-    - pytest -v -s tests/models/multimodal/processing/
-    - pytest -v -s tests/models/multimodal/test_mapping.py
-    - python3 examples/offline_inference/basic/chat.py
-    - python3 examples/offline_inference/vision_language.py --model-type qwen2_5_vl
-    # Whisper needs spawn method to avoid deadlock
-    - VLLM_WORKER_MULTIPROC_METHOD=spawn python3 examples/offline_inference/audio_language.py --model-type whisper
@@ -1,22 +0,0 @@
-group: Models - Distributed
-depends_on: 
-  - image-build
-steps:
- label: Distributed Model Tests (2 GPUs)
-  timeout_in_minutes: 50
-  working_dir: "/vllm-workspace/tests"
-  num_gpus: 2
-  source_file_dependencies:
-  - vllm/model_executor/model_loader/sharded_state_loader.py
-  - vllm/model_executor/models/
-  - tests/basic_correctness/
-  - tests/model_executor/model_loader/test_sharded_state_loader.py
-  - tests/models/
-  commands:
-  - TARGET_TEST_SUITE=L4 pytest basic_correctness/ -v -s -m 'distributed(num_gpus=2)'
-  - CUDA_VISIBLE_DEVICES=0,1 pytest -v -s model_executor/model_loader/test_sharded_state_loader.py
-  # Avoid importing model tests that cause CUDA reinitialization error
-  - pytest models/test_transformers.py -v -s -m 'distributed(num_gpus=2)'
-  - pytest models/language -v -s -m 'distributed(num_gpus=2)'
-  - pytest models/multimodal -v -s -m 'distributed(num_gpus=2)' --ignore models/multimodal/generation/test_whisper.py
-  - VLLM_WORKER_MULTIPROC_METHOD=spawn pytest models/multimodal/generation/test_whisper.py -v -s -m 'distributed(num_gpus=2)'
@@ -1,91 +0,0 @@
-group: Models - Language
-depends_on: 
-  - image-build
-steps:
- label: Language Models Tests (Standard)
-  timeout_in_minutes: 25
-  mirror_hardwares: [amdexperimental]
-  torch_nightly: true
-  source_file_dependencies:
-  - vllm/
-  - tests/models/language
-  commands:
-    # Test standard language models, excluding a subset of slow tests
-    - pip freeze | grep -E 'torch'
-    - pytest -v -s models/language -m 'core_model and (not slow_test)'
-
- label: Language Models Tests (Extra Standard) %N
-  timeout_in_minutes: 45
-  mirror_hardwares: [amdexperimental]
-  torch_nightly: true
-  source_file_dependencies:
-  - vllm/model_executor/models/
-  - tests/models/language/pooling/test_embedding.py
-  - tests/models/language/generation/test_common.py
-  - tests/models/language/pooling/test_classification.py
-  commands:
-    # Shard slow subset of standard language models tests. Only run when model
-    # source is modified, or when specified test files are modified
-    - pip freeze | grep -E 'torch'
-    - pytest -v -s models/language -m 'core_model and slow_test' --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
-  parallelism: 2
-
- label: Language Models Tests (Hybrid) %N
-  timeout_in_minutes: 75
-  mirror_hardwares: [amdexperimental]
-  torch_nightly: true
-  source_file_dependencies:
-  - vllm/
-  - tests/models/language/generation
-  commands:
-    # Install fast path packages for testing against transformers
-    # Note: also needed to run plamo2 model in vLLM
-    - uv pip install --system --no-build-isolation 'git+https://github.com/state-spaces/mamba@v2.2.5'
-    - uv pip install --system --no-build-isolation 'git+https://github.com/Dao-AILab/causal-conv1d@v1.5.2'
-    # Shard hybrid language model tests
-    - pytest -v -s models/language/generation -m hybrid_model --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --shard-id=$$BUILDKITE_PARALLEL_JOB
-  parallelism: 2
-
- label: Language Models Test (Extended Generation) # 80min
-  timeout_in_minutes: 110
-  mirror_hardwares: [amdexperimental]
-  optional: true
-  source_file_dependencies:
-  - vllm/
-  - tests/models/language/generation
-  commands:
-    # Install fast path packages for testing against transformers
-    # Note: also needed to run plamo2 model in vLLM
-    - uv pip install --system --no-build-isolation 'git+https://github.com/state-spaces/mamba@v2.2.5'
-    - uv pip install --system --no-build-isolation 'git+https://github.com/Dao-AILab/causal-conv1d@v1.5.2'
-    - pytest -v -s models/language/generation -m '(not core_model) and (not hybrid_model)'
-
- label: Language Models Test (PPL)
-  timeout_in_minutes: 110
-  mirror_hardwares: [amdexperimental]
-  optional: true
-  source_file_dependencies:
-  - vllm/
-  - tests/models/language/generation_ppl_test
-  commands:
-    - pytest -v -s models/language/generation_ppl_test
-
- label: Language Models Test (Extended Pooling)  # 36min
-  timeout_in_minutes: 50
-  mirror_hardwares: [amdexperimental]
-  optional: true
-  source_file_dependencies:
-  - vllm/
-  - tests/models/language/pooling
-  commands:
-    - pytest -v -s models/language/pooling -m 'not core_model'
-
- label: Language Models Test (MTEB)
-  timeout_in_minutes: 110
-  mirror_hardwares: [amdexperimental]
-  optional: true
-  source_file_dependencies:
-  - vllm/
-  - tests/models/language/pooling_mteb_test
-  commands:
-    - pytest -v -s models/language/pooling_mteb_test
@@ -1,79 +0,0 @@
-group: Models - Multimodal
-depends_on: 
-  - image-build
-steps:
- label: Multi-Modal Models (Standard) # 60min
-  timeout_in_minutes: 80
-  source_file_dependencies:
-  - vllm/
-  - tests/models/multimodal
-  commands:
-    - pip install git+https://github.com/TIGER-AI-Lab/Mantis.git
-    - pip freeze | grep -E 'torch'
-    - pytest -v -s models/multimodal -m core_model --ignore models/multimodal/generation/test_whisper.py --ignore models/multimodal/processing
-    - cd .. && VLLM_WORKER_MULTIPROC_METHOD=spawn pytest -v -s tests/models/multimodal/generation/test_whisper.py -m core_model  # Otherwise, mp_method="spawn" doesn't work
-
- label: Multi-Modal Processor Test (CPU)
-  timeout_in_minutes: 60
-  source_file_dependencies:
-  - vllm/
-  - tests/models/multimodal
-  no_gpu: true
-  commands:
-    - pip install git+https://github.com/TIGER-AI-Lab/Mantis.git
-    - pytest -v -s models/multimodal/processing --ignore models/multimodal/processing/test_tensor_schema.py
-
- label: Multi-Modal Processor # 44min
-  timeout_in_minutes: 60
-  source_file_dependencies:
-  - vllm/
-  - tests/models/multimodal
-  commands:
-    - pip install git+https://github.com/TIGER-AI-Lab/Mantis.git
-    - pytest -v -s models/multimodal/processing/test_tensor_schema.py
-
- label: Multi-Modal Accuracy Eval (Small Models) # 50min
-  timeout_in_minutes: 70
-  working_dir: "/vllm-workspace/.buildkite/lm-eval-harness"
-  source_file_dependencies:
-  - vllm/multimodal/
-  - vllm/inputs/
-  - vllm/v1/core/
-  commands:
-  - pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-mm-small.txt --tp-size=1
-
- label: Multi-Modal Models (Extended) 1
-  optional: true
-  source_file_dependencies:
-  - vllm/
-  - tests/models/multimodal
-  commands:
-    - pip install git+https://github.com/TIGER-AI-Lab/Mantis.git
-    - pytest -v -s models/multimodal -m 'not core_model' --ignore models/multimodal/generation/test_common.py --ignore models/multimodal/processing
-
- label: Multi-Modal Models (Extended) 2
-  optional: true
-  source_file_dependencies:
-  - vllm/
-  - tests/models/multimodal
-  commands:
-    - pip install git+https://github.com/TIGER-AI-Lab/Mantis.git
-    - pytest -v -s models/multimodal/generation/test_common.py -m 'split(group=0) and not core_model'
-
- label: Multi-Modal Models (Extended) 3
-  optional: true
-  source_file_dependencies:
-  - vllm/
-  - tests/models/multimodal
-  commands:
-    - pip install git+https://github.com/TIGER-AI-Lab/Mantis.git
-    - pytest -v -s models/multimodal/generation/test_common.py -m 'split(group=1) and not core_model'
-
-# This test is used only in PR development phase to test individual models and should never run on main
- label: Custom Models
-  optional: true
-  commands:
-    - echo 'Testing custom models...'
-    # PR authors can temporarily add commands below to test individual models
-    # e.g. pytest -v -s models/encoder_decoder/vision_language/test_mllama.py
-    # *To avoid merge conflicts, remember to REMOVE (not just comment out) them before merging the PR*
@@ -1,34 +0,0 @@
-group: Plugins
-depends_on: 
-  - image-build
-steps:
- label: Plugin Tests (2 GPUs)
-  timeout_in_minutes: 60
-  working_dir: "/vllm-workspace/tests"
-  num_gpus: 2
-  source_file_dependencies:
-  - vllm/plugins/
-  - tests/plugins/
-  commands:
-  # begin platform plugin and general plugin tests, all the code in-between runs on dummy platform
-  - pip install -e ./plugins/vllm_add_dummy_platform
-  - pytest -v -s plugins_tests/test_platform_plugins.py
-  - pip uninstall vllm_add_dummy_platform -y
-  # end platform plugin tests
-  # begin io_processor plugins test, all the code in between uses the prithvi_io_processor plugin
-  - pip install -e ./plugins/prithvi_io_processor_plugin
-  - pytest -v -s plugins_tests/test_io_processor_plugins.py
-  - pip uninstall prithvi_io_processor_plugin -y
-  # end io_processor plugins test
-  # begin stat_logger plugins test
-  - pip install -e ./plugins/vllm_add_dummy_stat_logger
-  - pytest -v -s plugins_tests/test_stats_logger_plugins.py
-  - pip uninstall dummy_stat_logger -y
-  # end stat_logger plugins test
-  # other tests continue here:
-  - pytest -v -s plugins_tests/test_scheduler_plugins.py
-  - pip install -e ./plugins/vllm_add_dummy_model
-  - pytest -v -s distributed/test_distributed_oot.py
-  - pytest -v -s entrypoints/openai/test_oot_registration.py # it needs a clean process
-  - pytest -v -s models/test_oot_registration.py # it needs a clean process
-  - pytest -v -s plugins/lora_resolvers # unit tests for in-tree lora resolver plugins
@@ -1,52 +0,0 @@
-group: PyTorch
-depends_on: 
-  - image-build
-steps:
- label: PyTorch Compilation Unit Tests
-  timeout_in_minutes: 30
-  source_file_dependencies:
-    - vllm/
-    - tests/compile
-  commands:
-  # Run unit tests defined directly under compile/,
-  # not including subdirectories, which are usually heavier
-  # tests covered elsewhere.
-  # Use `find` to launch multiple instances of pytest so that
-  # they do not suffer from https://github.com/vllm-project/vllm/issues/28965
-  # However, find does not normally propagate error codes, so we combine it with xargs
-  # (using -0 for proper path handling)
-  - "find compile/ -maxdepth 1 -name 'test_*.py' -print0 | xargs -0 -n1 -I{} pytest -s -v '{}'"
-
- label: PyTorch Fullgraph Smoke Test
-  timeout_in_minutes: 30
-  source_file_dependencies:
-  - vllm/
-  - tests/compile
-  commands:
-  # Run smoke tests under fullgraph directory, except test_full_graph.py
-  # as it is a heavy test that is covered in other steps.
-  # Use `find` to launch multiple instances of pytest so that
-  # they do not suffer from https://github.com/vllm-project/vllm/issues/28965
-  - "find compile/fullgraph/ -name 'test_*.py' -not -name 'test_full_graph.py' -exec pytest -s -v {} \\;"
-
- label: PyTorch Fullgraph
-  timeout_in_minutes: 40
-  source_file_dependencies:
-  - vllm/
-  - tests/compile
-  commands:
-    # fp8 kv scales not supported on sm89, tested on Blackwell instead
-  - pytest -v -s compile/fullgraph/test_full_graph.py -k 'not test_fp8_kv_scale_compile'
-    # Limit to no custom ops to reduce running time
-    # Wrap with quotes to escape yaml and avoid starting -k string with a -
-  - "pytest -v -s compile/distributed/test_fusions_e2e.py -k 'TRITON and not +quant_fp8 and not Llama-4'"
-
- label: Pytorch Nightly Dependency Override Check # 2min
-  # if this test fails, it means the nightly torch version is not compatible with some
-  # of the dependencies. Please check the error message and add the package to whitelist
-  # in /vllm/tools/pre_commit/generate_nightly_torch_test.py
-  soft_fail: true
-  source_file_dependencies:
-  - requirements/nightly_torch_test.txt
-  commands:
-  - bash standalone_tests/pytorch_nightly_dependency.sh
@@ -1,46 +0,0 @@
-group: Quantization
-depends_on: 
-  - image-build
-steps:
- label: Quantization
-  timeout_in_minutes: 90
-  source_file_dependencies:
-  - csrc/
-  - vllm/model_executor/layers/quantization
-  - tests/quantization
-  commands:
-  # temporary install here since we need nightly, will move to requirements/test.in
-  # after torchao 0.12 release, and pin a working version of torchao nightly here
-
-  # since torchao nightly is only compatible with torch nightly currently
-  # https://github.com/pytorch/ao/issues/2919, we'll have to skip new torchao tests for now
-  # we can only upgrade after this is resolved
-  # TODO(jerryzh168): resolve the above comment
-  - uv pip install --system torchao==0.13.0 --index-url https://download.pytorch.org/whl/cu129
-  - uv pip install --system conch-triton-kernels
-  - VLLM_TEST_FORCE_LOAD_FORMAT=auto pytest -v -s quantization/ --ignore quantization/test_blackwell_moe.py
-
- label: Quantized MoE Test (B200)
-  timeout_in_minutes: 60
-  working_dir: "/vllm-workspace/"
-  gpu: b200
-  source_file_dependencies:
-  - tests/quantization/test_blackwell_moe.py
-  - vllm/model_executor/models/deepseek_v2.py
-  - vllm/model_executor/models/gpt_oss.py
-  - vllm/model_executor/models/llama4.py
-  - vllm/model_executor/layers/fused_moe
-  - vllm/model_executor/layers/quantization/compressed_tensors
-  - vllm/model_executor/layers/quantization/modelopt.py
-  - vllm/model_executor/layers/quantization/mxfp4.py
-  - vllm/v1/attention/backends/flashinfer.py
-  commands:
-    - pytest -s -v tests/quantization/test_blackwell_moe.py
-
- label: Quantized Models Test
-  timeout_in_minutes: 60
-  source_file_dependencies:
-  - vllm/model_executor/layers/quantization
-  - tests/models/quantization
-  commands:
-    - pytest -v -s models/quantization
@@ -1,14 +0,0 @@
-group: Samplers
-depends_on: 
-  - image-build
-steps:
- label: Samplers Test
-  timeout_in_minutes: 75
-  source_file_dependencies:
-  - vllm/model_executor/layers
-  - vllm/sampling_metadata.py
-  - tests/samplers
-  - tests/conftest.py
-  commands:
-    - pytest -v -s samplers
-    - VLLM_USE_FLASHINFER_SAMPLER=1 pytest -v -s samplers
@@ -1,25 +0,0 @@
-group: Weight Loading
-depends_on: 
-  - image-build
-steps:
- label: Weight Loading Multiple GPU  # 33min
-  timeout_in_minutes: 45
-  working_dir: "/vllm-workspace/tests"
-  num_gpus: 2
-  optional: true
-  source_file_dependencies:
-  - vllm/
-  - tests/weight_loading
-  commands:
-    - bash weight_loading/run_model_weight_loading_test.sh -c weight_loading/models.txt
-
- label: Weight Loading Multiple GPU - Large Models # optional
-  working_dir: "/vllm-workspace/tests"
-  num_gpus: 2
-  gpu: a100
-  optional: true
-  source_file_dependencies:
-  - vllm/
-  - tests/weight_loading
-  commands:
-    - bash weight_loading/run_model_weight_loading_test.sh -c weight_loading/models-large.txt
@@ -3,49 +3,44 @@

 # This lists cover the "core" components of vLLM that require careful review
 /vllm/attention @LucasWilkinson
-/vllm/attention/backends/abstract.py @WoosukKwon @zhuohan123 @youkaichao @alexm-redhat @njhill
-/vllm/executor/executor_base.py @zhuohan123 @youkaichao @alexm-redhat @njhill @22quinn
+/vllm/attention/backends/abstract.py @WoosukKwon @zhuohan123 @youkaichao @alexm-redhat @comaniac @njhill
+/vllm/executor/executor_base.py @zhuohan123 @youkaichao @alexm-redhat @comaniac @njhill @22quinn
 /vllm/model_executor/layers/fused_moe @mgoin @pavanimajety
 /vllm/model_executor/layers/quantization @mgoin @robertgshaw2-redhat @tlrmchlsmth @yewentao256 @pavanimajety
 /vllm/model_executor/layers/mamba @tdoublep
 /vllm/model_executor/model_loader @22quinn
-/vllm/model_executor/layers/batch_invariant.py @yewentao256 
-/vllm/multimodal @DarkLight1337 @ywang96 @NickLucche @tjtanaa
+/vllm/multimodal @DarkLight1337 @ywang96 @NickLucche
 /vllm/vllm_flash_attn @LucasWilkinson
 /vllm/lora @jeejeelee
 /vllm/reasoning @aarnphm @chaunceyjiang
 /vllm/entrypoints @aarnphm @chaunceyjiang
-/vllm/tool_parsers @aarnphm @chaunceyjiang
 /vllm/compilation @zou3519 @youkaichao @ProExpertProg
 /vllm/distributed/kv_transfer @NickLucche @ApostaC
 CMakeLists.txt @tlrmchlsmth @LucasWilkinson

 # Any change to the VllmConfig changes can have a large user-facing impact,
 # so spam a lot of people
-/vllm/config @WoosukKwon @youkaichao @robertgshaw2-redhat @mgoin @tlrmchlsmth @houseroad @hmellor @yewentao256 @ProExpertProg
-/vllm/config/cache.py @WoosukKwon @youkaichao @robertgshaw2-redhat @mgoin @tlrmchlsmth @houseroad @hmellor @yewentao256 @ProExpertProg @heheda12345
+/vllm/config @simon-mo @WoosukKwon @youkaichao @robertgshaw2-redhat @mgoin @tlrmchlsmth @houseroad @hmellor @yewentao256 @ProExpertProg
+/vllm/config/cache.py @simon-mo @WoosukKwon @youkaichao @robertgshaw2-redhat @mgoin @tlrmchlsmth @houseroad @hmellor @yewentao256 @ProExpertProg @heheda12345

 # vLLM V1
 /vllm/v1/attention @LucasWilkinson
 /vllm/v1/attention/backends/mla @pavanimajety
 /vllm/v1/attention/backends/flashinfer.py @mgoin @pavanimajety
 /vllm/v1/attention/backends/triton_attn.py @tdoublep
-/vllm/v1/core @WoosukKwon @robertgshaw2-redhat @njhill @ywang96 @alexm-redhat @heheda12345 @ApostaC
+/vllm/v1/core @WoosukKwon @robertgshaw2-redhat @njhill @ywang96 @comaniac @alexm-redhat @heheda12345 @ApostaC
 /vllm/v1/sample @22quinn @houseroad @njhill
 /vllm/v1/spec_decode @benchislett @luccafong
 /vllm/v1/structured_output @mgoin @russellb @aarnphm @benchislett
 /vllm/v1/kv_cache_interface.py @heheda12345
 /vllm/v1/offloading @ApostaC

-# Model runner V2
-/vllm/v1/worker/gpu @WoosukKwon
-
 # Test ownership
-/.buildkite/lm-eval-harness @mgoin 
+/.buildkite/lm-eval-harness @mgoin @simon-mo
 /tests/distributed/test_multi_node_assignment.py @youkaichao
 /tests/distributed/test_pipeline_parallel.py @youkaichao
 /tests/distributed/test_same_node.py @youkaichao
-/tests/entrypoints @DarkLight1337 @robertgshaw2-redhat @aarnphm @NickLucche
+/tests/entrypoints @DarkLight1337 @robertgshaw2-redhat @simon-mo @aarnphm @NickLucche
 /tests/evals @mgoin
 /tests/kernels @mgoin @tlrmchlsmth @WoosukKwon @yewentao256
 /tests/models @DarkLight1337 @ywang96
@@ -54,29 +49,18 @@ CMakeLists.txt @tlrmchlsmth @LucasWilkinson
 /tests/test_inputs.py @DarkLight1337 @ywang96
 /tests/v1/entrypoints/llm/test_struct_output_generate.py @mgoin @russellb @aarnphm
 /tests/v1/structured_output @mgoin @russellb @aarnphm
-/tests/v1/core @WoosukKwon @robertgshaw2-redhat @njhill @ywang96 @alexm-redhat @heheda12345 @ApostaC
+/tests/v1/core @WoosukKwon @robertgshaw2-redhat @njhill @ywang96 @comaniac @alexm-redhat @heheda12345 @ApostaC
 /tests/weight_loading @mgoin @youkaichao @yewentao256
 /tests/lora @jeejeelee
 /tests/models/language/generation/test_hybrid.py @tdoublep
 /tests/v1/kv_connector/nixl_integration @NickLucche
 /tests/v1/kv_connector @ApostaC
 /tests/v1/offloading @ApostaC
-/tests/v1/determinism @yewentao256 

-# Transformers modeling backend
+# Transformers backend
 /vllm/model_executor/models/transformers @hmellor
 /tests/models/test_transformers.py @hmellor

-# Observability
-/vllm/config/observability.py @markmc
-/vllm/v1/metrics @markmc
-/tests/v1/metrics @markmc
-/vllm/tracing.py @markmc
-/tests/v1/tracing/test_tracing.py @markmc
-/vllm/config/kv_events.py @markmc
-/vllm/distributed/kv_events.py @markmc
-/tests/distributed/test_events.py @markmc
-
 # Docs
 /docs/mkdocs @hmellor
 /docs/**/*.yml @hmellor
@@ -121,21 +105,11 @@ mkdocs.yaml @hmellor
 /vllm/attention/ops/triton_unified_attention.py @tdoublep

 # ROCm related: specify owner with write access to notify AMD folks for careful code review
-/vllm/**/*rocm* @tjtanaa
-/docker/Dockerfile.rocm* @gshtras @tjtanaa
-/vllm/v1/attention/backends/rocm*.py @gshtras @tjtanaa
-/vllm/v1/attention/backends/mla/rocm*.py @gshtras @tjtanaa
-/vllm/attention/ops/rocm*.py @gshtras @tjtanaa
-/vllm/model_executor/layers/fused_moe/rocm*.py @gshtras @tjtanaa
-/csrc/rocm @gshtras @tjtanaa
-/requirements/*rocm* @tjtanaa
-/tests/**/*rocm* @tjtanaa
-/docs/**/*rocm* @tjtanaa
-/vllm/**/*quark* @tjtanaa
-/tests/**/*quark* @tjtanaa
-/docs/**/*quark* @tjtanaa
-/vllm/**/*aiter* @tjtanaa
-/tests/**/*aiter* @tjtanaa
+/docker/Dockerfile.rocm* @gshtras
+/vllm/v1/attention/backends/rocm*.py @gshtras
+/vllm/v1/attention/backends/mla/rocm*.py @gshtras
+/vllm/attention/ops/rocm*.py @gshtras
+/vllm/model_executor/layers/fused_moe/rocm*.py @gshtras

 # TPU
 /vllm/v1/worker/tpu* @NickLucche
@@ -147,15 +121,9 @@ mkdocs.yaml @hmellor
 /requirements/kv_connectors.txt @NickLucche

 # Pooling models
-/examples/pooling @noooop
+/examples/*/pooling/ @noooop
 /tests/models/*/pooling* @noooop
 /tests/entrypoints/pooling @noooop
-/vllm/entrypoints/pooling @noooop
 /vllm/config/pooler.py @noooop
 /vllm/pooling_params.py @noooop
 /vllm/model_executor/layers/pooler.py @noooop
-
-# Security guide and policies
-/docs/usage/security.md @russellb
-/SECURITY.md @russellb
-/docs/contributing/vulnerability_management.md @russellb
@@ -14,52 +14,6 @@ pull_request_rules:
    comment:
      message: "Documentation preview: https://vllm--{{number}}.org.readthedocs.build/en/{{number}}/"

- name: comment-pre-commit-failure
-  description: Comment on PR when pre-commit check fails
-  conditions:
-    - status-failure=pre-commit
-    - -closed
-    - -draft
-  actions:
-    comment:
-      message: |
-        Hi @{{author}}, the pre-commit checks have failed. Please run:
-
-        ```bash 
-        uv pip install pre-commit
-        pre-commit install
-        pre-commit run --all-files
-        ```
-
-        Then, commit the changes and push to your branch.
-
-        For future commits, `pre-commit` will run automatically on changed files before each commit.
-
-        > [!TIP]
-        > <details>
-        > <summary>Is <code>mypy</code> or <code>markdownlint</code> failing?</summary>
-        > <br/>
-        > <code>mypy</code> and <code>markdownlint</code> are run differently in CI. If the failure is related to either of these checks, please use the following commands to run them locally:
-        >
-        > ```bash
-        > # For mypy (substitute "3.10" with the failing version if needed)
-        > pre-commit run --hook-stage manual mypy-3.10
-        > # For markdownlint
-        > pre-commit run --hook-stage manual markdownlint
-        > ```
-        > </details>
-
- name: comment-dco-failure
-  description: Comment on PR when DCO check fails
-  conditions:
-    - status-failure=dco
-    - -closed
-    - -draft
-  actions:
-    comment:
-      message: |
-        Hi @{{author}}, the DCO check has failed. Please click on DCO in the Checks section for instructions on how to resolve this.
-
 - name: label-ci-build
  description: Automatically apply ci/build label
  conditions:
@@ -186,7 +140,7 @@ pull_request_rules:
      - files~=^tests/entrypoints/test_context.py
      - files~=^vllm/model_executor/models/.*gpt[-_]?oss.*\.py
      - files~=^vllm/model_executor/layers/.*gpt[-_]?oss.*\.py
-      - files~=^vllm/entrypoints/openai/parser/harmony_utils.py
+      - files~=^vllm/entrypoints/harmony_utils.py
      - files~=^vllm/entrypoints/tool_server.py
      - files~=^vllm/entrypoints/tool.py
      - files~=^vllm/entrypoints/context.py
@@ -197,23 +151,6 @@ pull_request_rules:
      add:
        - gpt-oss

- name: label-nvidia
-  description: Automatically apply nvidia label
-  conditions:
-    - label != stale
-    - or:
-      - files~=cuda
-      - files~=cutlass
-      - files~=flashinfer
-      - files~=trtllm
-      - title~=(?i)NVIDIA
-      - title~=(?i)CUDA
-      - title~=(?i)CUTLASS
-  actions:
-    label:
-      add:
-        - nvidia
-
 - name: label-rocm
  description: Automatically apply rocm label
  conditions:
@@ -235,20 +172,6 @@ pull_request_rules:
      add:
        - rocm

- name: label-cpu
-  description: Automatically apply cpu label
-  conditions:
-    - label != stale
-    - files~=^(?!.*kv_offload)(?!.*cpu_offload).*\bcpu.*
-  actions:
-    label:
-      add:
-        - cpu
-    assign:
-      users:
-        - "fadara01"
-        - "aditew01"
-
 - name: label-structured-output
  description: Automatically apply structured-output label
  conditions:
@@ -349,18 +272,6 @@ pull_request_rules:
      add:
        - tool-calling

- name: auto-rebase if approved, ready, and 40 commits behind main
-  conditions:
-    - base = main
-    - label=ready
-    - "#approved-reviews-by >= 1"
-    - "#commits-behind >= 40"
-    - -closed
-    - -draft
-    - -conflict
-  actions:
-    rebase: {}
-
 - name: ping author on conflicts and add 'needs-rebase' label
  conditions:
    - label != stale
@@ -430,4 +341,4 @@ pull_request_rules:
  actions:
    label:
      add:
-        - kv-connector
+        - kv-connector
@@ -13,10 +13,10 @@ jobs:

    steps:
      - name: Checkout repository
-        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2

      - name: Set up Python
-        uses: actions/setup-python@83679a892e2d95755f2dac6acb0bfd1e9ac5d548 # v6.1.0
+        uses: actions/setup-python@e797f83bcb11b83ae66e0230d6156d7c80228e7c # v6.0.0
        with:
          python-version: '3.12'

@@ -105,31 +105,6 @@ jobs:
                  }
                ],
              },
-              cpu: {
-                // Keyword search - matches whole words only (with word boundaries)
-                keywords: [
-                  {
-                    term: "CPU Backend",
-                    searchIn: "title"
-                  },
-                  {
-                    term: "x86",
-                    searchIn: "title"
-                  },
-                  {
-                    term: "ARM",
-                    searchIn: "title"
-                  },
-                  {
-                    term: "Apple Silicon",
-                    searchIn: "title"
-                  },
-                  {
-                    term: "IBM Z",
-                    searchIn: "title"
-                  },
-                ],
-              },
              // Add more label configurations here as needed
              // example: {
              //   keywords: [...],
@@ -1,80 +0,0 @@
-name: macOS Apple Silicon Smoke Test
-
-on:
-  push:
-    branches:
-      - main
-  workflow_dispatch:  # Manual trigger
-
-jobs:
-  macos-m1-smoke-test:
-    runs-on: macos-latest
-    timeout-minutes: 30
-
-    steps:
-      - uses: actions/checkout@v6.0.1
-
-      - uses: astral-sh/setup-uv@v7
-        with:
-          enable-cache: true
-          cache-dependency-glob: |
-            requirements/**/*.txt
-            pyproject.toml
-          python-version: '3.12'
-
-      - name: Create virtual environment
-        run: |
-          uv venv
-          echo "$GITHUB_WORKSPACE/.venv/bin" >> "$GITHUB_PATH"
-
-      - name: Install dependencies and build vLLM
-        run: |
-          uv pip install -r requirements/cpu.txt --index-strategy unsafe-best-match
-          uv pip install -e .
-        env:
-          CMAKE_BUILD_PARALLEL_LEVEL: 4
-
-      - name: Verify installation
-        run: |
-          python -c "import vllm; print(f'vLLM version: {vllm.__version__}')"
-
-      - name: Smoke test vllm serve
-        run: |
-          # Start server in background
-          vllm serve Qwen/Qwen3-0.6B \
-            --max-model-len=2K \
-            --load-format=dummy \
-            --hf-overrides '{"num_hidden_layers": 2}' \
-            --enforce-eager \
-            --port 8000 &
-
-          SERVER_PID=$!
-
-          # Wait for server to start
-          for i in {1..30}; do
-            if curl -s http://localhost:8000/health > /dev/null; then
-              echo "Server started successfully"
-              break
-            fi
-            if [ "$i" -eq 30 ]; then
-              echo "Server failed to start"
-              kill "$SERVER_PID"
-              exit 1
-            fi
-            sleep 2
-          done
-
-          # Test health endpoint
-          curl -f http://localhost:8000/health
-
-          # Test completion
-          curl -f http://localhost:8000/v1/completions \
-            -H "Content-Type: application/json" \
-            -d '{
-              "model": "Qwen/Qwen3-0.6B",
-              "prompt": "Hello",
-              "max_tokens": 5
-            }'
-
-          # Cleanup
-          kill "$SERVER_PID"
@@ -16,8 +16,8 @@ jobs:
  pre-commit:
    runs-on: ubuntu-latest
    steps:
-    - uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1
-    - uses: actions/setup-python@83679a892e2d95755f2dac6acb0bfd1e9ac5d548 # v6.1.0
+    - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+    - uses: actions/setup-python@e797f83bcb11b83ae66e0230d6156d7c80228e7c # v6.0.0
      with:
        python-version: "3.12"
    - run: echo "::add-matcher::.github/workflows/matchers/actionlint.json"
@@ -7,15 +7,13 @@ on:

 jobs:
  close-issues-and-pull-requests:
-    # Prevents triggering on forks or other repos
-    if: github.repository == 'vllm-project/vllm'
    permissions:
      issues: write
      pull-requests: write
      actions: write
    runs-on: ubuntu-latest
    steps:
-      - uses: actions/stale@997185467fa4f803885201cee163a9f38240193d # v10.1.1
+      - uses: actions/stale@5f858e3efba33a5ca4407a664cc011ad407f2008 # v10.1.0
        with:
          # Increasing this value ensures that changes to this workflow
          # propagate to all issues and PRs in days rather than months
@@ -4,9 +4,6 @@
 # vllm-flash-attn built from source
 vllm/vllm_flash_attn/*

-# OpenAI triton kernels copied from source
-vllm/third_party/triton_kernels/*
-
 # triton jit
 .triton

@@ -3,9 +3,10 @@ MD007:
 MD013: false
 MD024:
  siblings_only: true
-MD031:
-  list_items: false
 MD033: false
+MD045: false
 MD046: false
+MD051: false
 MD052: false
+MD053: false
 MD059: false
@@ -38,7 +38,7 @@ repos:
  rev: 0.9.1
  hooks:
    - id: pip-compile
-      args: [requirements/test.in, -o, requirements/test.txt, --index-strategy, unsafe-best-match, --torch-backend, cu129, --python-platform, x86_64-manylinux_2_28, --python-version, "3.12"]
+      args: [requirements/test.in, -o, requirements/test.txt, --index-strategy, unsafe-best-match, --torch-backend, cu129, --python-platform, x86_64-manylinux_2_28]
      files: ^requirements/test\.(in|txt)$
 - repo: local
  hooks:
@@ -39,13 +39,6 @@ set(PYTHON_SUPPORTED_VERSIONS "3.10" "3.11" "3.12" "3.13")
 # Supported AMD GPU architectures.
 set(HIP_SUPPORTED_ARCHS "gfx906;gfx908;gfx90a;gfx942;gfx950;gfx1030;gfx1100;gfx1101;gfx1200;gfx1201;gfx1150;gfx1151")

-# ROCm installation prefix. Default to /opt/rocm but allow override via
-# -DROCM_PATH=/your/rocm/path when invoking cmake.
-if(NOT DEFINED ROCM_PATH)
-  set(ROCM_PATH "/opt/rocm" CACHE PATH "ROCm installation prefix")
-else()
-  set(ROCM_PATH ${ROCM_PATH} CACHE PATH "ROCm installation prefix" FORCE)
-endif()
 #
 # Supported/expected torch versions for CUDA/ROCm.
 #
@@ -56,8 +49,8 @@ endif()
 # requirements.txt files and should be kept consistent.  The ROCm torch
 # versions are derived from docker/Dockerfile.rocm
 #
-set(TORCH_SUPPORTED_VERSION_CUDA "2.9.1")
-set(TORCH_SUPPORTED_VERSION_ROCM "2.9.1")
+set(TORCH_SUPPORTED_VERSION_CUDA "2.9.0")
+set(TORCH_SUPPORTED_VERSION_ROCM "2.9.0")

 #
 # Try to find python package with an executable that exactly matches
@@ -136,7 +129,7 @@ elseif(HIP_FOUND)

  # ROCm 5.X and 6.X
  if (ROCM_VERSION_DEV_MAJOR GREATER_EQUAL 5 AND
-      Torch_VERSION VERSION_LESS ${TORCH_SUPPORTED_VERSION_ROCM})
+      NOT Torch_VERSION VERSION_EQUAL ${TORCH_SUPPORTED_VERSION_ROCM})
    message(WARNING "Pytorch version >= ${TORCH_SUPPORTED_VERSION_ROCM} "
      "expected for ROCm build, saw ${Torch_VERSION} instead.")
  endif()
@@ -244,28 +237,11 @@ set_gencode_flags_for_srcs(
  SRCS "${VLLM_CUMEM_EXT_SRC}"
  CUDA_ARCHS "${CUDA_ARCHS}")

-if(VLLM_GPU_LANG STREQUAL "CUDA" OR VLLM_GPU_LANG STREQUAL "HIP")
+if(VLLM_GPU_LANG STREQUAL "CUDA")
  message(STATUS "Enabling cumem allocator extension.")
-  if(VLLM_GPU_LANG STREQUAL "CUDA")
-    # link against cuda driver library
-    list(APPEND CUMEM_LIBS CUDA::cuda_driver)
-  else()
-    # link against rocm driver library. Prefer an absolute path to
-    # libamdhip64.so inside ${ROCM_PATH}/lib if available, otherwise fall
-    # back to linking by name "amdhip64".
-    find_library(AMDHIP64_LIB
-      NAMES amdhip64 libamdhip64.so
-      PATHS ${ROCM_PATH}/lib
-      NO_DEFAULT_PATH)
-    if(AMDHIP64_LIB)
-      message(STATUS "Found libamdhip64 at ${AMDHIP64_LIB}")
-      list(APPEND CUMEM_LIBS ${AMDHIP64_LIB})
-    else()
-      message(WARNING "libamdhip64 not found in ${ROCM_PATH}/lib; falling back to linking 'amdhip64' by name")
-      list(APPEND CUMEM_LIBS amdhip64)
-    endif()
-  endif()
-  define_extension_target(
+  # link against cuda driver library
+  list(APPEND CUMEM_LIBS CUDA::cuda_driver)
+  define_gpu_extension_target(
    cumem_allocator
    DESTINATION vllm
    LANGUAGE CXX
@@ -289,7 +265,6 @@ set(VLLM_EXT_SRC
  "csrc/pos_encoding_kernels.cu"
  "csrc/activation_kernels.cu"
  "csrc/layernorm_kernels.cu"
-  "csrc/fused_qknorm_rope_kernel.cu"
  "csrc/layernorm_quant_kernels.cu"
  "csrc/sampler.cu"
  "csrc/cuda_view.cu"
@@ -307,7 +282,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
  SET(CUTLASS_ENABLE_HEADERS_ONLY ON CACHE BOOL "Enable only the header library")

  # Set CUTLASS_REVISION. Used for FetchContent. Also fixes some bogus messages when building.
-  set(CUTLASS_REVISION "v4.2.1")
+  set(CUTLASS_REVISION "v4.2.1" CACHE STRING "CUTLASS revision to use")

  # Use the specified CUTLASS source directory for compilation if VLLM_CUTLASS_SRC_DIR is provided
  if (DEFINED ENV{VLLM_CUTLASS_SRC_DIR})
@@ -354,22 +329,9 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
  # Only build Marlin kernels if we are building for at least some compatible archs.
  # Keep building Marlin for 9.0 as there are some group sizes and shapes that
  # are not supported by Machete yet.
-
-  # marlin arches for fp16 output
-  cuda_archs_loose_intersection(MARLIN_ARCHS "8.0+PTX" "${CUDA_ARCHS}")
-  # marlin has limited support for turing
-  cuda_archs_loose_intersection(MARLIN_SM75_ARCHS "7.5" "${CUDA_ARCHS}")
-  # marlin arches for bf16 output (we need 9.0 for bf16 atomicAdd PTX)
-  cuda_archs_loose_intersection(MARLIN_BF16_ARCHS "8.0+PTX;9.0+PTX" "${CUDA_ARCHS}")
-  # marlin arches for fp8 input
-  # - sm80 doesn't support fp8 computation
-  # - sm90 and sm100 don't support QMMA.16832.F32.E4M3.E4M3 SAAS instruction
-  # so we only enable fp8 computation for SM89 (e.g. RTX 40x0)  and 12.0 (e.g. RTX 50x0)
-  cuda_archs_loose_intersection(MARLIN_FP8_ARCHS "8.9;12.0" "${CUDA_ARCHS}")
-  # marlin arches for other files
-  cuda_archs_loose_intersection(MARLIN_OTHER_ARCHS "7.5;8.0+PTX" "${CUDA_ARCHS}")
-
-  if (MARLIN_OTHER_ARCHS)
+  # 9.0 for latest bf16 atomicAdd PTX
+  cuda_archs_loose_intersection(MARLIN_ARCHS "8.0;8.7;9.0+PTX" "${CUDA_ARCHS}")
+  if (MARLIN_ARCHS)

    #
    # For the Marlin kernels we automatically generate sources for various
@@ -378,18 +340,16 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
    set(MARLIN_GEN_SCRIPT
      ${CMAKE_CURRENT_SOURCE_DIR}/csrc/quantization/gptq_marlin/generate_kernels.py)
    file(MD5 ${MARLIN_GEN_SCRIPT} MARLIN_GEN_SCRIPT_HASH)
-    list(JOIN CUDA_ARCHS "," CUDA_ARCHS_STR)
-    set(MARLIN_GEN_SCRIPT_HASH_AND_ARCH "${MARLIN_GEN_SCRIPT_HASH}(ARCH:${CUDA_ARCHS_STR})")

-    message(STATUS "Marlin generation script hash: ${MARLIN_GEN_SCRIPT_HASH_AND_ARCH}")
-    message(STATUS "Last run Marlin generate script hash: $CACHE{MARLIN_GEN_SCRIPT_HASH_AND_ARCH}")
+    message(STATUS "Marlin generation script hash: ${MARLIN_GEN_SCRIPT_HASH}")
+    message(STATUS "Last run Marlin generate script hash: $CACHE{MARLIN_GEN_SCRIPT_HASH}")

-    if (NOT DEFINED CACHE{MARLIN_GEN_SCRIPT_HASH_AND_ARCH}
-        OR NOT $CACHE{MARLIN_GEN_SCRIPT_HASH_AND_ARCH} STREQUAL ${MARLIN_GEN_SCRIPT_HASH_AND_ARCH})
+    if (NOT DEFINED CACHE{MARLIN_GEN_SCRIPT_HASH}
+        OR NOT $CACHE{MARLIN_GEN_SCRIPT_HASH} STREQUAL ${MARLIN_GEN_SCRIPT_HASH})
      execute_process(
        COMMAND ${CMAKE_COMMAND} -E env
-        PYTHONPATH=$ENV{PYTHONPATH}
-          ${Python_EXECUTABLE} ${MARLIN_GEN_SCRIPT} ${CUDA_ARCHS_STR}
+        PYTHONPATH=$PYTHONPATH
+          ${Python_EXECUTABLE} ${MARLIN_GEN_SCRIPT}
        RESULT_VARIABLE marlin_generation_result
        OUTPUT_VARIABLE marlin_generation_result
        OUTPUT_FILE ${CMAKE_CURRENT_BINARY_DIR}/marlin_generation.log
@@ -402,76 +362,40 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
                            "\nCheck the log for details: "
                            "${CMAKE_CURRENT_BINARY_DIR}/marlin_generation.log")
      else()
-        set(MARLIN_GEN_SCRIPT_HASH_AND_ARCH ${MARLIN_GEN_SCRIPT_HASH_AND_ARCH}
-            CACHE STRING "Last run Marlin generate script hash and arch" FORCE)
+        set(MARLIN_GEN_SCRIPT_HASH ${MARLIN_GEN_SCRIPT_HASH}
+            CACHE STRING "Last run Marlin generate script hash" FORCE)
        message(STATUS "Marlin generation completed successfully.")
      endif()
    else()
      message(STATUS "Marlin generation script has not changed, skipping generation.")
    endif()

-    if (MARLIN_ARCHS)
-      file(GLOB MARLIN_TEMPLATE_KERNEL_SRC "csrc/quantization/gptq_marlin/sm80_kernel_*_float16.cu")
-      set_gencode_flags_for_srcs(
-        SRCS "${MARLIN_TEMPLATE_KERNEL_SRC}"
-        CUDA_ARCHS "${MARLIN_ARCHS}")
-      if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8)
-        set_source_files_properties(${MARLIN_TEMPLATE_KERNEL_SRC}
-          PROPERTIES COMPILE_FLAGS "-static-global-template-stub=false")
-      endif()
-      list(APPEND VLLM_EXT_SRC ${MARLIN_TEMPLATE_KERNEL_SRC})
-
-      file(GLOB MARLIN_TEMPLATE_BF16_KERNEL_SRC "csrc/quantization/gptq_marlin/sm80_kernel_*_bfloat16.cu")
-      set_gencode_flags_for_srcs(
-        SRCS "${MARLIN_TEMPLATE_BF16_KERNEL_SRC}"
-        CUDA_ARCHS "${MARLIN_BF16_ARCHS}")
-      if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8)
-        set_source_files_properties(${MARLIN_TEMPLATE_BF16_KERNEL_SRC}
-          PROPERTIES COMPILE_FLAGS "-static-global-template-stub=false")
-      endif()
-      list(APPEND VLLM_EXT_SRC ${MARLIN_TEMPLATE_BF16_KERNEL_SRC})
+    file(GLOB MARLIN_TEMPLATE_KERNEL_SRC "csrc/quantization/gptq_marlin/kernel_*.cu")
+    set_gencode_flags_for_srcs(
+      SRCS "${MARLIN_TEMPLATE_KERNEL_SRC}"
+      CUDA_ARCHS "${MARLIN_ARCHS}")
+    if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8)
+      set_source_files_properties(${MARLIN_TEMPLATE_KERNEL_SRC}
+        PROPERTIES COMPILE_FLAGS "-static-global-template-stub=false")
    endif()

-    if (MARLIN_SM75_ARCHS) 
-      file(GLOB MARLIN_TEMPLATE_SM75_KERNEL_SRC "csrc/quantization/gptq_marlin/sm75_kernel_*.cu")
-      set_gencode_flags_for_srcs(
-        SRCS "${MARLIN_TEMPLATE_SM75_KERNEL_SRC}"
-        CUDA_ARCHS "${MARLIN_SM75_ARCHS}")
-      if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8)
-        set_source_files_properties(${MARLIN_TEMPLATE_SM75_KERNEL_SRC}
-          PROPERTIES COMPILE_FLAGS "-static-global-template-stub=false")
-      endif()
-      list(APPEND VLLM_EXT_SRC ${MARLIN_TEMPLATE_SM75_KERNEL_SRC})
-    endif()
-
-    if (MARLIN_FP8_ARCHS) 
-      file(GLOB MARLIN_TEMPLATE_FP8_KERNEL_SRC "csrc/quantization/gptq_marlin/sm89_kernel_*.cu")
-      set_gencode_flags_for_srcs(
-        SRCS "${MARLIN_TEMPLATE_FP8_KERNEL_SRC}"
-        CUDA_ARCHS "${MARLIN_FP8_ARCHS}")
-      if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8)
-        set_source_files_properties(${MARLIN_TEMPLATE_FP8_KERNEL_SRC}
-          PROPERTIES COMPILE_FLAGS "-static-global-template-stub=false")
-      endif()
-      list(APPEND VLLM_EXT_SRC ${MARLIN_TEMPLATE_FP8_KERNEL_SRC})
-    endif()
+    list(APPEND VLLM_EXT_SRC ${MARLIN_TEMPLATE_KERNEL_SRC})

    set(MARLIN_SRCS
       "csrc/quantization/marlin/sparse/marlin_24_cuda_kernel.cu"
       "csrc/quantization/gptq_marlin/gptq_marlin.cu"
-       "csrc/quantization/gptq_marlin/marlin_int4_fp8_preprocess.cu"
       "csrc/quantization/gptq_marlin/gptq_marlin_repack.cu"
       "csrc/quantization/gptq_marlin/awq_marlin_repack.cu")
    set_gencode_flags_for_srcs(
      SRCS "${MARLIN_SRCS}"
-      CUDA_ARCHS "${MARLIN_OTHER_ARCHS}")
+      CUDA_ARCHS "${MARLIN_ARCHS}")
    if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8)
-      set_source_files_properties(${MARLIN_SRCS}
+      set_source_files_properties("csrc/quantization/gptq_marlin/gptq_marlin.cu"
        PROPERTIES COMPILE_FLAGS "-static-global-template-stub=false")
    endif()
    list(APPEND VLLM_EXT_SRC "${MARLIN_SRCS}")

-    message(STATUS "Building Marlin kernels for archs: ${MARLIN_OTHER_ARCHS}")
+    message(STATUS "Building Marlin kernels for archs: ${MARLIN_ARCHS}")
  else()
    message(STATUS "Not building Marlin kernels as no compatible archs found"
                   " in CUDA target architectures")
@@ -563,9 +487,9 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
  # The cutlass_scaled_mm kernels for Blackwell SM100 (c3x, i.e. CUTLASS 3.x)
  # require CUDA 12.8 or later
  if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 13.0)
-    cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0f;11.0f" "${CUDA_ARCHS}")
+    cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0f;11.0f;12.0f" "${CUDA_ARCHS}")
  else()
-    cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0a;10.1a;10.3a" "${CUDA_ARCHS}")
+    cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0a;10.1a;10.3a;12.0a;12.1a" "${CUDA_ARCHS}")
  endif()
  if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8 AND SCALED_MM_ARCHS)
    set(SRCS
@@ -655,15 +579,12 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
    set(SRCS
      "csrc/quantization/fp4/nvfp4_quant_kernels.cu"
      "csrc/quantization/fp4/activation_nvfp4_quant_fusion_kernels.cu"
-      "csrc/quantization/fp4/nvfp4_experts_quant.cu"
-      "csrc/quantization/fp4/nvfp4_scaled_mm_sm120_kernels.cu"
-      "csrc/quantization/fp4/nvfp4_blockwise_moe_kernel.cu")
+      "csrc/quantization/fp4/nvfp4_scaled_mm_sm120_kernels.cu")
    set_gencode_flags_for_srcs(
      SRCS "${SRCS}"
      CUDA_ARCHS "${FP4_ARCHS}")
    list(APPEND VLLM_EXT_SRC "${SRCS}")
    list(APPEND VLLM_GPU_FLAGS "-DENABLE_NVFP4_SM120=1")
-    list(APPEND VLLM_GPU_FLAGS "-DENABLE_CUTLASS_MOE_SM120=1")
    message(STATUS "Building NVFP4 for archs: ${FP4_ARCHS}")
  else()
    message(STATUS "Not building NVFP4 as no compatible archs were found.")
@@ -673,9 +594,9 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")

  # FP4 Archs and flags
  if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 13.0)
-    cuda_archs_loose_intersection(FP4_ARCHS "10.0f;11.0f" "${CUDA_ARCHS}")
+    cuda_archs_loose_intersection(FP4_ARCHS "10.0f;11.0f;12.0f" "${CUDA_ARCHS}")
  else()
-    cuda_archs_loose_intersection(FP4_ARCHS "10.0a;10.1a;10.3a" "${CUDA_ARCHS}")
+    cuda_archs_loose_intersection(FP4_ARCHS "10.0a;10.1a;12.0a;12.1a" "${CUDA_ARCHS}")
  endif()
  if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8 AND FP4_ARCHS)
    set(SRCS
@@ -749,7 +670,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
  if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 13.0)
    cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0f;11.0f" "${CUDA_ARCHS}")
  else()
-    cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0a;10.1a;10.3a" "${CUDA_ARCHS}")
+    cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0a" "${CUDA_ARCHS}")
  endif()
  if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8 AND SCALED_MM_ARCHS)
    set(SRCS "csrc/quantization/w8a8/cutlass/moe/grouped_mm_c3x_sm100.cu")
@@ -795,9 +716,27 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
  endif()

  if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 13.0)
-    cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0f;11.0f" "${CUDA_ARCHS}")
+    cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0f;11.0f;12.0f" "${CUDA_ARCHS}")
  else()
-    cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0a;10.1a;10.3a" "${CUDA_ARCHS}")
+    cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0a;10.1a;10.3a;12.0a;12.1a" "${CUDA_ARCHS}")
+  endif()
+  if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8 AND SCALED_MM_ARCHS)
+    set(SRCS "csrc/quantization/w8a8/cutlass/moe/blockwise_scaled_group_mm_sm100.cu")
+    set_gencode_flags_for_srcs(
+      SRCS "${SRCS}"
+      CUDA_ARCHS "${SCALED_MM_ARCHS}")
+    list(APPEND VLLM_EXT_SRC "${SRCS}")
+    list(APPEND VLLM_GPU_FLAGS "-DENABLE_CUTLASS_MOE_SM100=1")
+    message(STATUS "Building blockwise_scaled_group_mm_sm100 for archs: ${SCALED_MM_ARCHS}")
+  else()
+    if (NOT ${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8 AND SCALED_MM_ARCHS)
+      message(STATUS "Not building blockwise_scaled_group_mm_sm100 kernels as CUDA Compiler version is "
+                     "not >= 12.8, we recommend upgrading to CUDA 12.8 or later "
+                     "if you intend on running FP8 quantized MoE models on Blackwell.")
+    else()
+      message(STATUS "Not building blockwise_scaled_group_mm_sm100 as no compatible archs found "
+                     "in CUDA target architectures")
+    endif()
  endif()

  #
@@ -822,7 +761,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
        OR NOT $CACHE{MACHETE_GEN_SCRIPT_HASH} STREQUAL ${MACHETE_GEN_SCRIPT_HASH})
      execute_process(
        COMMAND ${CMAKE_COMMAND} -E env
-        PYTHONPATH=${CMAKE_CURRENT_SOURCE_DIR}/csrc/cutlass_extensions/:${CUTLASS_DIR}/python/:${VLLM_PYTHON_PATH}:$ENV{PYTHONPATH}
+        PYTHONPATH=${CMAKE_CURRENT_SOURCE_DIR}/csrc/cutlass_extensions/:${CUTLASS_DIR}/python/:${VLLM_PYTHON_PATH}:$PYTHONPATH
          ${Python_EXECUTABLE} ${MACHETE_GEN_SCRIPT}
        RESULT_VARIABLE machete_generation_result
        OUTPUT_VARIABLE machete_generation_output
@@ -874,10 +813,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
  cuda_archs_loose_intersection(W4A8_ARCHS "9.0a" "${CUDA_ARCHS}")
  if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.0 AND W4A8_ARCHS)
    set(SRCS
-       "csrc/quantization/cutlass_w4a8/w4a8_mm_entry.cu"
-       "csrc/quantization/cutlass_w4a8/w4a8_grouped_mm_entry.cu"
-       "csrc/quantization/cutlass_w4a8/w4a8_utils.cu"
-       )
+       "csrc/quantization/cutlass_w4a8/w4a8_mm_entry.cu")

    set_gencode_flags_for_srcs(
      SRCS "${SRCS}"
@@ -900,7 +836,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
  endif()

  # Hadacore kernels
-  cuda_archs_loose_intersection(HADACORE_ARCHS "8.0+PTX;9.0+PTX" "${CUDA_ARCHS}")
+  cuda_archs_loose_intersection(HADACORE_ARCHS "8.0;8.9;9.0" "${CUDA_ARCHS}")
  if(HADACORE_ARCHS)
    set(SRCS "csrc/quantization/hadamard/hadacore/hadamard_transform_cuda.cu")
    set_gencode_flags_for_srcs(
@@ -922,7 +858,7 @@ if (VLLM_GPU_LANG STREQUAL "HIP")
 endif()

 message(STATUS "Enabling C extension.")
-define_extension_target(
+define_gpu_extension_target(
  _C
  DESTINATION vllm
  LANGUAGE ${VLLM_GPU_LANG}
@@ -947,6 +883,7 @@ target_compile_definitions(_C PRIVATE CUTLASS_ENABLE_DIRECT_CUDA_DRIVER_CALL=1)
 set(VLLM_MOE_EXT_SRC
  "csrc/moe/torch_bindings.cpp"
  "csrc/moe/moe_align_sum_kernels.cu"
+  "csrc/moe/moe_lora_align_sum_kernels.cu"
  "csrc/moe/topk_softmax_kernels.cu")

 if(VLLM_GPU_LANG STREQUAL "CUDA")
@@ -976,20 +913,9 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
    CUDA_ARCHS "${CUDA_ARCHS}")

  list(APPEND VLLM_MOE_EXT_SRC "${VLLM_MOE_WNA16_SRC}")
-  # moe marlin arches
-  # note that we always set `use_atomic_add=False` for moe marlin now,
-  # so we don't need 9.0 for bf16 atomicAdd PTX
-  cuda_archs_loose_intersection(MARLIN_MOE_ARCHS "8.0+PTX" "${CUDA_ARCHS}")
-  # moe marlin has limited support for turing
-  cuda_archs_loose_intersection(MARLIN_MOE_SM75_ARCHS "7.5" "${CUDA_ARCHS}")
-  # moe marlin arches for fp8 input
-  # - sm80 doesn't support fp8 computation
-  # - sm90 and sm100 don't support QMMA.16832.F32.E4M3.E4M3 SAAS instruction
-  # so we only enable fp8 computation for SM89 (e.g. RTX 40x0)  and 12.0 (e.g. RTX 50x0)
-  cuda_archs_loose_intersection(MARLIN_MOE_FP8_ARCHS "8.9;12.0" "${CUDA_ARCHS}")
-  # moe marlin arches for other files
-  cuda_archs_loose_intersection(MARLIN_MOE_OTHER_ARCHS "7.5;8.0+PTX" "${CUDA_ARCHS}")
-  if (MARLIN_MOE_OTHER_ARCHS)
+  # 9.0 for latest bf16 atomicAdd PTX
+  cuda_archs_loose_intersection(MARLIN_MOE_ARCHS "8.0;8.7;9.0+PTX" "${CUDA_ARCHS}")
+  if (MARLIN_MOE_ARCHS)

    #
    # For the Marlin MOE kernels we automatically generate sources for various
@@ -998,18 +924,16 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
    set(MOE_MARLIN_GEN_SCRIPT
      ${CMAKE_CURRENT_SOURCE_DIR}/csrc/moe/marlin_moe_wna16/generate_kernels.py)
    file(MD5 ${MOE_MARLIN_GEN_SCRIPT} MOE_MARLIN_GEN_SCRIPT_HASH)
-    list(JOIN CUDA_ARCHS "," CUDA_ARCHS_STR)
-    set(MOE_MARLIN_GEN_SCRIPT_HASH_AND_ARCH "${MOE_MARLIN_GEN_SCRIPT_HASH}(ARCH:${CUDA_ARCHS_STR})")

-    message(STATUS "Marlin MOE generation script hash with arch: ${MOE_MARLIN_GEN_SCRIPT_HASH_AND_ARCH}")
-    message(STATUS "Last run Marlin MOE generate script hash with arch: $CACHE{MOE_MARLIN_GEN_SCRIPT_HASH_AND_ARCH}")
+    message(STATUS "Marlin MOE generation script hash: ${MOE_MARLIN_GEN_SCRIPT_HASH}")
+    message(STATUS "Last run Marlin MOE generate script hash: $CACHE{MOE_MARLIN_GEN_SCRIPT_HASH}")

-    if (NOT DEFINED CACHE{MOE_MARLIN_GEN_SCRIPT_HASH_AND_ARCH}
-        OR NOT $CACHE{MOE_MARLIN_GEN_SCRIPT_HASH_AND_ARCH} STREQUAL ${MOE_MARLIN_GEN_SCRIPT_HASH_AND_ARCH})
+    if (NOT DEFINED CACHE{MOE_MARLIN_GEN_SCRIPT_HASH}
+        OR NOT $CACHE{MOE_MARLIN_GEN_SCRIPT_HASH} STREQUAL ${MOE_MARLIN_GEN_SCRIPT_HASH})
      execute_process(
        COMMAND ${CMAKE_COMMAND} -E env
-        PYTHONPATH=$ENV{PYTHONPATH}
-          ${Python_EXECUTABLE} ${MOE_MARLIN_GEN_SCRIPT} ${CUDA_ARCHS_STR}
+        PYTHONPATH=$PYTHONPATH
+          ${Python_EXECUTABLE} ${MOE_MARLIN_GEN_SCRIPT}
        RESULT_VARIABLE moe_marlin_generation_result
        OUTPUT_VARIABLE moe_marlin_generation_output
        OUTPUT_FILE ${CMAKE_CURRENT_BINARY_DIR}/moe_marlin_generation.log
@@ -1022,7 +946,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
                            "\nCheck the log for details: "
                            "${CMAKE_CURRENT_BINARY_DIR}/moe_marlin_generation.log")
      else()
-        set(MOE_MARLIN_GEN_SCRIPT_HASH_AND_ARCH ${MOE_MARLIN_GEN_SCRIPT_HASH_AND_ARCH}
+        set(MOE_MARLIN_GEN_SCRIPT_HASH ${MOE_MARLIN_GEN_SCRIPT_HASH}
            CACHE STRING "Last run Marlin MOE generate script hash" FORCE)
        message(STATUS "Marlin MOE generation completed successfully.")
      endif()
@@ -1030,53 +954,18 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
      message(STATUS "Marlin MOE generation script has not changed, skipping generation.")
    endif()

-    if (MARLIN_MOE_ARCHS)
-      file(GLOB MARLIN_MOE_SRC "csrc/moe/marlin_moe_wna16/sm80_kernel_*.cu")
-      set_gencode_flags_for_srcs(
-        SRCS "${MARLIN_MOE_SRC}"
-        CUDA_ARCHS "${MARLIN_MOE_ARCHS}")
-      if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8)
-        set_source_files_properties(${MARLIN_MOE_SRC}
-          PROPERTIES COMPILE_FLAGS "-static-global-template-stub=false")
-      endif()
-      list(APPEND VLLM_MOE_EXT_SRC ${MARLIN_MOE_SRC})
-    endif()
-
-    if (MARLIN_MOE_SM75_ARCHS) 
-      file(GLOB MARLIN_MOE_SM75_SRC "csrc/moe/marlin_moe_wna16/sm75_kernel_*.cu")
-      set_gencode_flags_for_srcs(
-        SRCS "${MARLIN_MOE_SM75_SRC}"
-        CUDA_ARCHS "${MARLIN_MOE_SM75_ARCHS}")
-      if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8)
-        set_source_files_properties(${MARLIN_MOE_SM75_SRC}
-          PROPERTIES COMPILE_FLAGS "-static-global-template-stub=false")
-      endif()
-      list(APPEND VLLM_MOE_EXT_SRC ${MARLIN_MOE_SM75_SRC})
-    endif()
-
-    if (MARLIN_MOE_FP8_ARCHS)
-      file(GLOB MARLIN_MOE_FP8_SRC "csrc/moe/marlin_moe_wna16/sm89_kernel_*.cu")
-      set_gencode_flags_for_srcs(
-        SRCS "${MARLIN_MOE_FP8_SRC}"
-        CUDA_ARCHS "${MARLIN_MOE_FP8_ARCHS}")
-      if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8)
-        set_source_files_properties(${MARLIN_MOE_FP8_SRC}
-          PROPERTIES COMPILE_FLAGS "-static-global-template-stub=false")
-      endif()
-      list(APPEND VLLM_MOE_EXT_SRC ${MARLIN_MOE_FP8_SRC})
-    endif()
-
-    set(MARLIN_MOE_OTHER_SRC "csrc/moe/marlin_moe_wna16/ops.cu")
+    file(GLOB MOE_WNAA16_MARLIN_SRC "csrc/moe/marlin_moe_wna16/*.cu")
    set_gencode_flags_for_srcs(
-      SRCS "${MARLIN_MOE_OTHER_SRC}"
-      CUDA_ARCHS "${MARLIN_MOE_OTHER_ARCHS}")
+      SRCS "${MOE_WNAA16_MARLIN_SRC}"
+      CUDA_ARCHS "${MARLIN_MOE_ARCHS}")
    if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8)
-      set_source_files_properties(${MARLIN_MOE_OTHER_SRC}
+      set_source_files_properties(${MOE_WNAA16_MARLIN_SRC}
        PROPERTIES COMPILE_FLAGS "-static-global-template-stub=false")
    endif()
-    list(APPEND VLLM_MOE_EXT_SRC "${MARLIN_MOE_OTHER_SRC}")

-    message(STATUS "Building Marlin MOE kernels for archs: ${MARLIN_MOE_OTHER_ARCHS}")
+    list(APPEND VLLM_MOE_EXT_SRC ${MOE_WNAA16_MARLIN_SRC})
+
+    message(STATUS "Building Marlin MOE kernels for archs: ${MARLIN_MOE_ARCHS}")
  else()
    message(STATUS "Not building Marlin MOE kernels as no compatible archs found"
                   " in CUDA target architectures")
@@ -1084,7 +973,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
 endif()

 message(STATUS "Enabling moe extension.")
-define_extension_target(
+define_gpu_extension_target(
  _moe_C
  DESTINATION vllm
  LANGUAGE ${VLLM_GPU_LANG}
@@ -1105,7 +994,7 @@ if(VLLM_GPU_LANG STREQUAL "HIP")
    "csrc/rocm/skinny_gemms.cu"
    "csrc/rocm/attention.cu")

-  define_extension_target(
+  define_gpu_extension_target(
    _rocm_C
    DESTINATION vllm
    LANGUAGE ${VLLM_GPU_LANG}
@@ -1116,11 +1005,6 @@ if(VLLM_GPU_LANG STREQUAL "HIP")
    WITH_SOABI)
 endif()

-# For CUDA and HIP builds also build the triton_kernels external package.
-if(VLLM_GPU_LANG STREQUAL "CUDA" OR VLLM_GPU_LANG STREQUAL "HIP")
-    include(cmake/external_projects/triton_kernels.cmake)
-endif()
-
 # For CUDA we also build and ship some external projects.
 if (VLLM_GPU_LANG STREQUAL "CUDA")
    include(cmake/external_projects/flashmla.cmake)
@@ -14,8 +14,48 @@ Easy, fast, and cheap LLM serving for everyone
 | <a href="https://docs.vllm.ai"><b>Documentation</b></a> | <a href="https://blog.vllm.ai/"><b>Blog</b></a> | <a href="https://arxiv.org/abs/2309.06180"><b>Paper</b></a> | <a href="https://x.com/vllm_project"><b>Twitter/X</b></a> | <a href="https://discuss.vllm.ai"><b>User Forum</b></a> | <a href="https://slack.vllm.ai"><b>Developer Slack</b></a> |
 </p>

-🔥 We have built a vllm website to help you get started with vllm. Please visit [vllm.ai](https://vllm.ai) to learn more.
-For events, please visit [vllm.ai/events](https://vllm.ai/events) to join us.
+---
+Join us at the [PyTorch Conference, October 22-23](https://events.linuxfoundation.org/pytorch-conference/) and [Ray Summit, November 3-5](https://www.anyscale.com/ray-summit/2025) in San Francisco for our latest updates on vLLM and to meet the vLLM team! Register now for the largest vLLM community events of the year!
+
+---
+
+*Latest News* 🔥
+
+- [2025/10] We hosted [vLLM Shanghai Meetup](https://mp.weixin.qq.com/s/__xb4OyOsImz-9eAVrdlcg) focused on hands-on vLLM inference optimization! Please find the meetup slides [here](https://drive.google.com/drive/folders/1KqwjsFJLfEsC8wlDugnrR61zsWHt94Q6).
+- [2025/09] We hosted [vLLM Toronto Meetup](https://luma.com/e80e0ymm) focused on tackling inference at scale and speculative decoding with speakers from NVIDIA and Red Hat! Please find the meetup slides [here](https://docs.google.com/presentation/d/1IYJYmJcu9fLpID5N5RbW_vO0XLo0CGOR14IXOjB61V8/edit?usp=sharing).
+- [2025/08] We hosted [vLLM Shenzhen Meetup](https://mp.weixin.qq.com/s/k8ZBO1u2_2odgiKWH_GVTQ) focusing on the ecosystem around vLLM! Please find the meetup slides [here](https://drive.google.com/drive/folders/1Ua2SVKVSu-wp5vou_6ElraDt2bnKhiEA).
+- [2025/08] We hosted [vLLM Singapore Meetup](https://www.sginnovate.com/event/vllm-sg-meet). We shared V1 updates, disaggregated serving and MLLM speedups with speakers from Embedded LLM, AMD, WekaIO, and A*STAR. Please find the meetup slides [here](https://drive.google.com/drive/folders/1ncf3GyqLdqFaB6IeB834E5TZJPLAOiXZ?usp=sharing).
+- [2025/08] We hosted [vLLM Shanghai Meetup](https://mp.weixin.qq.com/s/pDmAXHcN7Iqc8sUKgJgGtg) focusing on building, developing, and integrating with vLLM! Please find the meetup slides [here](https://drive.google.com/drive/folders/1OvLx39wnCGy_WKq8SiVKf7YcxxYI3WCH).
+- [2025/05] vLLM is now a hosted project under PyTorch Foundation! Please find the announcement [here](https://pytorch.org/blog/pytorch-foundation-welcomes-vllm/).
+- [2025/01] We are excited to announce the alpha release of vLLM V1: A major architectural upgrade with 1.7x speedup! Clean code, optimized execution loop, zero-overhead prefix caching, enhanced multimodal support, and more. Please check out our blog post [here](https://blog.vllm.ai/2025/01/27/v1-alpha-release.html).
+
+<details>
+<summary>Previous News</summary>
+
+- [2025/08] We hosted [vLLM Korea Meetup](https://luma.com/cgcgprmh) with Red Hat and Rebellions! We shared the latest advancements in vLLM along with project spotlights from the vLLM Korea community. Please find the meetup slides [here](https://drive.google.com/file/d/1bcrrAE1rxUgx0mjIeOWT6hNe2RefC5Hm/view).
+- [2025/08] We hosted [vLLM Beijing Meetup](https://mp.weixin.qq.com/s/dgkWg1WFpWGO2jCdTqQHxA) focusing on large-scale LLM deployment! Please find the meetup slides [here](https://drive.google.com/drive/folders/1Pid6NSFLU43DZRi0EaTcPgXsAzDvbBqF) and the recording [here](https://www.chaspark.com/#/live/1166916873711665152).
+- [2025/05] We hosted [NYC vLLM Meetup](https://lu.ma/c1rqyf1f)! Please find the meetup slides [here](https://docs.google.com/presentation/d/1_q_aW_ioMJWUImf1s1YM-ZhjXz8cUeL0IJvaquOYBeA/edit?usp=sharing).
+- [2025/04] We hosted [Asia Developer Day](https://www.sginnovate.com/event/limited-availability-morning-evening-slots-remaining-inaugural-vllm-asia-developer-day)! Please find the meetup slides from the vLLM team [here](https://docs.google.com/presentation/d/19cp6Qu8u48ihB91A064XfaXruNYiBOUKrBxAmDOllOo/edit?usp=sharing).
+- [2025/03] We hosted [vLLM x Ollama Inference Night](https://lu.ma/vllm-ollama)! Please find the meetup slides from the vLLM team [here](https://docs.google.com/presentation/d/16T2PDD1YwRnZ4Tu8Q5r6n53c5Lr5c73UV9Vd2_eBo4U/edit?usp=sharing).
+- [2025/03] We hosted [the first vLLM China Meetup](https://mp.weixin.qq.com/s/n77GibL2corAtQHtVEAzfg)! Please find the meetup slides from vLLM team [here](https://docs.google.com/presentation/d/1REHvfQMKGnvz6p3Fd23HhSO4c8j5WPGZV0bKYLwnHyQ/edit?usp=sharing).
+- [2025/03] We hosted [the East Coast vLLM Meetup](https://lu.ma/7mu4k4xx)! Please find the meetup slides [here](https://docs.google.com/presentation/d/1NHiv8EUFF1NLd3fEYODm56nDmL26lEeXCaDgyDlTsRs/edit#slide=id.g31441846c39_0_0).
+- [2025/02] We hosted [the ninth vLLM meetup](https://lu.ma/h7g3kuj9) with Meta! Please find the meetup slides from vLLM team [here](https://docs.google.com/presentation/d/1jzC_PZVXrVNSFVCW-V4cFXb6pn7zZ2CyP_Flwo05aqg/edit?usp=sharing) and AMD [here](https://drive.google.com/file/d/1Zk5qEJIkTmlQ2eQcXQZlljAx3m9s7nwn/view?usp=sharing). The slides from Meta will not be posted.
+- [2025/01] We hosted [the eighth vLLM meetup](https://lu.ma/zep56hui) with Google Cloud! Please find the meetup slides from vLLM team [here](https://docs.google.com/presentation/d/1epVkt4Zu8Jz_S5OhEHPc798emsYh2BwYfRuDDVEF7u4/edit?usp=sharing), and Google Cloud team [here](https://drive.google.com/file/d/1h24pHewANyRL11xy5dXUbvRC9F9Kkjix/view?usp=sharing).
+- [2024/12] vLLM joins [pytorch ecosystem](https://pytorch.org/blog/vllm-joins-pytorch)! Easy, Fast, and Cheap LLM Serving for Everyone!
+- [2024/11] We hosted [the seventh vLLM meetup](https://lu.ma/h0qvrajz) with Snowflake! Please find the meetup slides from vLLM team [here](https://docs.google.com/presentation/d/1e3CxQBV3JsfGp30SwyvS3eM_tW-ghOhJ9PAJGK6KR54/edit?usp=sharing), and Snowflake team [here](https://docs.google.com/presentation/d/1qF3RkDAbOULwz9WK5TOltt2fE9t6uIc_hVNLFAaQX6A/edit?usp=sharing).
+- [2024/10] We have just created a developer slack ([slack.vllm.ai](https://slack.vllm.ai)) focusing on coordinating contributions and discussing features. Please feel free to join us there!
+- [2024/10] Ray Summit 2024 held a special track for vLLM! Please find the opening talk slides from the vLLM team [here](https://docs.google.com/presentation/d/1B_KQxpHBTRa_mDF-tR6i8rWdOU5QoTZNcEg2MKZxEHM/edit?usp=sharing). Learn more from the [talks](https://www.youtube.com/playlist?list=PLzTswPQNepXl6AQwifuwUImLPFRVpksjR) from other vLLM contributors and users!
+- [2024/09] We hosted [the sixth vLLM meetup](https://lu.ma/87q3nvnh) with NVIDIA! Please find the meetup slides [here](https://docs.google.com/presentation/d/1wrLGwytQfaOTd5wCGSPNhoaW3nq0E-9wqyP7ny93xRs/edit?usp=sharing).
+- [2024/07] We hosted [the fifth vLLM meetup](https://lu.ma/lp0gyjqr) with AWS! Please find the meetup slides [here](https://docs.google.com/presentation/d/1RgUD8aCfcHocghoP3zmXzck9vX3RCI9yfUAB2Bbcl4Y/edit?usp=sharing).
+- [2024/07] In partnership with Meta, vLLM officially supports Llama 3.1 with FP8 quantization and pipeline parallelism! Please check out our blog post [here](https://blog.vllm.ai/2024/07/23/llama31.html).
+- [2024/06] We hosted [the fourth vLLM meetup](https://lu.ma/agivllm) with Cloudflare and BentoML! Please find the meetup slides [here](https://docs.google.com/presentation/d/1iJ8o7V2bQEi0BFEljLTwc5G1S10_Rhv3beed5oB0NJ4/edit?usp=sharing).
+- [2024/04] We hosted [the third vLLM meetup](https://robloxandvllmmeetup2024.splashthat.com/) with Roblox! Please find the meetup slides [here](https://docs.google.com/presentation/d/1A--47JAK4BJ39t954HyTkvtfwn0fkqtsL8NGFuslReM/edit?usp=sharing).
+- [2024/01] We hosted [the second vLLM meetup](https://lu.ma/ygxbpzhl) with IBM! Please find the meetup slides [here](https://docs.google.com/presentation/d/12mI2sKABnUw5RBWXDYY-HtHth4iMSNcEoQ10jDQbxgA/edit?usp=sharing).
+- [2023/10] We hosted [the first vLLM meetup](https://lu.ma/first-vllm-meetup) with a16z! Please find the meetup slides [here](https://docs.google.com/presentation/d/1QL-XPFXiFpDBh86DbEegFXBXFXjix4v032GhShbKf3s/edit?usp=sharing).
+- [2023/08] We would like to express our sincere gratitude to [Andreessen Horowitz](https://a16z.com/2023/08/30/supporting-the-open-source-ai-community/) (a16z) for providing a generous grant to support the open-source development and research of vLLM.
+- [2023/06] We officially released vLLM! FastChat-vLLM integration has powered [LMSYS Vicuna and Chatbot Arena](https://chat.lmsys.org) since mid-April. Check out our [blog post](https://vllm.ai).
+
+</details>

 ---

@@ -43,7 +83,7 @@ vLLM is flexible and easy to use with:
 - Tensor, pipeline, data and expert parallelism support for distributed inference
 - Streaming outputs
 - OpenAI-compatible API server
- Support for NVIDIA GPUs, AMD CPUs and GPUs, Intel CPUs and GPUs, PowerPC CPUs, Arm CPUs, and TPU. Additionally, support for diverse hardware plugins such as Intel Gaudi, IBM Spyre and Huawei Ascend.
+- Support for NVIDIA GPUs, AMD CPUs and GPUs, Intel CPUs and GPUs, PowerPC CPUs, and TPU. Additionally, support for diverse hardware plugins such as Intel Gaudi, IBM Spyre and Huawei Ascend.
 - Prefix caching support
 - Multi-LoRA support

@@ -75,6 +115,47 @@ Visit our [documentation](https://docs.vllm.ai/en/latest/) to learn more.
 We welcome and value any contributions and collaborations.
 Please check out [Contributing to vLLM](https://docs.vllm.ai/en/latest/contributing/index.html) for how to get involved.

+## Sponsors
+
+vLLM is a community project. Our compute resources for development and testing are supported by the following organizations. Thank you for your support!
+
+<!-- Note: Please sort them in alphabetical order. -->
+<!-- Note: Please keep these consistent with docs/community/sponsors.md -->
+Cash Donations:
+
+- a16z
+- Dropbox
+- Sequoia Capital
+- Skywork AI
+- ZhenFund
+
+Compute Resources:
+
+- Alibaba Cloud
+- AMD
+- Anyscale
+- AWS
+- Crusoe Cloud
+- Databricks
+- DeepInfra
+- Google Cloud
+- Intel
+- Lambda Lab
+- Nebius
+- Novita AI
+- NVIDIA
+- Replicate
+- Roblox
+- RunPod
+- Trainy
+- UC Berkeley
+- UC San Diego
+- Volcengine
+
+Slack Sponsor: Anyscale
+
+We also have an official fundraising venue through [OpenCollective](https://opencollective.com/vllm). We plan to use the fund to support the development, maintenance, and adoption of vLLM.
+
 ## Citation

 If you use vLLM for your research, please cite our [paper](https://arxiv.org/abs/2309.06180):
@@ -95,7 +176,7 @@ If you use vLLM for your research, please cite our [paper](https://arxiv.org/abs
 - For discussing with fellow users, please use the [vLLM Forum](https://discuss.vllm.ai)
 - For coordinating contributions and development, please use [Slack](https://slack.vllm.ai)
 - For security disclosures, please use GitHub's [Security Advisories](https://github.com/vllm-project/vllm/security/advisories) feature
- For collaborations and partnerships, please contact us at [collaboration@vllm.ai](mailto:collaboration@vllm.ai)
+- For collaborations and partnerships, please contact us at [vllm-questions@lists.berkeley.edu](mailto:vllm-questions@lists.berkeley.edu)
 <!-- --8<-- [end:contact-us] -->

 ## Media Kit
@@ -83,7 +83,7 @@ MIN_CACHE_HIT_PCT=0
 MAX_LATENCY_ALLOWED_MS=100000000000 # A very large number
 ```

-### 2. Maximize Throughput with a Latency Requirement
+#### 2. Maximize Throughput with a Latency Requirement

 - **Goal**: Find the best server parameters when P99 end-to-end latency must be below 500ms.
 - **Configuration**:
@@ -96,7 +96,7 @@ MIN_CACHE_HIT_PCT=0
 MAX_LATENCY_ALLOWED_MS=500
 ```

-### 3. Maximize Throughput with Prefix Caching and Latency Requirements
+#### 3. Maximize Throughput with Prefix Caching and Latency Requirements

 - **Goal**: Find the best server parameters assuming a 60% prefix cache hit rate and a latency requirement of 500ms.
 - **Configuration**:
@@ -18,11 +18,6 @@ MIN_CACHE_HIT_PCT=${MIN_CACHE_HIT_PCT:-0}
 MAX_LATENCY_ALLOWED_MS=${MAX_LATENCY_ALLOWED_MS:-100000000000}
 NUM_SEQS_LIST=${NUM_SEQS_LIST:-"128 256"}
 NUM_BATCHED_TOKENS_LIST=${NUM_BATCHED_TOKENS_LIST:-"512 1024 2048 4096"}
-HOSTNAME=$(hostname)
-if [[ -z "$HOSTNAME" ]]; then
-    echo "Error: Failed to determine hostname." >&2
-    exit 1
-fi

 LOG_FOLDER="$BASE/auto-benchmark/$TAG"
 RESULT="$LOG_FOLDER/result.txt"
@@ -87,7 +82,6 @@ start_server() {
        "$MODEL"
        "--disable-log-requests"
        "--port" "8004"
-        "--host" "$HOSTNAME"
        "--gpu-memory-utilization" "$gpu_memory_utilization"
        "--max-num-seqs" "$max_num_seqs"
        "--max-num-batched-tokens" "$max_num_batched_tokens"
@@ -102,9 +96,8 @@ start_server() {
    # This correctly passes each element as a separate argument.
    if [[ -n "$profile_dir" ]]; then
        # Start server with profiling enabled
-        local profile_config_json="{\"profiler\": \"torch\", \"torch_profiler_dir\": \"$profile_dir\"}"
-        VLLM_SERVER_DEV_MODE=1 \
-            vllm serve --profiler-config "$profile_config_json" "${common_args_array[@]}" > "$vllm_log" 2>&1 &
+        VLLM_SERVER_DEV_MODE=1 VLLM_TORCH_PROFILER_DIR=$profile_dir \
+            vllm serve "${common_args_array[@]}" > "$vllm_log" 2>&1 &
    else
        # Start server without profiling
        VLLM_SERVER_DEV_MODE=1 \
@@ -119,7 +112,7 @@ start_server() {
        # since that we should always have permission to send signal to the server process.
        kill -0 $server_pid 2> /dev/null || break

-        RESPONSE=$(curl -s -X GET "http://${HOSTNAME}:8004/health" -w "%{http_code}" -o /dev/stdout)
+        RESPONSE=$(curl -s -X GET "http://0.0.0.0:8004/health" -w "%{http_code}" -o /dev/stdout)
        STATUS_CODE=$(echo "$RESPONSE" | tail -n 1)
        if [[ "$STATUS_CODE" -eq 200 ]]; then
            server_started=1
@@ -179,7 +172,6 @@ run_benchmark() {
        --goodput e2el:$MAX_LATENCY_ALLOWED_MS \
        --num-prompts 1000 \
        --random-prefix-len $prefix_len \
-        --host "$HOSTNAME" \
        --port 8004 &> "$bm_log"
    throughput=$(grep "Request throughput (req/s):" "$bm_log" | sed 's/[^0-9.]//g')
    e2el=$(grep "P99 E2EL (ms):" "$bm_log" | awk '{print $NF}')
@@ -195,7 +187,7 @@ run_benchmark() {
        request_rate=$((${throughput%.*} + 1))
        while ((request_rate > 0)); do
            # clear prefix cache
-            curl -X POST http://${HOSTNAME}:8004/reset_prefix_cache
+            curl -X POST http://0.0.0.0:8004/reset_prefix_cache
            sleep 5
            bm_log="$LOG_FOLDER/bm_log_${max_num_seqs}_${max_num_batched_tokens}_requestrate_${request_rate}.txt"
            vllm bench serve \
@@ -211,7 +203,6 @@ run_benchmark() {
                --goodput e2el:$MAX_LATENCY_ALLOWED_MS \
                --num-prompts 100 \
                --random-prefix-len $prefix_len \
-                --host "$HOSTNAME" \
                --port 8004 &> "$bm_log"
            throughput=$(grep "Request throughput (req/s):" "$bm_log" | sed 's/[^0-9.]//g')
            e2el=$(grep "P99 E2EL (ms):" "$bm_log" | awk '{print $NF}')
@@ -312,7 +303,6 @@ if (( $(echo "$best_throughput > 0" | bc -l) )); then
        --goodput e2el:$MAX_LATENCY_ALLOWED_MS \
        --num-prompts 100 \
        --random-prefix-len $prefix_len \
-        --host "$HOSTNAME" \
        --port 8004 \
        --profile &> "$bm_log"
 else
@@ -620,7 +620,7 @@ def get_tokenizer(
        kwargs["use_fast"] = False
    if tokenizer_mode == "mistral":
        try:
-            from vllm.tokenizers.mistral import MistralTokenizer
+            from vllm.transformers_utils.tokenizer import MistralTokenizer
        except ImportError as e:
            raise ImportError(
                "MistralTokenizer requires vllm package.\n"
@@ -1,380 +0,0 @@
-#!/usr/bin/env python3
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-"""
-Benchmark to measure the performance overhead of VLLM_BATCH_INVARIANT mode.
-
-This benchmark runs the same workload twice:
-1. With VLLM_BATCH_INVARIANT=0 (baseline)
-2. With VLLM_BATCH_INVARIANT=1 (batch invariant mode)
-
-And reports the timing and throughput metrics for comparison.
-
-Environment variables:
-    VLLM_BENCH_MODEL: Model to benchmark (default: "Qwen/Qwen3-1.7B")
-    VLLM_BENCH_TP_SIZE: Tensor parallel size (default: 1, use 8 for deepseek)
-    VLLM_BENCH_BATCH_SIZE: Max batch size (default: 128)
-    VLLM_BENCH_NUM_TRIALS: Number of trials to run (default: 5)
-    VLLM_BENCH_MIN_PROMPT: Min prompt length in words (default: 1024)
-    VLLM_BENCH_MAX_PROMPT: Max prompt length in words (default: 2048)
-    VLLM_BENCH_MAX_TOKENS: Max tokens to generate (default: 128)
-    VLLM_BENCH_TEMPERATURE: Temperature for sampling (default: 0.0)
-    VLLM_BENCH_GPU_MEMORY_UTILIZATION: GPU memory utilization (default: 0.4)
-    VLLM_BENCH_MAX_MODEL_LEN: Max model length (default: 5120)
-    VLLM_BENCH_BACKEND: Attention backend (default: FLASH_ATTN)
-
-Example usage:
-    # Benchmark qwen3 (default)
-    python benchmarks/benchmark_batch_invariance.py
-
-    # Benchmark deepseek with 8 GPUs
-    VLLM_BENCH_MODEL="deepseek-ai/DeepSeek-V3" VLLM_BENCH_TP_SIZE=8 \\
-        python benchmarks/benchmark_batch_invariance.py
-
-    # Quick test with fewer trials
-    VLLM_BENCH_NUM_TRIALS=2 VLLM_BENCH_BATCH_SIZE=32 \\
-        python benchmarks/benchmark_batch_invariance.py
-"""
-
-import contextlib
-import os
-import random
-import time
-
-from vllm import LLM, SamplingParams
-from vllm.platforms import current_platform
-
-
-def _random_prompt(min_words: int = 1024, max_words: int = 1024 * 2) -> str:
-    """Generate a random prompt for benchmarking."""
-    prompt_templates = [
-        "Question: What is the capital of France?\nAnswer: The capital of France is",
-        "Q: How does photosynthesis work?\nA: Photosynthesis is the process by which",
-        "User: Can you explain quantum mechanics?\nAssistant: Quantum mechanics is",
-        "Once upon a time in a distant galaxy, there lived",
-        "The old man walked slowly down the street, remembering",
-        "In the year 2157, humanity finally discovered",
-        "To implement a binary search tree in Python, first we need to",
-        "The algorithm works by iterating through the array and",
-        "Here's how to optimize database queries using indexing:",
-        "The Renaissance was a period in European history that",
-        "Climate change is caused by several factors including",
-        "The human brain contains approximately 86 billion neurons which",
-        "I've been thinking about getting a new laptop because",
-        "Yesterday I went to the store and bought",
-        "My favorite thing about summer is definitely",
-    ]
-
-    base_prompt = random.choice(prompt_templates)
-
-    if max_words < min_words:
-        max_words = min_words
-    target_words = random.randint(min_words, max_words)
-
-    if target_words > 50:
-        padding_text = (
-            " This is an interesting topic that deserves more explanation. "
-            * (target_words // 50)
-        )
-        base_prompt = base_prompt + padding_text
-
-    return base_prompt
-
-
-def run_benchmark_with_batch_invariant(
-    model: str,
-    tp_size: int,
-    max_batch_size: int,
-    num_trials: int,
-    min_prompt: int,
-    max_prompt: int,
-    max_tokens: int,
-    temperature: float,
-    gpu_mem_util: float,
-    max_model_len: int,
-    backend: str,
-    batch_invariant: bool,
-    seed: int = 12345,
-) -> dict:
-    """
-    Run the benchmark with the specified configuration.
-
-    Returns a dict with timing and throughput metrics.
-    """
-    random.seed(seed)
-
-    # Set environment variables
-    if batch_invariant:
-        os.environ["VLLM_BATCH_INVARIANT"] = "1"
-    else:
-        os.environ["VLLM_BATCH_INVARIANT"] = "0"
-
-    print(f"\n{'=' * 80}")
-    print(f"BENCHMARK: VLLM_BATCH_INVARIANT={int(batch_invariant)}")
-    print(f"  Model: {model}")
-    print(f"  TP Size: {tp_size}")
-    print(f"  Backend: {backend}")
-    print(f"  Max Batch Size: {max_batch_size}")
-    print(f"  Trials: {num_trials}")
-    print(f"  Max Tokens: {max_tokens}")
-    print(f"{'=' * 80}\n")
-
-    sampling = SamplingParams(
-        temperature=temperature,
-        top_p=0.95,
-        max_tokens=max_tokens,
-        seed=20240919,
-    )
-
-    needle_prompt = "There once was a "
-
-    llm = None
-    try:
-        # Create LLM engine
-        start_init = time.perf_counter()
-        llm = LLM(
-            model=model,
-            max_num_seqs=max_batch_size,
-            gpu_memory_utilization=gpu_mem_util,
-            max_model_len=max_model_len,
-            dtype="bfloat16",
-            tensor_parallel_size=tp_size,
-            attention_config={"backend": backend},
-            enable_prefix_caching=False,
-        )
-        init_time = time.perf_counter() - start_init
-        print(f"Engine initialization time: {init_time:.2f}s\n")
-
-        # Generate baseline
-        print("Generating baseline (warmup)...")
-        baseline_out = llm.generate([needle_prompt], sampling)
-        assert len(baseline_out) == 1
-        baseline_text = baseline_out[0].outputs[0].text
-        print(f"Baseline output: '{baseline_text[:50]}...'\n")
-
-        # Run trials and measure timing
-        trial_times: list[float] = []
-        total_tokens = 0
-        total_prompts = 0
-
-        for trial in range(num_trials):
-            # Create a batch
-            prompts: list[str] = []
-            batch_size = random.randint(max_batch_size // 2, max_batch_size)
-            needle_pos = random.randint(0, batch_size - 1)
-            for i in range(batch_size):
-                if i == needle_pos:
-                    prompts.append(needle_prompt)
-                else:
-                    prompts.append(_random_prompt(min_prompt, max_prompt))
-
-            # Measure time for this trial
-            start_time = time.perf_counter()
-            outputs = llm.generate(prompts, sampling)
-            trial_time = time.perf_counter() - start_time
-
-            trial_times.append(trial_time)
-            total_prompts += len(prompts)
-
-            # Count tokens
-            for output in outputs:
-                if output.outputs:
-                    total_tokens += len(output.outputs[0].token_ids)
-
-            print(
-                f"Trial {trial + 1}/{num_trials}: "
-                f"batch_size={batch_size}, "
-                f"time={trial_time:.2f}s"
-            )
-
-            # Verify needle output still matches
-            needle_output = outputs[needle_pos]
-            assert needle_output.prompt == needle_prompt
-
-        # Compute statistics
-        avg_time = sum(trial_times) / len(trial_times)
-        min_time = min(trial_times)
-        max_time = max(trial_times)
-        throughput = total_tokens / sum(trial_times)
-        prompts_per_sec = total_prompts / sum(trial_times)
-
-        print(f"\n{'=' * 80}")
-        print("RESULTS:")
-        print(f"  Average time per trial: {avg_time:.2f}s")
-        print(f"  Min time: {min_time:.2f}s")
-        print(f"  Max time: {max_time:.2f}s")
-        print(f"  Total tokens generated: {total_tokens}")
-        print(f"  Total prompts processed: {total_prompts}")
-        print(f"  Throughput: {throughput:.2f} tokens/s")
-        print(f"  Prompts/s: {prompts_per_sec:.2f}")
-        print(f"{'=' * 80}\n")
-
-        return {
-            "init_time": init_time,
-            "avg_time": avg_time,
-            "min_time": min_time,
-            "max_time": max_time,
-            "total_tokens": total_tokens,
-            "total_prompts": total_prompts,
-            "throughput": throughput,
-            "prompts_per_sec": prompts_per_sec,
-            "trial_times": trial_times,
-        }
-
-    finally:
-        # Cleanup
-        if llm is not None:
-            with contextlib.suppress(Exception):
-                llm.shutdown()
-
-
-def main():
-    # Check platform support
-    if not (current_platform.is_cuda() and current_platform.has_device_capability(90)):
-        print("ERROR: Requires CUDA and >= Hopper (SM90)")
-        print(f"Current platform: {current_platform.device_type}")
-        if current_platform.is_cuda():
-            print(f"Device capability: {current_platform.get_device_capability()}")
-        return 1
-
-    # Read configuration from environment
-    model = os.getenv("VLLM_BENCH_MODEL", "Qwen/Qwen3-1.7B")
-    tp_size = int(os.getenv("VLLM_BENCH_TP_SIZE", "1"))
-    max_batch_size = int(os.getenv("VLLM_BENCH_BATCH_SIZE", "128"))
-    num_trials = int(os.getenv("VLLM_BENCH_NUM_TRIALS", "5"))
-    min_prompt = int(os.getenv("VLLM_BENCH_MIN_PROMPT", "1024"))
-    max_prompt = int(os.getenv("VLLM_BENCH_MAX_PROMPT", "2048"))
-    max_tokens = int(os.getenv("VLLM_BENCH_MAX_TOKENS", "128"))
-    temperature = float(os.getenv("VLLM_BENCH_TEMPERATURE", "0.0"))
-    gpu_mem_util = float(os.getenv("VLLM_BENCH_GPU_MEMORY_UTILIZATION", "0.4"))
-    max_model_len = int(os.getenv("VLLM_BENCH_MAX_MODEL_LEN", "5120"))
-    backend = os.getenv("VLLM_BENCH_BACKEND", "FLASH_ATTN")
-
-    print("\n" + "=" * 80)
-    print("VLLM BATCH INVARIANCE BENCHMARK")
-    print("=" * 80)
-    print("\nConfiguration:")
-    print(f"  Model: {model}")
-    print(f"  Tensor Parallel Size: {tp_size}")
-    print(f"  Attention Backend: {backend}")
-    print(f"  Max Batch Size: {max_batch_size}")
-    print(f"  Number of Trials: {num_trials}")
-    print(f"  Prompt Length Range: {min_prompt}-{max_prompt} words")
-    print(f"  Max Tokens to Generate: {max_tokens}")
-    print(f"  Temperature: {temperature}")
-    print(f"  GPU Memory Utilization: {gpu_mem_util}")
-    print(f"  Max Model Length: {max_model_len}")
-    print("=" * 80)
-
-    # Run benchmark WITHOUT batch invariance (baseline)
-    print("\n" + "=" * 80)
-    print("PHASE 1: Running WITHOUT batch invariance (baseline)")
-    print("=" * 80)
-    baseline_results = run_benchmark_with_batch_invariant(
-        model=model,
-        tp_size=tp_size,
-        max_batch_size=max_batch_size,
-        num_trials=num_trials,
-        min_prompt=min_prompt,
-        max_prompt=max_prompt,
-        max_tokens=max_tokens,
-        temperature=temperature,
-        gpu_mem_util=gpu_mem_util,
-        max_model_len=max_model_len,
-        backend=backend,
-        batch_invariant=False,
-    )
-
-    # Run benchmark WITH batch invariance
-    print("\n" + "=" * 80)
-    print("PHASE 2: Running WITH batch invariance")
-    print("=" * 80)
-    batch_inv_results = run_benchmark_with_batch_invariant(
-        model=model,
-        tp_size=tp_size,
-        max_batch_size=max_batch_size,
-        num_trials=num_trials,
-        min_prompt=min_prompt,
-        max_prompt=max_prompt,
-        max_tokens=max_tokens,
-        temperature=temperature,
-        gpu_mem_util=gpu_mem_util,
-        max_model_len=max_model_len,
-        backend=backend,
-        batch_invariant=True,
-    )
-
-    # Compare results
-    print("\n" + "=" * 80)
-    print("COMPARISON: Batch Invariance vs Baseline")
-    print("=" * 80)
-
-    init_overhead_pct = (
-        (batch_inv_results["init_time"] - baseline_results["init_time"])
-        / baseline_results["init_time"]
-        * 100
-    )
-    time_overhead_pct = (
-        (batch_inv_results["avg_time"] - baseline_results["avg_time"])
-        / baseline_results["avg_time"]
-        * 100
-    )
-    throughput_change_pct = (
-        (batch_inv_results["throughput"] - baseline_results["throughput"])
-        / baseline_results["throughput"]
-        * 100
-    )
-
-    print("\nInitialization Time:")
-    print(f"  Baseline:         {baseline_results['init_time']:.2f}s")
-    print(f"  Batch Invariant:  {batch_inv_results['init_time']:.2f}s")
-    print(f"  Overhead:         {init_overhead_pct:+.2f}%")
-
-    print("\nAverage Trial Time:")
-    print(f"  Baseline:         {baseline_results['avg_time']:.2f}s")
-    print(f"  Batch Invariant:  {batch_inv_results['avg_time']:.2f}s")
-    print(f"  Overhead:         {time_overhead_pct:+.2f}%")
-
-    print("\nThroughput (tokens/s):")
-    print(f"  Baseline:         {baseline_results['throughput']:.2f}")
-    print(f"  Batch Invariant:  {batch_inv_results['throughput']:.2f}")
-    print(f"  Change:           {throughput_change_pct:+.2f}%")
-
-    print("\nPrompts/s:")
-    print(f"  Baseline:         {baseline_results['prompts_per_sec']:.2f}")
-    print(f"  Batch Invariant:  {batch_inv_results['prompts_per_sec']:.2f}")
-
-    print("\n" + "=" * 80)
-    print("SUMMARY")
-    print("=" * 80)
-    if time_overhead_pct > 0:
-        print(
-            f"Batch invariance mode adds approximately {time_overhead_pct:.1f}% "
-            "overhead"
-        )
-    else:
-        print(
-            f"Batch invariance mode is approximately {-time_overhead_pct:.1f}% "
-            "faster (unexpected!)"
-        )
-
-    if abs(throughput_change_pct) < 1.0:
-        print("Throughput difference is negligible (< 1%)")
-    elif throughput_change_pct < 0:
-        print(
-            f"Throughput decreased by {-throughput_change_pct:.1f}% "
-            "with batch invariance"
-        )
-    else:
-        print(
-            f"Throughput increased by {throughput_change_pct:.1f}% "
-            "with batch invariance (unexpected!)"
-        )
-
-    print("=" * 80 + "\n")
-
-    return 0
-
-
-if __name__ == "__main__":
-    exit(main())
@@ -1,120 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-"""
-Micro benchmark comparing built-in hash(), SHA-256, and xxHash.
-
-This focuses on a single test payload shaped like the prefix-cache hash input:
-    (32-byte bytes object, 32-int tuple)
-
-Usage:
-    python benchmarks/hash_micro_benchmark.py --iterations 20000
-"""
-
-from __future__ import annotations
-
-import argparse
-import random
-import statistics
-import time
-from collections.abc import Callable, Iterable
-
-from vllm.utils.hashing import sha256, xxhash
-
-
-def _generate_test_data(seed: int) -> tuple[bytes, tuple[int, ...]]:
-    """Generate a deterministic test payload."""
-    random.seed(seed)
-    bytes_data = bytes(random.getrandbits(8) for _ in range(32))
-    int_tuple = tuple(random.randint(1, 1_000_000) for _ in range(32))
-    return (bytes_data, int_tuple)
-
-
-def _benchmark_func(func: Callable[[tuple], object], data: tuple, iterations: int):
-    """Return (avg_seconds, std_seconds) for hashing `data` `iterations` times."""
-    times: list[float] = []
-
-    # Warm-up to avoid first-run noise.
-    for _ in range(200):
-        func(data)
-
-    for _ in range(iterations):
-        start = time.perf_counter()
-        func(data)
-        end = time.perf_counter()
-        times.append(end - start)
-
-    avg = statistics.mean(times)
-    std = statistics.stdev(times) if len(times) > 1 else 0.0
-    return avg, std
-
-
-def _run_benchmarks(
-    benchmarks: Iterable[tuple[str, Callable[[tuple], object]]],
-    data: tuple,
-    iterations: int,
-):
-    """Yield (name, avg, std) for each benchmark, skipping unavailable ones."""
-    for name, func in benchmarks:
-        try:
-            avg, std = _benchmark_func(func, data, iterations)
-        except ModuleNotFoundError as exc:
-            print(f"Skipping {name}: {exc}")
-            continue
-        yield name, avg, std
-
-
-def builtin_hash(data: tuple) -> int:
-    """Wrapper for Python's built-in hash()."""
-    return hash(data)
-
-
-def main() -> None:
-    parser = argparse.ArgumentParser(description=__doc__)
-    parser.add_argument(
-        "--iterations",
-        type=int,
-        default=10_000,
-        help="Number of measured iterations per hash function.",
-    )
-    parser.add_argument(
-        "--seed", type=int, default=42, help="Random seed for test payload."
-    )
-    args = parser.parse_args()
-
-    data = _generate_test_data(args.seed)
-    benchmarks = (
-        ("SHA256 (pickle)", sha256),
-        ("xxHash (pickle)", xxhash),
-        ("built-in hash()", builtin_hash),
-    )
-
-    print("=" * 60)
-    print("HASH FUNCTION MICRO BENCHMARK")
-    print("=" * 60)
-    print("Test data: (32-byte bytes object, 32-int tuple)")
-    print(f"Iterations: {args.iterations:,}")
-    print("=" * 60)
-
-    results = list(_run_benchmarks(benchmarks, data, args.iterations))
-    builtin_entry = next((r for r in results if r[0] == "built-in hash()"), None)
-
-    print("\nResults:")
-    for name, avg, std in results:
-        print(f"  {name:16s}: {avg * 1e6:8.2f} ± {std * 1e6:6.2f} μs")
-
-    if builtin_entry:
-        _, builtin_avg, _ = builtin_entry
-        print("\n" + "=" * 60)
-        print("SUMMARY (relative to built-in hash())")
-        print("=" * 60)
-        for name, avg, _ in results:
-            if name == "built-in hash()":
-                continue
-            speed_ratio = avg / builtin_avg
-            print(f"• {name} is {speed_ratio:.1f}x slower than built-in hash()")
-    else:
-        print("\nBuilt-in hash() result missing; cannot compute speed ratios.")
-
-
-if __name__ == "__main__":
-    main()
@@ -32,11 +32,12 @@ def benchmark_propose(args):

        model_config = ModelConfig(
            model="facebook/opt-125m",
+            task="generate",
            max_model_len=args.num_token + args.num_spec_token,
            tokenizer="facebook/opt-125m",
            tokenizer_mode="auto",
            dtype="auto",
-            seed=0,
+            seed=None,
            trust_remote_code=False,
        )
        proposer = NgramProposer(
@@ -107,10 +108,7 @@ def benchmark_batched_propose(args):
        device_config=DeviceConfig(device=current_platform.device_type),
        parallel_config=ParallelConfig(),
        load_config=LoadConfig(),
-        scheduler_config=SchedulerConfig(
-            max_model_len=model_config.max_model_len,
-            is_encoder_decoder=model_config.is_encoder_decoder,
-        ),
+        scheduler_config=SchedulerConfig(),
    )

    # monkey patch vllm.v1.worker.gpu_model_runner.get_pp_group
@@ -1,110 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-
-"""
-Simple benchmark to compare prefix-cache block hashing algorithms.
-
-Example:
-    python benchmark_prefix_block_hash.py --num-blocks 20000 --block-size 32
-"""
-
-from __future__ import annotations
-
-import argparse
-import random
-import statistics
-import sys
-import time
-from collections.abc import Callable, Iterable, Sequence
-
-from vllm.utils.hashing import get_hash_fn_by_name
-from vllm.v1.core.kv_cache_utils import BlockHash, hash_block_tokens, init_none_hash
-
-SUPPORTED_ALGOS = ("sha256", "sha256_cbor", "xxhash", "xxhash_cbor")
-
-
-def _generate_blocks(
-    num_blocks: int, block_size: int, vocab_size: int, seed: int
-) -> list[list[int]]:
-    rng = random.Random(seed)
-    return [
-        [rng.randrange(vocab_size) for _ in range(block_size)]
-        for _ in range(num_blocks)
-    ]
-
-
-def _hash_all_blocks(
-    hash_fn: Callable[[object], bytes],
-    blocks: Iterable[Sequence[int]],
-) -> float:
-    parent_hash: BlockHash | None = None
-    start = time.perf_counter()
-    for block in blocks:
-        parent_hash = hash_block_tokens(hash_fn, parent_hash, block, extra_keys=None)
-    end = time.perf_counter()
-    return end - start
-
-
-def _benchmark(
-    hash_algo: str,
-    blocks: list[list[int]],
-    trials: int,
-) -> tuple[float, float, float] | None:
-    try:
-        hash_fn = get_hash_fn_by_name(hash_algo)
-        init_none_hash(hash_fn)
-        timings = [_hash_all_blocks(hash_fn, blocks) for _ in range(trials)]
-    except ModuleNotFoundError as exc:
-        print(f"Skipping {hash_algo}: {exc}", file=sys.stderr)
-        return None
-
-    avg = statistics.mean(timings)
-    best = min(timings)
-    # throughput: tokens / second
-    tokens_hashed = len(blocks) * len(blocks[0])
-    throughput = tokens_hashed / best
-    return avg, best, throughput
-
-
-def main() -> None:
-    parser = argparse.ArgumentParser(description=__doc__)
-    parser.add_argument("--num-blocks", type=int, default=10000, help="Block count.")
-    parser.add_argument("--block-size", type=int, default=32, help="Tokens per block.")
-    parser.add_argument(
-        "--vocab-size", type=int, default=32000, help="Token id range [0, vocab_size)."
-    )
-    parser.add_argument("--seed", type=int, default=0, help="Random seed.")
-    parser.add_argument(
-        "--trials", type=int, default=5, help="Number of timed trials per algorithm."
-    )
-    parser.add_argument(
-        "--algorithms",
-        nargs="+",
-        default=SUPPORTED_ALGOS,
-        choices=SUPPORTED_ALGOS,
-        help="Hash algorithms to benchmark.",
-    )
-    args = parser.parse_args()
-
-    blocks = _generate_blocks(
-        args.num_blocks, args.block_size, args.vocab_size, args.seed
-    )
-    print(
-        f"Benchmarking {len(args.algorithms)} algorithms on "
-        f"{args.num_blocks} blocks (block size={args.block_size})."
-    )
-
-    for algo in args.algorithms:
-        result = _benchmark(algo, blocks, args.trials)
-        if result is None:
-            continue
-
-        avg, best, throughput = result
-        print(
-            f"{algo:14s} avg: {avg:.6f}s  best: {best:.6f}s  "
-            f"throughput: {throughput / 1e6:.2f}M tokens/s"
-        )
-
-
-if __name__ == "__main__":
-    main()
@@ -40,7 +40,7 @@ from vllm.engine.arg_utils import EngineArgs
 from vllm.utils.argparse_utils import FlexibleArgumentParser

 try:
-    from vllm.tokenizers import get_tokenizer
+    from vllm.transformers_utils.tokenizer import get_tokenizer
 except ImportError:
    from backend_request_func import get_tokenizer

@@ -69,7 +69,7 @@ def sample_tokens(tokenizer: PreTrainedTokenizerBase, length: int) -> list[int]:

    # Remove the special tokens.
    return random.choices(
-        [v for v in vocab.values() if v not in all_special_ids],
+        [v for k, v in vocab.items() if k not in all_special_ids],
        k=length,
    )

@@ -46,7 +46,7 @@ from tqdm.asyncio import tqdm
 from transformers import PreTrainedTokenizerBase

 try:
-    from vllm.tokenizers import get_tokenizer
+    from vllm.transformers_utils.tokenizer import get_tokenizer
 except ImportError:
    from backend_request_func import get_tokenizer

@@ -574,7 +574,7 @@ async def benchmark(
    )
    print(
        "{:<40} {:<10.2f}".format(
-            "Total token throughput (tok/s):", metrics.total_token_throughput
+            "Total Token throughput (tok/s):", metrics.total_token_throughput
        )
    )

@@ -963,7 +963,8 @@ def create_argument_parser():
    parser.add_argument(
        "--profile",
        action="store_true",
-        help="Use vLLM Profiling. --profiler-config must be provided on the server.",
+        help="Use Torch Profiler. The endpoint must be launched with "
+        "VLLM_TORCH_PROFILER_DIR to enable profiler.",
    )
    parser.add_argument(
        "--result-dir",
@@ -5,12 +5,11 @@ import argparse
 import asyncio
 import logging
 import os
-import time
-import uuid
-from urllib.parse import urlparse

 import aiohttp
 from quart import Quart, Response, make_response, request
+from rate_limiter import RateLimiter
+from request_queue import RequestQueue

 # Configure logging
 logging.basicConfig(level=logging.INFO)
@@ -25,8 +24,26 @@ def parse_args():
    parser.add_argument(
        "--timeout",
        type=float,
-        default=6 * 60 * 60,
-        help="Timeout for backend service requests in seconds (default: 21600)",
+        default=300,
+        help="Timeout for backend service requests in seconds (default: 300)",
+    )
+    parser.add_argument(
+        "--max-concurrent",
+        type=int,
+        default=100,
+        help="Maximum concurrent requests to backend services (default: 100)",
+    )
+    parser.add_argument(
+        "--queue-size",
+        type=int,
+        default=500,
+        help="Maximum number of requests in the queue (default: 500)",
+    )
+    parser.add_argument(
+        "--rate-limit",
+        type=int,
+        default=40,
+        help="Maximum requests per second (default: 40)",
    )
    parser.add_argument(
        "--port",
@@ -37,32 +54,14 @@ def parse_args():
    parser.add_argument(
        "--prefill-url",
        type=str,
-        default="http://localhost:8100",
-        help="Prefill service base URL (protocol + host[:port])",
+        default="http://localhost:8100/v1/completions",
+        help="Prefill service endpoint URL",
    )
    parser.add_argument(
        "--decode-url",
        type=str,
-        default="http://localhost:8200",
-        help="Decode service base URL (protocol + host[:port])",
-    )
-    parser.add_argument(
-        "--kv-host",
-        type=str,
-        default="localhost",
-        help="Hostname or IP used by KV transfer (default: localhost)",
-    )
-    parser.add_argument(
-        "--prefill-kv-port",
-        type=int,
-        default=14579,
-        help="Prefill KV port (default: 14579)",
-    )
-    parser.add_argument(
-        "--decode-kv-port",
-        type=int,
-        default=14580,
-        help="Decode KV port (default: 14580)",
+        default="http://localhost:8200/v1/completions",
+        help="Decode service endpoint URL",
    )

    return parser.parse_args()
@@ -74,129 +73,70 @@ def main():

    # Initialize configuration using command line parameters
    AIOHTTP_TIMEOUT = aiohttp.ClientTimeout(total=args.timeout)
+    MAX_CONCURRENT_REQUESTS = args.max_concurrent
+    REQUEST_QUEUE_SIZE = args.queue_size
+    RATE_LIMIT = args.rate_limit
    PREFILL_SERVICE_URL = args.prefill_url
    DECODE_SERVICE_URL = args.decode_url
    PORT = args.port

-    PREFILL_KV_ADDR = f"{args.kv_host}:{args.prefill_kv_port}"
-    DECODE_KV_ADDR = f"{args.kv_host}:{args.decode_kv_port}"
-
-    logger.info(
-        "Proxy resolved KV addresses -> prefill: %s, decode: %s",
-        PREFILL_KV_ADDR,
-        DECODE_KV_ADDR,
-    )
-
    app = Quart(__name__)

-    # Attach the configuration object to the application instance so helper
-    # coroutines can read the resolved backend URLs and timeouts without using
-    # globals.
+    # Initialize the rate limiter and request queue
+    rate_limiter = RateLimiter(RATE_LIMIT)
+    request_queue = RequestQueue(MAX_CONCURRENT_REQUESTS, REQUEST_QUEUE_SIZE)
+
+    # Attach the configuration object to the application instance
    app.config.update(
        {
            "AIOHTTP_TIMEOUT": AIOHTTP_TIMEOUT,
+            "rate_limiter": rate_limiter,
+            "request_queue": request_queue,
            "PREFILL_SERVICE_URL": PREFILL_SERVICE_URL,
            "DECODE_SERVICE_URL": DECODE_SERVICE_URL,
-            "PREFILL_KV_ADDR": PREFILL_KV_ADDR,
-            "DECODE_KV_ADDR": DECODE_KV_ADDR,
        }
    )

-    def _normalize_base_url(url: str) -> str:
-        """Remove any trailing slash so path joins behave predictably."""
-        return url.rstrip("/")
+    # Start queue processing on app startup
+    @app.before_serving
+    async def startup():
+        """Start request processing task when app starts serving"""
+        asyncio.create_task(request_queue.process())

-    def _get_host_port(url: str) -> str:
-        """Return the hostname:port portion for logging and KV headers."""
-        parsed = urlparse(url)
-        host = parsed.hostname or "localhost"
-        port = parsed.port
-        if port is None:
-            port = 80 if parsed.scheme == "http" else 443
-        return f"{host}:{port}"
+    async def forward_request(url, data):
+        """Forward request to backend service with rate limiting and error handling"""
+        headers = {"Authorization": f"Bearer {os.environ.get('OPENAI_API_KEY')}"}

-    PREFILL_BASE = _normalize_base_url(PREFILL_SERVICE_URL)
-    DECODE_BASE = _normalize_base_url(DECODE_SERVICE_URL)
-    KV_TARGET = _get_host_port(DECODE_SERVICE_URL)
-
-    def _build_headers(request_id: str) -> dict[str, str]:
-        """Construct the headers expected by vLLM's P2P disagg connector."""
-        headers: dict[str, str] = {"X-Request-Id": request_id, "X-KV-Target": KV_TARGET}
-        api_key = os.environ.get("OPENAI_API_KEY")
-        if api_key:
-            headers["Authorization"] = f"Bearer {api_key}"
-        return headers
-
-    async def _run_prefill(
-        request_path: str,
-        payload: dict,
-        headers: dict[str, str],
-        request_id: str,
-    ):
-        url = f"{PREFILL_BASE}{request_path}"
-        start_ts = time.perf_counter()
-        logger.info("[prefill] start request_id=%s url=%s", request_id, url)
-        try:
-            async with (
-                aiohttp.ClientSession(timeout=AIOHTTP_TIMEOUT) as session,
-                session.post(url=url, json=payload, headers=headers) as resp,
-            ):
-                if resp.status != 200:
-                    error_text = await resp.text()
-                    raise RuntimeError(
-                        f"Prefill backend error {resp.status}: {error_text}"
-                    )
-                await resp.read()
-                logger.info(
-                    "[prefill] done request_id=%s status=%s elapsed=%.2fs",
-                    request_id,
-                    resp.status,
-                    time.perf_counter() - start_ts,
-                )
-        except asyncio.TimeoutError as exc:
-            raise RuntimeError(f"Prefill service timeout at {url}") from exc
-        except aiohttp.ClientError as exc:
-            raise RuntimeError(f"Prefill service unavailable at {url}") from exc
-
-    async def _stream_decode(
-        request_path: str,
-        payload: dict,
-        headers: dict[str, str],
-        request_id: str,
-    ):
-        url = f"{DECODE_BASE}{request_path}"
-        # Stream tokens from the decode service once the prefill stage has
-        # materialized KV caches on the target workers.
-        logger.info("[decode] start request_id=%s url=%s", request_id, url)
-        try:
-            async with (
-                aiohttp.ClientSession(timeout=AIOHTTP_TIMEOUT) as session,
-                session.post(url=url, json=payload, headers=headers) as resp,
-            ):
-                if resp.status != 200:
-                    error_text = await resp.text()
-                    logger.error(
-                        "Decode backend error %s - %s", resp.status, error_text
-                    )
-                    err_msg = (
-                        '{"error": "Decode backend error ' + str(resp.status) + '"}'
-                    )
-                    yield err_msg.encode()
-                    return
-                logger.info(
-                    "[decode] streaming response request_id=%s status=%s",
-                    request_id,
-                    resp.status,
-                )
-                async for chunk_bytes in resp.content.iter_chunked(1024):
-                    yield chunk_bytes
-                logger.info("[decode] finished streaming request_id=%s", request_id)
-        except asyncio.TimeoutError:
-            logger.error("Decode service timeout at %s", url)
-            yield b'{"error": "Decode service timeout"}'
-        except aiohttp.ClientError as exc:
-            logger.error("Decode service error at %s: %s", url, exc)
-            yield b'{"error": "Decode service unavailable"}'
+        # Use rate limiter as context manager
+        async with (
+            rate_limiter,
+            aiohttp.ClientSession(timeout=AIOHTTP_TIMEOUT) as session,
+        ):
+            try:
+                async with session.post(
+                    url=url, json=data, headers=headers
+                ) as response:
+                    if response.status == 200:
+                        # Stream response chunks
+                        async for chunk_bytes in response.content.iter_chunked(1024):
+                            yield chunk_bytes
+                    else:
+                        # Handle backend service errors
+                        error_text = await response.text()
+                        logger.error(
+                            "Backend service error: %s - %s",
+                            response.status,
+                            error_text,
+                        )
+                        yield b'{"error": "Backend service error"}'
+            except aiohttp.ClientError as e:
+                # Handle connection errors
+                logger.error("Connection error to %s: %s", url, str(e))
+                yield b'{"error": "Service unavailable"}'
+            except asyncio.TimeoutError:
+                # Handle timeout errors
+                logger.error("Timeout connecting to %s", url)
+                yield b'{"error": "Service timeout"}'

    async def process_request():
        """Process a single request through prefill and decode stages"""
@@ -206,27 +146,13 @@ def main():
            # Create prefill request (max_tokens=1)
            prefill_request = original_request_data.copy()
            prefill_request["max_tokens"] = 1
-            if "max_completion_tokens" in prefill_request:
-                prefill_request["max_completion_tokens"] = 1

            # Execute prefill stage
-            # The request id encodes both KV socket addresses so the backend can
-            # shuttle tensors directly via NCCL once the prefill response
-            # completes.
-            request_id = (
-                f"___prefill_addr_{PREFILL_KV_ADDR}___decode_addr_"
-                f"{DECODE_KV_ADDR}_{uuid.uuid4().hex}"
-            )
-
-            headers = _build_headers(request_id)
-            await _run_prefill(request.path, prefill_request, headers, request_id)
+            async for _ in forward_request(PREFILL_SERVICE_URL, prefill_request):
+                continue

            # Execute decode stage and stream response
-            # Pass the unmodified user request so the decode phase can continue
-            # sampling with the already-populated KV cache.
-            generator = _stream_decode(
-                request.path, original_request_data, headers, request_id
-            )
+            generator = forward_request(DECODE_SERVICE_URL, original_request_data)
            response = await make_response(generator)
            response.timeout = None  # Disable timeout for streaming response
            return response
@@ -242,10 +168,23 @@ def main():
    @app.route("/v1/completions", methods=["POST"])
    async def handle_request():
        """Handle incoming API requests with concurrency and rate limiting"""
+        # Create task for request processing
+        task = asyncio.create_task(process_request())
+
+        # Enqueue request or reject if queue is full
+        if not await request_queue.enqueue(task):
+            return Response(
+                response=b'{"error": "Server busy, try again later"}',
+                status=503,
+                content_type="application/json",
+            )
+
        try:
-            return await process_request()
+            # Return the response from the processing task
+            return await task
        except asyncio.CancelledError:
-            logger.warning("Request cancelled")
+            # Handle task cancellation (timeout or queue full)
+            logger.warning("Request cancelled due to timeout or queue full")
            return Response(
                response=b'{"error": "Request cancelled"}',
                status=503,
@@ -14,9 +14,6 @@ from tqdm import tqdm

 import vllm._custom_ops as ops
 from vllm.model_executor.layers.layernorm import RMSNorm
-from vllm.model_executor.layers.quantization.utils.fp8_utils import (
-    per_token_group_quant_fp8,
-)


@dataclass
@@ -25,7 +22,6 @@ class bench_params_t:
    hidden_size: int
    add_residual: bool
    dtype: torch.dtype
-    group_size: list[int]

    def description(self):
        return (
@@ -33,7 +29,6 @@ class bench_params_t:
            f"x D {self.hidden_size} "
            f"x R {self.add_residual} "
            f"x DT {self.dtype}"
-            f"x GS {self.group_size}"
        )


@@ -43,11 +38,10 @@ def get_bench_params() -> list[bench_params_t]:
    HIDDEN_SIZES = list(range(1024, 8129, 1024))
    ADD_RESIDUAL = [True, False]
    DTYPES = [torch.bfloat16, torch.float]
-    GROUP_SIZES = [[1, 64], [1, 128]]

-    combinations = product(NUM_TOKENS, HIDDEN_SIZES, ADD_RESIDUAL, DTYPES, GROUP_SIZES)
+    combinations = product(NUM_TOKENS, HIDDEN_SIZES, ADD_RESIDUAL, DTYPES)
    bench_params = list(
-        map(lambda x: bench_params_t(x[0], x[1], x[2], x[3], x[4]), combinations)
+        map(lambda x: bench_params_t(x[0], x[1], x[2], x[3]), combinations)
    )
    return bench_params

@@ -58,7 +52,6 @@ def unfused_int8_impl(
    x: torch.Tensor,
    residual: torch.Tensor | None,
    quant_dtype: torch.dtype,
-    group_size: list[int],
 ):
    # Norm
    torch_out = None
@@ -76,7 +69,6 @@ def unfused_fp8_impl(
    x: torch.Tensor,
    residual: torch.Tensor | None,
    quant_dtype: torch.dtype,
-    group_size: list[int],
 ):
    # Norm
    torch_out = None
@@ -89,63 +81,23 @@ def unfused_fp8_impl(
    torch_out, _ = ops.scaled_fp8_quant(torch_out)


-def unfused_groupwise_fp8_impl(
-    rms_norm_layer: RMSNorm,
-    x: torch.Tensor,
-    residual: torch.Tensor | None,
-    quant_dtype: torch.dtype,
-    group_size: list[int],
-):
-    # Norm
-    torch_out = None
-    if residual is None:
-        torch_out = rms_norm_layer.forward_cuda(x, residual)
-    else:
-        torch_out, _ = rms_norm_layer.forward_cuda(x, residual)
-
-    # Quant
-    torch_out, _ = per_token_group_quant_fp8(
-        torch_out, group_size=group_size[1], use_ue8m0=False
-    )
-
-
 def fused_impl(
    rms_norm_layer: RMSNorm,  # this stores the weights
    x: torch.Tensor,
    residual: torch.Tensor | None,
    quant_dtype: torch.dtype,
-    group_size: list[int],
 ):
    out, _ = ops.rms_norm_dynamic_per_token_quant(
        x, rms_norm_layer.weight, 1e-6, quant_dtype, residual=residual
    )


-def fused_groupwise_impl(
-    rms_norm_layer: RMSNorm,  # this stores the weights
-    x: torch.Tensor,
-    residual: torch.Tensor | None,
-    quant_dtype: torch.dtype,
-    group_size: list[int],
-):
-    out, _ = ops.rms_norm_per_block_quant(
-        x,
-        rms_norm_layer.weight,
-        1e-6,
-        quant_dtype,
-        group_size,
-        residual=residual,
-        is_scale_transposed=True,
-    )
-
-
 # Bench functions
 def bench_fn(
    rms_norm_layer: RMSNorm,
    x: torch.Tensor,
    residual: torch.Tensor,
    quant_dtype: torch.dtype,
-    group_size: list[int],
    label: str,
    sub_label: str,
    fn: Callable,
@@ -158,11 +110,10 @@ def bench_fn(
        "x": x,
        "residual": residual,
        "quant_dtype": quant_dtype,
-        "group_size": group_size,
        "fn": fn,
    }
    return TBenchmark.Timer(
-        stmt="fn(rms_norm_layer, x, residual, quant_dtype, group_size)",
+        stmt="fn(rms_norm_layer, x, residual, quant_dtype)",
        globals=globals,
        label=label,
        sub_label=sub_label,
@@ -196,7 +147,6 @@ def bench(params: bench_params_t, label: str, sub_label: str) -> Iterable[TMeasu
            x,
            residual,
            torch.int8,
-            params.group_size,
            label,
            sub_label,
            unfused_int8_impl,
@@ -211,7 +161,6 @@ def bench(params: bench_params_t, label: str, sub_label: str) -> Iterable[TMeasu
            x,
            residual,
            torch.float8_e4m3fn,
-            params.group_size,
            label,
            sub_label,
            unfused_fp8_impl,
@@ -226,7 +175,6 @@ def bench(params: bench_params_t, label: str, sub_label: str) -> Iterable[TMeasu
            x,
            residual,
            torch.int8,
-            params.group_size,
            label,
            sub_label,
            fused_impl,
@@ -241,7 +189,6 @@ def bench(params: bench_params_t, label: str, sub_label: str) -> Iterable[TMeasu
            x,
            residual,
            torch.float8_e4m3fn,
-            params.group_size,
            label,
            sub_label,
            fused_impl,
@@ -249,36 +196,6 @@ def bench(params: bench_params_t, label: str, sub_label: str) -> Iterable[TMeasu
        )
    )

-    # unfused groupwise fp8 impl.
-    timers.append(
-        bench_fn(
-            layer,
-            x,
-            residual,
-            torch.float8_e4m3fn,
-            params.group_size,
-            label,
-            sub_label,
-            unfused_groupwise_fp8_impl,
-            "unfused_groupwise_fp8_impl",
-        )
-    )
-
-    # fused groupwise fp8 impl.
-    timers.append(
-        bench_fn(
-            layer,
-            x,
-            residual,
-            torch.float8_e4m3fn,
-            params.group_size,
-            label,
-            sub_label,
-            fused_groupwise_impl,
-            "fused_groupwise_fp8_impl",
-        )
-    )
-
    print_timers(timers)

    return timers
@@ -1,18 +1,10 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project

-import os
-
-# Disable DeepGEMM for this benchmark to use CUTLASS
-os.environ["VLLM_USE_DEEP_GEMM"] = "0"
-
 import torch

 from vllm.model_executor.layers.quantization.utils.fp8_utils import (
-    W8A8BlockFp8LinearOp,
-)
-from vllm.model_executor.layers.quantization.utils.quant_utils import (
-    GroupShape,
+    apply_w8a8_block_fp8_linear,
 )
 from vllm.model_executor.layers.quantization.utils.w8a8_utils import (
    CUTLASS_BLOCK_FP8_SUPPORTED,
@@ -47,14 +39,13 @@ def build_w8a8_block_fp8_runner(M, N, K, block_size, device, use_cutlass):
    fp8_info = torch.finfo(torch.float8_e4m3fn)
    fp8_max, fp8_min = fp8_info.max, fp8_info.min

-    # Create random input tensor (bfloat16, will be quantized by W8A8BlockFp8LinearOp)
+    # Create random FP8 tensors
    A_ref = (torch.rand(M, K, dtype=torch.bfloat16, device=device) - 0.5) * 2 * fp8_max

-    # Create quantized weight tensor
    B_ref = (torch.rand(N, K, dtype=torch.bfloat16, device=device) - 0.5) * 2 * fp8_max
    B = B_ref.clamp(min=fp8_min, max=fp8_max).to(torch.float8_e4m3fn)

-    # Create weight scales
+    # Create scales
    block_n, block_k = block_size[0], block_size[1]
    n_tiles = (N + block_n - 1) // block_n
    k_tiles = (K + block_k - 1) // block_k
@@ -64,25 +55,19 @@ def build_w8a8_block_fp8_runner(M, N, K, block_size, device, use_cutlass):
        * factor_for_scale
    )

-    # Create W8A8BlockFp8LinearOp instance
-    weight_group_shape = GroupShape(block_n, block_k)
-    act_quant_group_shape = GroupShape(1, block_k)  # Per-token, per-group quantization
-
-    linear_op = W8A8BlockFp8LinearOp(
-        weight_group_shape=weight_group_shape,
-        act_quant_group_shape=act_quant_group_shape,
-        cutlass_block_fp8_supported=use_cutlass,
-        use_aiter_and_is_supported=False,
-    )
+    # SM90 CUTLASS requires row-major format for scales
+    if use_cutlass and current_platform.is_device_capability(90):
+        Bs = Bs.T.contiguous()

    def run():
-        return linear_op.apply(
-            input=A_ref,
-            weight=B,
-            weight_scale=Bs,
-            input_scale=None,
-            bias=None,
-        )
+        if use_cutlass:
+            return apply_w8a8_block_fp8_linear(
+                A_ref, B, block_size, Bs, cutlass_block_fp8_supported=True
+            )
+        else:
+            return apply_w8a8_block_fp8_linear(
+                A_ref, B, block_size, Bs, cutlass_block_fp8_supported=False
+            )

    return run

@@ -1,177 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-import argparse
-import copy
-import itertools
-
-import torch
-from weight_shapes import WEIGHT_SHAPES
-
-from vllm import _custom_ops as ops
-from vllm.platforms import current_platform
-from vllm.scalar_type import scalar_types
-from vllm.triton_utils import triton
-from vllm.utils.flashinfer import flashinfer_fp4_quantize
-
-if not current_platform.has_device_capability(100):
-    raise RuntimeError("NVFP4 requires compute capability of 10.0 (Blackwell)")
-
-FLOAT4_E2M1_MAX = scalar_types.float4_e2m1f.max()
-FLOAT8_E4M3_MAX = torch.finfo(torch.float8_e4m3fn).max
-
-PROVIDER_CFGS = {
-    "vllm": dict(backend="vllm", enabled=True),
-    "flashinfer": dict(backend="flashinfer", enabled=True),
-}
-
-_enabled = [k for k, v in PROVIDER_CFGS.items() if v["enabled"]]
-
-
-def compute_global_scale(tensor: torch.Tensor) -> torch.Tensor:
-    """Compute global scale for FP4 quantization."""
-    amax = torch.abs(tensor).max().to(torch.float32)
-    return FLOAT8_E4M3_MAX * FLOAT4_E2M1_MAX / amax
-
-
-@triton.testing.perf_report(
-    triton.testing.Benchmark(
-        x_names=["batch_size"],
-        x_vals=[1, 16, 32, 64, 128, 256, 512, 1024, 2048, 4096],
-        x_log=False,
-        line_arg="provider",
-        line_vals=_enabled,
-        line_names=_enabled,
-        ylabel="us (lower is better)",
-        plot_name="NVFP4 Input Quantization Latency (us)",
-        args={},
-    )
-)
-def benchmark(batch_size, provider, N, K):
-    M = batch_size
-    device = "cuda"
-    dtype = torch.bfloat16
-
-    # Create input tensor
-    a = torch.randn((M, K), device=device, dtype=dtype)
-
-    # Compute global scale for activation
-    a_global_scale = compute_global_scale(a)
-
-    quantiles = [0.5, 0.2, 0.8]
-
-    cfg = PROVIDER_CFGS[provider]
-
-    if cfg["backend"] == "vllm":
-        # vLLM's FP4 quantization
-        ms, min_ms, max_ms = triton.testing.do_bench_cudagraph(
-            lambda: ops.scaled_fp4_quant(a, a_global_scale),
-            quantiles=quantiles,
-        )
-    elif cfg["backend"] == "flashinfer":
-        # FlashInfer's FP4 quantization
-        # Use is_sf_swizzled_layout=True to match vLLM's output format
-        ms, min_ms, max_ms = triton.testing.do_bench_cudagraph(
-            lambda: flashinfer_fp4_quantize(
-                a, a_global_scale, is_sf_swizzled_layout=True
-            ),
-            quantiles=quantiles,
-        )
-
-    # Convert ms to us for better readability at small batch sizes
-    to_us = lambda t_ms: t_ms * 1000
-    return to_us(ms), to_us(max_ms), to_us(min_ms)
-
-
-def prepare_shapes(args):
-    out = []
-    for model, tp_size in itertools.product(args.models, args.tp_sizes):
-        for KN, tp_dim in copy.deepcopy(WEIGHT_SHAPES[model]):
-            KN[tp_dim] //= tp_size
-            KN.append(model)
-            out.append(KN)
-    return out
-
-
-def _test_accuracy_once(M: int, K: int, dtype: torch.dtype, device: str):
-    """Test accuracy between vLLM and FlashInfer FP4 quantization."""
-    # Create input tensor
-    a = torch.randn((M, K), device=device, dtype=dtype)
-
-    # Compute global scale
-    a_global_scale = compute_global_scale(a)
-
-    # vLLM quantization
-    vllm_fp4, vllm_scale = ops.scaled_fp4_quant(a, a_global_scale)
-
-    # FlashInfer quantization (with swizzled layout to match vLLM's output)
-    flashinfer_fp4, flashinfer_scale = flashinfer_fp4_quantize(
-        a, a_global_scale, is_sf_swizzled_layout=True
-    )
-    flashinfer_scale = flashinfer_scale.view(torch.float8_e4m3fn)
-
-    # Compare outputs
-    torch.testing.assert_close(
-        vllm_fp4,
-        flashinfer_fp4,
-    )
-    print(f"M={M}, K={K}, dtype={dtype}: PASSED")
-
-
-def test_accuracy():
-    """Run accuracy tests across various shapes."""
-    print("\n" + "=" * 60)
-    print("Running accuracy tests: vLLM vs FlashInfer")
-    print("=" * 60)
-
-    device = "cuda"
-    dtype = torch.bfloat16
-
-    # Test various batch sizes and hidden dimensions
-    Ms = [1, 1024]
-    Ks = [4096]
-
-    for M in Ms:
-        for K in Ks:
-            _test_accuracy_once(M, K, dtype, device)
-
-    print("\nAll accuracy tests passed!")
-
-
-if __name__ == "__main__":
-    parser = argparse.ArgumentParser(
-        description="Benchmark NVFP4 quantization: vLLM vs FlashInfer"
-    )
-    parser.add_argument(
-        "--models",
-        nargs="+",
-        type=str,
-        default=["meta-llama/Llama-3.1-8B-Instruct"],
-        choices=list(WEIGHT_SHAPES.keys()),
-    )
-    parser.add_argument("--tp-sizes", nargs="+", type=int, default=[1])
-    parser.add_argument(
-        "--save-path",
-        type=str,
-        default=None,
-        help="Path to save benchmark results",
-    )
-    parser.add_argument(
-        "--accuracy",
-        action="store_true",
-        help="Run accuracy tests",
-    )
-    args = parser.parse_args()
-
-    if args.accuracy:
-        test_accuracy()
-
-    for K, N, model in prepare_shapes(args):
-        print(f"\n{model}, N={N} K={K}")
-        benchmark.run(
-            print_data=True,
-            save_path=args.save_path,
-            N=N,
-            K=K,
-        )
-
-    print("\nBenchmark finished!")
@@ -1,244 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-
-from dataclasses import dataclass
-from enum import Enum
-from itertools import product
-from typing import Any
-
-import torch
-import torch.utils.benchmark as TBenchmark
-from torch.utils.benchmark import Measurement as TMeasurement
-
-from vllm.model_executor.layers.quantization.utils.fp8_utils import (
-    _per_token_group_quant_fp8_colmajor,
-    silu_mul_per_token_group_quant_fp8_colmajor,
-)
-from vllm.triton_utils import triton
-from vllm.utils.deep_gemm import is_deep_gemm_e8m0_used
-
-from .utils import ArgPool, Bench, CudaGraphBenchParams
-
-GROUP_SIZE = 128
-FLOAT8_T = torch.float8_e4m3fn
-
-
-def print_timers(timers: list[TMeasurement], cuda_graph_nops: int):
-    print(
-        f"Note : The timings reported above is for {cuda_graph_nops} "
-        "consecutive invocations of the benchmarking functions. "
-        f"Please divide by {cuda_graph_nops} for single invocation "
-        "timings."
-    )
-    compare = TBenchmark.Compare(timers)
-    compare.print()
-
-
-class ImplType(Enum):
-    SILU_MUL_PER_TOKEN_GROUP_QUANT_FP8_COLMAJOR = 1
-    REFERENCE = 2
-
-    def get_impl(self):
-        if self == ImplType.SILU_MUL_PER_TOKEN_GROUP_QUANT_FP8_COLMAJOR:
-            return silu_mul_per_token_group_quant_fp8_colmajor
-        elif self == ImplType.REFERENCE:
-            return reference
-        raise ValueError(f"Unrecognized ImplType {self}")
-
-
-@dataclass
-class BenchmarkTensors:
-    input: torch.Tensor
-    output: torch.Tensor
-
-    # Reference act output tensor
-    ref_act_out: torch.Tensor
-    ref_quant_out: torch.Tensor
-
-    @staticmethod
-    def make(T: int, N: int) -> "BenchmarkTensors":
-        assert T % GROUP_SIZE == 0
-        assert N % (GROUP_SIZE * 2) == 0
-
-        input = torch.rand((T, N), dtype=torch.bfloat16, device="cuda")
-
-        # silu_mul_per_token_group_quant_fp8_colmajor output.
-        output = torch.rand((T, N // 2), dtype=torch.bfloat16, device="cuda").to(
-            FLOAT8_T
-        )
-
-        # reference output.
-        ref_act_out = torch.empty((T, N // 2), dtype=torch.bfloat16, device="cuda")
-        ref_quant_out = torch.empty(
-            (T, N // 2), dtype=torch.bfloat16, device="cuda"
-        ).to(FLOAT8_T)
-
-        return BenchmarkTensors(
-            input=input,
-            output=output,
-            ref_act_out=ref_act_out,
-            ref_quant_out=ref_quant_out,
-        )
-
-    @property
-    def T(self):
-        return self.input.size(0)
-
-    @property
-    def N(self):
-        return self.input.size(1)
-
-    def make_impl_kwargs(self, impl_type: ImplType) -> dict[str, Any]:
-        if impl_type == ImplType.SILU_MUL_PER_TOKEN_GROUP_QUANT_FP8_COLMAJOR:
-            return {
-                "input": self.input,
-                "output": self.output,
-                "use_ue8m0": is_deep_gemm_e8m0_used(),
-            }
-        elif impl_type == ImplType.REFERENCE:
-            return {
-                "input": self.input,
-                "act_out": self.ref_act_out,
-                "quant_out": self.ref_quant_out,
-                "use_ue8m0": is_deep_gemm_e8m0_used(),
-            }
-        raise ValueError(f"Unrecognized impl_type {impl_type}")
-
-
-def reference_quant(x: torch.Tensor, quant_out: torch.Tensor, use_ue8m0: bool):
-    """
-    Reference triton quant kernel from,
-    vllm.model_executor.layers.quantization.utils.fp8_utils
-    """
-    assert quant_out.size() == x.size()
-    # Allocate the scale tensor column-major format.
-    shape = (x.shape[-1] // GROUP_SIZE,) + x.shape[:-1]
-    x_q = quant_out
-    x_s = torch.empty(shape, device=x.device, dtype=torch.float32).permute(-1, -2)
-
-    M = x.numel() // GROUP_SIZE
-    N = GROUP_SIZE
-    BLOCK = triton.next_power_of_2(N)
-    # heuristics for number of warps
-    num_warps = min(max(BLOCK // 256, 1), 8)
-    num_stages = 1
-
-    finfo = torch.finfo(FLOAT8_T)
-    fp8_min = finfo.min
-    fp8_max = finfo.max
-
-    _per_token_group_quant_fp8_colmajor[(M,)](
-        x,
-        x_q,
-        x_s,
-        GROUP_SIZE,
-        x.shape[1],
-        x.stride(0),
-        x_s.stride(1),
-        eps=1e-10,
-        fp8_min=fp8_min,
-        fp8_max=fp8_max,
-        use_ue8m0=use_ue8m0,
-        BLOCK=BLOCK,
-        num_warps=num_warps,
-        num_stages=num_stages,
-    )
-    return x_q, x_s
-
-
-def reference(
-    input: torch.Tensor,
-    act_out: torch.Tensor,
-    quant_out: torch.Tensor,
-    use_ue8m0: bool,
-) -> tuple[torch.Tensor, torch.Tensor]:
-    torch.ops._C.silu_and_mul(act_out, input)
-    return reference_quant(act_out, quant_out, use_ue8m0)
-
-
-def bench_impl(
-    bench_tensors: list[BenchmarkTensors], impl_type: ImplType
-) -> TMeasurement:
-    T = bench_tensors[0].T
-    N = bench_tensors[0].N
-
-    arg_pool_size = len(bench_tensors)
-    kwargs_list = [bt.make_impl_kwargs(impl_type) for bt in bench_tensors]
-
-    # warmup
-    for kwargs in kwargs_list:
-        impl_type.get_impl()(**kwargs)
-    torch.cuda.synchronize()
-
-    # Merge into a single kwargs and qualify arguments as ArgPool
-    kwargs = {k: ArgPool([]) for k in kwargs_list[0]}
-    for _kwargs in kwargs_list:
-        for k, v in _kwargs.items():
-            kwargs[k].values.append(v)
-
-    cuda_graph_params = None
-    cuda_graph_params = CudaGraphBenchParams(arg_pool_size)
-    timer = None
-    with Bench(
-        cuda_graph_params,
-        "silu-mul-quant",
-        f"num_tokens={T}, N={N}",
-        impl_type.name,
-        impl_type.get_impl(),
-        **kwargs,
-    ) as bench:
-        timer = bench.run()
-    return timer
-
-
-def test_correctness(T: int, N: int):
-    print(f"Testing num_tokens={T}, N={N} ...")
-
-    bench_tensor = BenchmarkTensors.make(T, N)
-
-    def output_from_impl(impl: ImplType) -> tuple[torch.Tensor, torch.Tensor]:
-        return impl.get_impl()(**bench_tensor.make_impl_kwargs(impl))
-
-    # reference output
-    ref_out_q, ref_out_s = output_from_impl(ImplType.REFERENCE)
-
-    # test ouptut
-    out_q, out_s = output_from_impl(
-        ImplType.SILU_MUL_PER_TOKEN_GROUP_QUANT_FP8_COLMAJOR
-    )
-
-    torch.testing.assert_close(ref_out_q.to(torch.float32), out_q.to(torch.float32))
-    torch.testing.assert_close(ref_out_s, out_s)
-
-
-def run(Ts: list[int], Ns: list[int], arg_pool_size: int) -> list[TMeasurement]:
-    timers = []
-    for N, T in product(Ns, Ts):
-        test_correctness(T, N)
-
-        bench_tensors: list[BenchmarkTensors] = [
-            BenchmarkTensors.make(T, N) for _ in range(arg_pool_size)
-        ]
-
-        silu_mul_quant_timer = bench_impl(
-            bench_tensors, ImplType.SILU_MUL_PER_TOKEN_GROUP_QUANT_FP8_COLMAJOR
-        )
-        timers.append(silu_mul_quant_timer)
-        reference_timer = bench_impl(bench_tensors, ImplType.REFERENCE)
-        timers.append(reference_timer)
-
-        print_timers(
-            [silu_mul_quant_timer, reference_timer], cuda_graph_nops=arg_pool_size
-        )
-
-    print_timers(timers, cuda_graph_nops=arg_pool_size)
-
-    return timers
-
-
-if __name__ == "__main__":
-    T = [128 * i for i in range(1, 16)] + [2048 * i for i in range(1, 65)]
-    N = [2048, 4096, 8192]
-
-    print(f"T = {T}, N = {N}")
-    run(T, N, arg_pool_size=8)
@@ -13,8 +13,8 @@ from vllm.triton_utils import triton
 from vllm.utils.argparse_utils import FlexibleArgumentParser
 from vllm.utils.torch_utils import STR_DTYPE_TO_TORCH_DTYPE

-batch_size_range = [1, 16, 128]
-seq_len_range = [1, 16, 64, 1024, 4096]
+batch_size_range = [1, 16, 32, 64, 128]
+seq_len_range = [1, 16, 64, 128, 256, 512, 1024, 2048, 4096]
 intermediate_size = [3072, 9728, 12288]
 configs = list(itertools.product(batch_size_range, seq_len_range, intermediate_size))

@@ -255,8 +255,8 @@ def bench_run(
        torch.cuda.synchronize()

        # Timing
-        start_event = torch.Event(enable_timing=True)
-        end_event = torch.Event(enable_timing=True)
+        start_event = torch.cuda.Event(enable_timing=True)
+        end_event = torch.cuda.Event(enable_timing=True)

        latencies = []
        for _ in range(num_iters):
@@ -293,7 +293,7 @@ class CommunicatorBenchmark:
                    graph = torch.cuda.CUDAGraph()
                    graph_pool = torch.cuda.graph_pool_handle()
                    set_graph_pool_id(graph_pool)
-                    with torch.cuda.graph(graph, pool=graph_pool, stream=stream):
+                    with torch.cuda.graph(graph, pool=graph_pool):
                        for _ in range(CUDA_GRAPH_CAPTURE_CYCLES):
                            allreduce_fn(graph_input)

@@ -16,8 +16,8 @@ from vllm.model_executor.layers.fused_moe.fused_moe import (
 from vllm.utils.argparse_utils import FlexibleArgumentParser

 DEFAULT_MODELS = [
-    "mistralai/Mixtral-8x7B-Instruct-v0.1",
-    "deepseek-ai/DeepSeek-V2-Lite",
+    "nm-testing/Mixtral-8x7B-Instruct-v0.1",
+    "nm-testing/deepseekv2-lite",
    "ibm-granite/granite-3.0-1b-a400m",
    "ibm-granite/granite-3.0-3b-a800m",
 ]
@@ -237,7 +237,6 @@ def marlin_create_bench_fn(bt: BenchmarkTensors) -> Callable:
            b_q_weight=w_q,
            b_bias=None,
            b_scales=w_s,
-            a_scales=None,
            global_scale=None,
            b_zeros=w_zp,
            g_idx=g_idx,
@@ -263,7 +263,7 @@ def bench_run(

    results.append(
        benchmark.Timer(
-            stmt="output = gptq_marlin_gemm(a, None, marlin_q_w, marlin_s, None, marlin_s2, marlin_zp, marlin_g_idx, marlin_sort_indices, marlin_workspace.scratch, quant_type, size_m, size_n, size_k, is_k_full, False, False, False)",  # noqa: E501
+            stmt="output = gptq_marlin_gemm(a, None, marlin_q_w, marlin_s, marlin_s2, marlin_zp, marlin_g_idx, marlin_sort_indices, marlin_workspace.scratch, quant_type, size_m, size_n, size_k, is_k_full, False, False, False)",  # noqa: E501
            globals=globals,
            label=label,
            sub_label=sub_label,
@@ -273,7 +273,7 @@ def bench_run(

    results.append(
        benchmark.Timer(
-            stmt="output = gptq_marlin_gemm(a, None, marlin_q_w, marlin_s, None, marlin_s2, marlin_zp, marlin_g_idx, marlin_sort_indices, marlin_workspace.scratch, quant_type, size_m, size_n, size_k, is_k_full, False, True, False)",  # noqa: E501
+            stmt="output = gptq_marlin_gemm(a, None, marlin_q_w, marlin_s, marlin_s2, marlin_zp, marlin_g_idx, marlin_sort_indices, marlin_workspace.scratch, quant_type, size_m, size_n, size_k, is_k_full, False, True, False)",  # noqa: E501
            globals=globals,
            label=label,
            sub_label=sub_label,
@@ -1,150 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-"""
-Benchmark script comparing torch.cat vs direct copy for k_nope/k_pe concatenation
-in MLA (Multi-head Latent Attention) prefill.
-
-This validates that the optimization from commit 8d4142bd is beneficial across
-various batch sizes, not just the originally tested batch size of 32768.
-"""
-
-import time
-from collections.abc import Callable
-
-import torch
-
-# DeepSeek-V3 MLA dimensions
-NUM_HEADS = 128
-QK_NOPE_HEAD_DIM = 128
-PE_DIM = 64
-
-
-def cat_method(k_nope: torch.Tensor, k_pe: torch.Tensor) -> torch.Tensor:
-    """Original torch.cat approach with expand."""
-    return torch.cat((k_nope, k_pe.expand((*k_nope.shape[:-1], -1))), dim=-1)
-
-
-def direct_copy_method(k_nope: torch.Tensor, k_pe: torch.Tensor) -> torch.Tensor:
-    """Optimized direct copy approach (avoids expand + cat overhead)."""
-    k = torch.empty(
-        (*k_nope.shape[:-1], k_nope.shape[-1] + k_pe.shape[-1]),
-        dtype=k_nope.dtype,
-        device=k_nope.device,
-    )
-    k[..., : k_nope.shape[-1]] = k_nope
-    k[..., k_nope.shape[-1] :] = k_pe
-    return k
-
-
-def benchmark_method(
-    method: Callable,
-    k_nope: torch.Tensor,
-    k_pe: torch.Tensor,
-    num_warmup: int = 10,
-    num_iters: int = 100,
-) -> float:
-    """Benchmark a concatenation method and return mean latency in ms."""
-    # Warmup
-    for _ in range(num_warmup):
-        _ = method(k_nope, k_pe)
-    torch.cuda.synchronize()
-
-    # Benchmark
-    start = time.perf_counter()
-    for _ in range(num_iters):
-        _ = method(k_nope, k_pe)
-    torch.cuda.synchronize()
-    end = time.perf_counter()
-
-    return (end - start) / num_iters * 1000  # Convert to ms
-
-
-@torch.inference_mode()
-def run_benchmark(dtype: torch.dtype, dtype_name: str):
-    """Run benchmark for a specific dtype."""
-    torch.set_default_device("cuda")
-
-    # Batch sizes to test (powers of 2 from 32 to 65536)
-    batch_sizes = [32, 64, 128, 256, 512, 1024, 2048, 4096, 8192, 16384, 32768, 65536]
-
-    print("=" * 80)
-    print("Benchmark: torch.cat vs direct copy for MLA k_nope/k_pe concatenation")
-    print("=" * 80)
-    print(
-        f"Tensor shapes: k_nope=[B, {NUM_HEADS}, {QK_NOPE_HEAD_DIM}], "
-        f"k_pe=[B, 1, {PE_DIM}]"
-    )
-    print(f"dtype: {dtype_name}")
-    print()
-    print(
-        f"{'Batch Size':>12} | {'cat (ms)':>10} | {'direct (ms)':>12} | "
-        f"{'Speedup':>8} | {'Reduction':>10}"
-    )
-    print("-" * 70)
-
-    results = []
-    for batch_size in batch_sizes:
-        # Create input tensors (generate in float32 then convert for FP8 compatibility)
-        k_nope = torch.randn(
-            batch_size, NUM_HEADS, QK_NOPE_HEAD_DIM, dtype=torch.float32, device="cuda"
-        ).to(dtype)
-        k_pe = torch.randn(
-            batch_size, 1, PE_DIM, dtype=torch.float32, device="cuda"
-        ).to(dtype)
-
-        # Benchmark both methods
-        cat_time = benchmark_method(cat_method, k_nope, k_pe)
-        direct_time = benchmark_method(direct_copy_method, k_nope, k_pe)
-
-        speedup = cat_time / direct_time
-        reduction = (1 - direct_time / cat_time) * 100
-
-        results.append((batch_size, cat_time, direct_time, speedup, reduction))
-
-        print(
-            f"{batch_size:>12} | {cat_time:>10.3f} | {direct_time:>12.3f} | "
-            f"{speedup:>7.2f}x | {reduction:>9.1f}%"
-        )
-
-    print("=" * 80)
-
-    # Summary statistics
-    speedups = [r[3] for r in results]
-    print("\nSpeedup summary:")
-    print(f"  Min:  {min(speedups):.2f}x")
-    print(f"  Max:  {max(speedups):.2f}x")
-    print(f"  Mean: {sum(speedups) / len(speedups):.2f}x")
-
-    # Find crossover point
-    crossover_batch = None
-    for batch_size, _, _, speedup, _ in results:
-        if speedup >= 1.0:
-            crossover_batch = batch_size
-            break
-
-    print("\nConclusion:")
-    if crossover_batch:
-        print(f"  - Direct copy becomes beneficial at batch size >= {crossover_batch}")
-    # Filter for large batches (>= 512 which is typical for prefill)
-    large_batch_speedups = [r[3] for r in results if r[0] >= 512]
-    if large_batch_speedups:
-        avg_large = sum(large_batch_speedups) / len(large_batch_speedups)
-        print(f"  - For batch sizes >= 512: avg speedup = {avg_large:.2f}x")
-    print("  - MLA prefill typically uses large batches, so optimization is effective")
-
-    return results
-
-
-@torch.inference_mode()
-def main():
-    # Test bfloat16
-    print("\n")
-    run_benchmark(torch.bfloat16, "bfloat16")
-
-    # Test float8_e4m3fn
-    print("\n")
-    run_benchmark(torch.float8_e4m3fn, "float8_e4m3fn")
-
-
-if __name__ == "__main__":
-    main()
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Duncan MossandAlexander Matveev	7ddf862297	lint Signed-off-by: Duncan Moss <djm.moss@gmail.com>	2025-11-04 19:54:45 +00:00
Duncan MossandAlexander Matveev	4e97071d84	add comments Signed-off-by: Duncan Moss <djm.moss@gmail.com>	2025-11-04 19:54:43 +00:00
Duncan MossandAlexander Matveev	f37a31d9fe	tp and ep support Signed-off-by: Duncan Moss <djm.moss@gmail.com>	2025-11-04 19:53:57 +00:00
Duncan MossandAlexander Matveev	ee45599ec8	finished integration Signed-off-by: Duncan Moss <djm.moss@gmail.com>	2025-11-04 19:52:25 +00:00
				`@@ -1 +0,0 @@`
				`Meta-Llama-4-Maverick-17B-128E-Instruct-FP8.yaml`