p

Signed-off-by: khluu <khluu000@gmail.com>
p
2026-05-01 14:47:49 -07:00 · 2026-05-01 14:28:31 -07:00 · 2026-05-01 12:24:17 -07:00 · 2026-05-01 12:23:33 -07:00 · 2026-05-01 12:17:45 -07:00 · 2026-05-01 12:17:28 -07:00
271 changed files with 2857 additions and 12637 deletions
@@ -126,4 +126,5 @@ steps:
      'cd tests &&
      pytest -v -s lora/test_default_mm_loras.py && 
      (pytest -v -s lora/test_qwen3_unembed.py || true) &&
+      (pytest -v -s lora/test_qwenvl.py || true) &&
      pytest -v -s lora/test_whisper.py'
@@ -388,18 +388,18 @@ steps:
    - python3 basic/offline_inference/embed.py
    - python3 basic/offline_inference/score.py
    # Multi-modal models
-    - python3 generate/multimodal/audio_language_offline.py --seed 0
-    - python3 generate/multimodal/vision_language_offline.py --seed 0
-    - python3 generate/multimodal/vision_language_multi_image_offline.py --seed 0
-    - python3 generate/multimodal/encoder_decoder_multimodal_offline.py --model-type whisper --seed 0
+    - 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
    # Pooling models
    - python3 pooling/embed/vision_embedding_offline.py --seed 0
    # Features demo
-    - python3 features/automatic_prefix_caching/prefix_caching_offline.py
+    - 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 features/speculative_decoding/spec_decode_offline.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
-    - python3 features/speculative_decoding/spec_decode_offline.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
+    - 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
+    - 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

 #----------------------------------------------------------  mi250 · kernels  ----------------------------------------------------------#

@@ -1168,13 +1168,13 @@ steps:
  - vllm/v1/attention/backends/
  - vllm/v1/attention/selector.py
  - tests/distributed/test_context_parallel.py
-  - examples/features/data_parallel/data_parallel_offline.py
+  - examples/offline_inference/data_parallel.py
  - vllm/_aiter_ops.py
  - vllm/platforms/rocm.py
  commands:
  - export TORCH_NCCL_BLOCKING_WAIT=1
  - pytest -v -s tests/distributed/test_context_parallel.py
-  - VLLM_LOGGING_LEVEL=DEBUG python3 examples/features/data_parallel/data_parallel_offline.py --model=Qwen/Qwen1.5-MoE-A2.7B -tp=1 -dp=2 --max-model-len=2048 --all2all-backend=allgather_reducescatter --disable-nccl-for-dp-synchronization
+  - 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=allgather_reducescatter --disable-nccl-for-dp-synchronization

 - label: Distributed Tests (4xA100-4xMI300) # TBD
  timeout_in_minutes: 180
@@ -1203,7 +1203,7 @@ steps:
  - tests/distributed/test_torchrun_example.py
  - tests/distributed/test_torchrun_example_moe.py
  - examples/rl/
-  - tests/examples/features/data_parallel/data_parallel_offline.py
+  - tests/examples/offline_inference/data_parallel.py
  - vllm/platforms/rocm.py
  commands:
  - export TORCH_NCCL_BLOCKING_WAIT=1
@@ -1213,7 +1213,7 @@ steps:
  - PP_SIZE=2 TP_SIZE=2 torchrun --nproc-per-node=4 distributed/test_torchrun_example_moe.py
  - DP_SIZE=4 ENABLE_EP=1 torchrun --nproc-per-node=4 distributed/test_torchrun_example_moe.py
  - TP_SIZE=2 DP_SIZE=2 ENABLE_EP=1 torchrun --nproc-per-node=4 distributed/test_torchrun_example_moe.py
-  - python3 ../examples/features/data_parallel/data_parallel_offline.py --enforce-eager
+  - python3 ../examples/offline_inference/data_parallel.py --enforce-eager
  # rlhf examples
  - VLLM_ALLOW_INSECURE_SERIALIZATION=1 python3 ../examples/rl/rlhf_nccl.py
  - VLLM_ALLOW_INSECURE_SERIALIZATION=1 python3 ../examples/rl/rlhf_ipc.py
@@ -1266,7 +1266,7 @@ steps:
  optional: true
  working_dir: "/vllm-workspace/tests"
  source_file_dependencies:
-  - examples/features/torchrun/torchrun_dp_example_offline.py
+  - examples/offline_inference/torchrun_dp_example.py
  - vllm/config/parallel.py
  - vllm/distributed/
  - vllm/v1/engine/llm_engine.py
@@ -1275,7 +1275,7 @@ steps:
  - vllm/platforms/rocm.py
  commands:
  - export TORCH_NCCL_BLOCKING_WAIT=1
-  - torchrun --nproc-per-node=8 ../examples/features/torchrun/torchrun_dp_example_offline.py --tp-size=2 --pp-size=1 --dp-size=4 --enable-ep
+  - torchrun --nproc-per-node=8 ../examples/offline_inference/torchrun_dp_example.py --tp-size=2 --pp-size=1 --dp-size=4 --enable-ep

 #--------------------------------------------------------  mi300 · entrypoints  --------------------------------------------------------#

@@ -1647,18 +1647,18 @@ steps:
    - python3 basic/offline_inference/embed.py
    - python3 basic/offline_inference/score.py
    # Multi-modal models
-    - python3 generate/multimodal/audio_language_offline.py --seed 0
-    - python3 generate/multimodal/vision_language_offline.py --seed 0
-    - python3 generate/multimodal/vision_language_multi_image_offline.py --seed 0
-    - python3 generate/multimodal/encoder_decoder_multimodal_offline.py --model-type whisper --seed 0
+    - 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
    # Pooling models
    - python3 pooling/embed/vision_embedding_offline.py --seed 0
    # Features demo
-    - python3 features/automatic_prefix_caching/prefix_caching_offline.py
+    - 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 features/speculative_decoding/spec_decode_offline.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
-    - python3 features/speculative_decoding/spec_decode_offline.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
+    - 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
+    - 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

 #----------------------------------------------------------  mi300 · kernels  ----------------------------------------------------------#

@@ -1951,8 +1951,8 @@ steps:
  - pytest -v -s tests/models/multimodal/processing/
  - pytest -v -s tests/models/multimodal/test_mapping.py
  - python3 examples/basic/offline_inference/chat.py
-  - python3 examples/generate/multimodal/vision_language_offline.py --model-type qwen2_5_vl
-  - VLLM_WORKER_MULTIPROC_METHOD=spawn python3 examples/generate/multimodal/audio_language_offline.py --model-type whisper
+  - python3 examples/offline_inference/vision_language.py --model-type qwen2_5_vl
+  - VLLM_WORKER_MULTIPROC_METHOD=spawn python3 examples/offline_inference/audio_language.py --model-type whisper

 #-------------------------------------------------------  mi300 · quantization  --------------------------------------------------------#

@@ -2302,7 +2302,7 @@ steps:
  commands:
  - export TORCH_NCCL_BLOCKING_WAIT=1
  - VLLM_ALLOW_INSECURE_SERIALIZATION=1 python3 examples/rl/rlhf_async_new_apis.py
-  - VLLM_LOGGING_LEVEL=DEBUG python3 examples/features/data_parallel/data_parallel_offline.py --model=Qwen/Qwen1.5-MoE-A2.7B -tp=1 -dp=2 --max-model-len=2048 --all2all-backend=deepep_high_throughput
+  - 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
  - VLLM_ALLOW_INSECURE_SERIALIZATION=1 pytest -v -s tests/distributed/test_weight_transfer.py
  - pytest -v -s tests/distributed/test_packed_tensor.py
@@ -2713,7 +2713,7 @@ steps:
  - vllm/v1/attention/selector.py
  - tests/distributed/test_context_parallel.py
  - tests/v1/distributed/test_dbo.py
-  - examples/features/data_parallel/data_parallel_offline.py
+  - examples/offline_inference/data_parallel.py
  - vllm/_aiter_ops.py
  - vllm/platforms/rocm.py
  commands:
@@ -2930,18 +2930,18 @@ steps:
  - python3 basic/offline_inference/embed.py
  - python3 basic/offline_inference/score.py
  # Multi-modal models
-  - python3 generate/multimodal/audio_language_offline.py --seed 0
-  - python3 generate/multimodal/vision_language_offline.py --seed 0
-  - python3 generate/multimodal/vision_language_multi_image_offline.py --seed 0
-  - python3 generate/multimodal/encoder_decoder_multimodal_offline.py --model-type whisper --seed 0
+  - 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
  # Pooling models
  - python3 pooling/embed/vision_embedding_offline.py --seed 0
  # Features demo
-  - python3 features/automatic_prefix_caching/prefix_caching_offline.py
+  - 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 features/speculative_decoding/spec_decode_offline.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
-  - python3 features/speculative_decoding/spec_decode_offline.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
+  - 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
+  - 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

 #----------------------------------------------------------  mi355 · kernels  ----------------------------------------------------------#

@@ -88,8 +88,9 @@ steps:
  - vllm/distributed/
  - tests/distributed/test_torchrun_example.py
  - tests/distributed/test_torchrun_example_moe.py
+  - examples/offline_inference/rlhf_colocate.py
  - examples/rl/
-  - tests/examples/features/data_parallel/data_parallel_offline.py
+  - tests/examples/offline_inference/data_parallel.py
  commands:
  # https://github.com/NVIDIA/nccl/issues/1838
  - export NCCL_CUMEM_HOST_ENABLE=0
@@ -106,7 +107,7 @@ steps:
  # test with torchrun tp=2 and dp=2 with ep
  - TP_SIZE=2 DP_SIZE=2 ENABLE_EP=1 torchrun --nproc-per-node=4 tests/distributed/test_torchrun_example_moe.py
  # test with internal dp
-  - python3 examples/features/data_parallel/data_parallel_offline.py --enforce-eager
+  - python3 examples/offline_inference/data_parallel.py --enforce-eager
  # rlhf examples
  - VLLM_ALLOW_INSECURE_SERIALIZATION=1 python3 examples/rl/rlhf_nccl.py
  - VLLM_ALLOW_INSECURE_SERIALIZATION=1 python3 examples/rl/rlhf_ipc.py
@@ -158,7 +159,7 @@ steps:
  num_devices: 8
  working_dir: "/vllm-workspace/tests"
  source_file_dependencies:
-  - examples/features/torchrun/torchrun_dp_example_offline.py
+  - examples/offline_inference/torchrun_dp_example.py
  - vllm/config/parallel.py
  - vllm/distributed/
  - vllm/v1/engine/llm_engine.py
@@ -168,7 +169,7 @@ steps:
  # 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/features/torchrun/torchrun_dp_example_offline.py --tp-size=2 --pp-size=1 --dp-size=4 --enable-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)
  device: a100
@@ -193,7 +194,7 @@ steps:
  commands:
    - pytest -v -s tests/distributed/test_context_parallel.py
    - VLLM_ALLOW_INSECURE_SERIALIZATION=1 python3 examples/rl/rlhf_async_new_apis.py
-    - VLLM_USE_DEEP_GEMM=1 VLLM_LOGGING_LEVEL=DEBUG python3 examples/features/data_parallel/data_parallel_offline.py --model=Qwen/Qwen1.5-MoE-A2.7B -tp=1 -dp=2 --max-model-len=2048 --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=1 -dp=2 --max-model-len=2048 --all2all-backend=deepep_high_throughput
    - pytest -v -s tests/v1/distributed/test_dbo.py
    - VLLM_ALLOW_INSECURE_SERIALIZATION=1 pytest -v -s tests/distributed/test_weight_transfer.py
    - pytest -v -s tests/distributed/test_packed_tensor.py
@@ -221,9 +222,9 @@ steps:
  - vllm/executor/
  - vllm/model_executor/models/
  - tests/distributed/
-  - tests/examples/features/data_parallel/data_parallel_offline.py
+  - tests/examples/offline_inference/data_parallel.py
  commands:
-    - ./.buildkite/scripts/run-multi-node-test.sh /vllm-workspace/tests 2 2 $IMAGE_TAG "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/features/data_parallel/data_parallel_offline.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/features/data_parallel/data_parallel_offline.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"
+    - ./.buildkite/scripts/run-multi-node-test.sh /vllm-workspace/tests 2 2 $IMAGE_TAG "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: Pipeline + Context Parallelism (4 GPUs)
  timeout_in_minutes: 60
@@ -95,13 +95,11 @@ steps:
  - tests/kernels/moe/test_deepgemm.py
  - tests/kernels/moe/test_batched_deepgemm.py
  - tests/kernels/attention/test_deepgemm_attention.py
-  - tests/quantization/test_cutlass_w4a16.py
  commands:
    - pytest -v -s kernels/quantization/test_block_fp8.py
    - 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
-    - pytest -v -s quantization/test_cutlass_w4a16.py

 - label: Kernels (B200)
  timeout_in_minutes: 30
@@ -113,19 +113,19 @@ steps:
    - python3 basic/offline_inference/embed.py
    - python3 basic/offline_inference/score.py
    # for multi-modal models
-    - python3 generate/multimodal/audio_language_offline.py --seed 0
-    - python3 generate/multimodal/vision_language_offline.py --seed 0
-    - python3 generate/multimodal/vision_language_multi_image_offline.py --seed 0
-    - python3 generate/multimodal/encoder_decoder_multimodal_offline.py --model-type whisper --seed 0
+    - 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/embed/vision_embedding_offline.py --seed 0
    # for features demo
-    - python3 features/automatic_prefix_caching/prefix_caching_offline.py
+    - 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 features/speculative_decoding/spec_decode_offline.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
+    - 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 features/speculative_decoding/spec_decode_offline.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
+    - 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
@@ -31,9 +31,8 @@ steps:
    - vllm/v1/worker/gpu/
    - vllm/v1/core/sched/
    - vllm/v1/worker/gpu_worker.py
+    - examples/offline_inference/
    - examples/basic/offline_inference/
-    - examples/generate/multimodal/
-    - examples/features/
    - examples/pooling/embed/vision_embedding_offline.py
    - examples/others/tensorize_vllm_model.py
  commands:
@@ -45,19 +44,19 @@ steps:
    #- python3 basic/offline_inference/generate.py --model meta-llama/Llama-2-13b-chat-hf --cpu-offload-gb 10  # TODO
    #- python3 basic/offline_inference/embed.py   # TODO
    # for multi-modal models
-    - python3 generate/multimodal/audio_language_offline.py --seed 0
-    - python3 generate/multimodal/vision_language_offline.py --seed 0
-    - python3 generate/multimodal/vision_language_multi_image_offline.py --seed 0
-    - python3 generate/multimodal/encoder_decoder_multimodal_offline.py --model-type whisper --seed 0
+    - 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/embed/vision_embedding_offline.py --seed 0
    # for features demo
-    - python3 features/automatic_prefix_caching/prefix_caching_offline.py
+    - 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 features/speculative_decoding/spec_decode_offline.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
+    - 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 features/speculative_decoding/spec_decode_offline.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
+    - 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: Model Runner V2 Distributed (2 GPUs)
  timeout_in_minutes: 45
@@ -69,9 +69,9 @@ steps:
    - pytest -v -s tests/models/multimodal/processing/
    - pytest -v -s tests/models/multimodal/test_mapping.py
    - python3 examples/basic/offline_inference/chat.py
-    - python3 examples/generate/multimodal/vision_language_offline.py --model-type qwen2_5_vl
+    - 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/generate/multimodal/audio_language_offline.py --model-type whisper
+    - VLLM_WORKER_MULTIPROC_METHOD=spawn python3 examples/offline_inference/audio_language.py --model-type whisper

 - label: Transformers Backward Compatibility Models Test
  working_dir: "/vllm-workspace/"
@@ -83,7 +83,7 @@ steps:
    - 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/basic/offline_inference/chat.py
-    - python3 examples/generate/multimodal/vision_language_offline.py --model-type qwen2_5_vl
+    - 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/generate/multimodal/audio_language_offline.py --model-type whisper
+    - VLLM_WORKER_MULTIPROC_METHOD=spawn python3 examples/offline_inference/audio_language.py --model-type whisper
@@ -308,7 +308,8 @@ pull_request_rules:
      - files=benchmarks/benchmark_serving_structured_output.py
      - files=benchmarks/run_structured_output_benchmark.sh
      - files=docs/features/structured_outputs.md
-      - files=^examples/features/structured_outputs/
+      - files=examples/offline_inference/structured_outputs.py
+      - files=examples/online_serving/structured_outputs/structured_outputs.py
      - files~=^tests/v1/structured_output/
      - files=tests/entrypoints/llm/test_struct_output_generate.py
      - files~=^vllm/v1/structured_output/
@@ -324,7 +325,7 @@ pull_request_rules:
    - or:
      - files~=^vllm/v1/spec_decode/
      - files~=^tests/v1/spec_decode/
-      - files=^examples/features/speculative_decoding/
+      - files~=^examples/.*(spec_decode|mlpspeculator|eagle|speculation).*\.py
      - files~=^vllm/model_executor/models/.*eagle.*\.py
      - files=vllm/model_executor/models/mlp_speculator.py
      - files~=^vllm/transformers_utils/configs/(eagle|medusa|mlp_speculator)\.py
@@ -388,7 +389,11 @@ pull_request_rules:
      - files~=^tests/entrypoints/anthropic/.*tool.*
      - files~=^vllm/tool_parsers/
      - files=docs/features/tool_calling.md
-      - files~=^examples/tool_calling/
+      - files~=^examples/tool_chat_*
+      - files=examples/offline_inference/chat_with_tools.py
+      - files=examples/online_serving/openai_chat_completion_client_with_tools_required.py
+      - files=examples/online_serving/openai_chat_completion_tool_calls_with_reasoning.py
+      - files=examples/online_serving/openai_chat_completion_client_with_tools.py
  actions:
    label:
      add:
@@ -1,324 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-
-# Benchmarks FP8 vs BF16 ViT attention via FlashInfer cuDNN backend.
-#
-# == Usage Examples ==
-#
-# Benchmark mode (default, FlashInfer CUDAGraph Bench)
-#   python3 benchmark_vit_fp8_attn.py
-#
-# Profile mode (PyTorch profiler, saves TensorBoard traces):
-#   python3 benchmark_vit_fp8_attn.py --profile
-#   python3 benchmark_vit_fp8_attn.py --profile --profile-output-dir ./profile_traces
-#
-# Custom seq_lens:
-#   python3 benchmark_vit_fp8_attn.py --seq-lens 4096 8192 16384
-
-from functools import partial
-
-import numpy as np
-import torch
-from torch.profiler import ProfilerActivity, profile, record_function
-
-from vllm.utils.argparse_utils import FlexibleArgumentParser
-
-# Qwen3-VL defaults
-NUM_HEADS = 16
-HEAD_DIM = 72
-DEFAULT_SEQ_LENS = [2304, 4096, 8192, 16384]
-
-
-def _setup_fp8_attention(num_heads: int, head_dim: int) -> tuple:
-    """Create FP8 and BF16 attention modules + workspace."""
-    from types import SimpleNamespace
-    from unittest.mock import patch
-
-    from vllm.config import VllmConfig, set_current_vllm_config
-    from vllm.config.multimodal import MultiModalConfig
-    from vllm.model_executor.layers.attention.mm_encoder_attention import (
-        MMEncoderAttention,
-        _get_flashinfer_workspace_buffer,
-    )
-    from vllm.v1.attention.backends.registry import AttentionBackendEnum
-
-    old_dtype = torch.get_default_dtype()
-    torch.set_default_dtype(torch.bfloat16)
-
-    backend_patch = patch(
-        "vllm.model_executor.layers.attention.mm_encoder_attention"
-        ".get_vit_attn_backend",
-        return_value=AttentionBackendEnum.FLASHINFER,
-    )
-
-    # FP8 attention
-    mm_config_fp8 = MultiModalConfig(mm_encoder_attn_dtype="fp8")
-    vllm_config_fp8 = VllmConfig()
-    vllm_config_fp8.model_config = SimpleNamespace(multimodal_config=mm_config_fp8)
-    with set_current_vllm_config(vllm_config_fp8), backend_patch:
-        attn_fp8 = MMEncoderAttention(
-            num_heads=num_heads,
-            head_size=head_dim,
-            prefix="visual.blocks.0.attn",
-        ).to("cuda")
-
-    # BF16 attention (no FP8)
-    with set_current_vllm_config(VllmConfig()), backend_patch:
-        attn_bf16 = MMEncoderAttention(
-            num_heads=num_heads,
-            head_size=head_dim,
-            prefix="visual.blocks.0.attn",
-        ).to("cuda")
-
-    torch.set_default_dtype(old_dtype)
-
-    workspace = _get_flashinfer_workspace_buffer()
-    return attn_fp8, attn_bf16, workspace
-
-
-def _build_meta(
-    seq_len: int,
-    num_heads: int,
-    head_dim: int,
-    fp8: bool,
-):
-    """Build cu_seqlens, max_seqlen, sequence_lengths."""
-    from vllm.model_executor.layers.attention.mm_encoder_attention import (
-        MMEncoderAttention,
-    )
-    from vllm.utils.math_utils import round_up
-    from vllm.v1.attention.backends.registry import AttentionBackendEnum
-
-    cu_np = np.array([0, seq_len], dtype=np.int32)
-    fp8_padded = num_heads * round_up(head_dim, 16) if fp8 else None
-
-    seq_lengths = MMEncoderAttention.maybe_compute_seq_lens(
-        AttentionBackendEnum.FLASHINFER, cu_np, torch.device("cuda")
-    )
-    max_seqlen = torch.tensor(
-        MMEncoderAttention.compute_max_seqlen(AttentionBackendEnum.FLASHINFER, cu_np),
-        dtype=torch.int32,
-    )
-    cu_seqlens = MMEncoderAttention.maybe_recompute_cu_seqlens(
-        AttentionBackendEnum.FLASHINFER,
-        cu_np,
-        num_heads * head_dim,
-        1,
-        torch.device("cuda"),
-        fp8_padded_hidden_size=fp8_padded,
-    )
-    return cu_seqlens, max_seqlen, seq_lengths
-
-
-def run_benchmark(
-    seq_lens: list[int],
-    num_heads: int,
-    head_dim: int,
-    method: str,
-):
-    """Benchmark FP8 vs BF16 attention across seq_lens.
-
-    Uses FlashInfer GPU-level timing to measure pure kernel time,
-    excluding CPU launch overhead.
-    """
-    if method == "cupti":
-        from flashinfer.testing import bench_gpu_time_with_cupti as bench_fn
-
-        bench_fn = partial(bench_fn, use_cuda_graph=True, cold_l2_cache=False)
-    elif method == "cudagraph":
-        from flashinfer.testing import (
-            bench_gpu_time_with_cudagraph as bench_fn,
-        )
-
-        bench_fn = partial(bench_fn, cold_l2_cache=False)
-    else:
-        raise ValueError(f"Invalid method: {method}")
-
-    attn_fp8, attn_bf16, workspace = _setup_fp8_attention(num_heads, head_dim)
-
-    print(f"Timing method: {method}")
-    print(f"{'seq_len':>8} {'BF16 (us)':>12} {'FP8 (us)':>12} {'Speedup':>10}")
-    print("-" * 46)
-
-    for seq_len in seq_lens:
-        torch.manual_seed(42)
-
-        q = torch.randn(
-            seq_len,
-            num_heads,
-            head_dim,
-            device="cuda",
-            dtype=torch.bfloat16,
-        )
-        k = torch.randn_like(q)
-        v = torch.randn_like(q)
-
-        cu_fp8, max_s, seq_l = _build_meta(seq_len, num_heads, head_dim, fp8=True)
-        # we can reuse cu_fp8 for cu_bf16 since q, k, and v are contiguous
-        cu_bf16 = cu_fp8.clone()
-
-        def bf16_fn(q=q, k=k, v=v, cu=cu_bf16, ms=max_s, sl=seq_l):
-            attn_bf16._forward_flashinfer(q, k, v, cu, ms, sl)
-
-        def fp8_fn(q=q, k=k, v=v, cu=cu_fp8, ms=max_s, sl=seq_l):
-            attn_fp8._forward_flashinfer(q, k, v, cu, ms, sl)
-
-        # bench_fn returns List[float] of per-iteration times in ms
-        bf16_times = bench_fn(bf16_fn)
-        fp8_times = bench_fn(fp8_fn)
-
-        bf16_us = np.median(bf16_times) * 1e3  # ms -> us
-        fp8_us = np.median(fp8_times) * 1e3
-        speedup = bf16_us / fp8_us if fp8_us > 0 else float("inf")
-
-        print(f"{seq_len:>8} {bf16_us:>12.1f} {fp8_us:>12.1f} {speedup:>9.2f}x")
-
-
-def _make_trace_handler(output_dir: str, worker_name: str, label: str):
-    """Create a trace handler that saves to TensorBoard and prints summary."""
-
-    def handler(prof):
-        torch.profiler.tensorboard_trace_handler(output_dir, worker_name)(prof)
-        print(f"\n{'=' * 80}")
-        print(label)
-        print(f"{'=' * 80}")
-        print(prof.key_averages().table(sort_by="cuda_time_total", row_limit=20))
-
-    return handler
-
-
-def run_profile(
-    seq_len: int,
-    num_heads: int,
-    head_dim: int,
-    warmup: int,
-    output_dir: str,
-):
-    """Profile FP8 vs BF16 attention with PyTorch profiler."""
-    attn_fp8, attn_bf16, workspace = _setup_fp8_attention(num_heads, head_dim)
-
-    torch.manual_seed(42)
-    q = torch.randn(
-        seq_len,
-        num_heads,
-        head_dim,
-        device="cuda",
-        dtype=torch.bfloat16,
-    )
-    k = torch.randn_like(q)
-    v = torch.randn_like(q)
-
-    cu_fp8, max_s, seq_l = _build_meta(seq_len, num_heads, head_dim, fp8=True)
-    # we can reuse cu_fp8 for cu_bf16 since q, k, and v are contiguous
-    cu_bf16 = cu_fp8.clone()
-
-    sched = torch.profiler.schedule(wait=0, warmup=warmup, active=1)
-
-    # Profile BF16 (warmup handled by profiler schedule)
-    with profile(
-        activities=[ProfilerActivity.CPU, ProfilerActivity.CUDA],
-        schedule=sched,
-        on_trace_ready=_make_trace_handler(
-            output_dir,
-            f"bf16_h{head_dim}_s{seq_len}",
-            f"BF16 Attention (seq_len={seq_len}, heads={num_heads}, "
-            f"head_dim={head_dim})",
-        ),
-    ) as prof_bf16:
-        for _ in range(warmup + 1):
-            with record_function("bf16_attention"):
-                attn_bf16._forward_flashinfer(
-                    q.clone(), k.clone(), v.clone(), cu_bf16, max_s, seq_l
-                )
-                torch.accelerator.synchronize()
-            prof_bf16.step()
-
-    # Profile FP8 (warmup handled by profiler schedule)
-    with profile(
-        activities=[ProfilerActivity.CPU, ProfilerActivity.CUDA],
-        schedule=sched,
-        on_trace_ready=_make_trace_handler(
-            output_dir,
-            f"fp8_h{head_dim}_s{seq_len}",
-            f"FP8 Attention (seq_len={seq_len}, heads={num_heads}, "
-            f"head_dim={head_dim})",
-        ),
-    ) as prof_fp8:
-        for _ in range(warmup + 1):
-            with record_function("fp8_attention"):
-                attn_fp8._forward_flashinfer(
-                    q.clone(), k.clone(), v.clone(), cu_fp8, max_s, seq_l
-                )
-                torch.accelerator.synchronize()
-            prof_fp8.step()
-
-    print(f"\nTensorBoard traces saved to: {output_dir}")
-    print(f"View with: tensorboard --logdir={output_dir}")
-
-
-if __name__ == "__main__":
-    parser = FlexibleArgumentParser(description="Benchmark FP8 vs BF16 ViT attention.")
-    parser.add_argument(
-        "--seq-lens",
-        type=int,
-        nargs="+",
-        default=DEFAULT_SEQ_LENS,
-        help="Sequence lengths to benchmark",
-    )
-    parser.add_argument(
-        "--num-heads",
-        type=int,
-        default=NUM_HEADS,
-    )
-    parser.add_argument(
-        "--head-dim",
-        type=int,
-        default=HEAD_DIM,
-    )
-    parser.add_argument(
-        "--method",
-        choices=["cupti", "cudagraph"],
-        default="cudagraph",
-        help="GPU timing method: cupti (CUPTI kernel timing) or "
-        "cudagraph (CUDA graph capture/replay). Default: cudagraph",
-    )
-    parser.add_argument(
-        "--warmup",
-        type=int,
-        default=10,
-        help="Warmup iterations (profile mode only)",
-    )
-    parser.add_argument(
-        "--profile",
-        action="store_true",
-        help="Run PyTorch profiler instead of benchmark",
-    )
-    parser.add_argument(
-        "--profile-seq-len",
-        type=int,
-        default=8192,
-        help="Sequence length for profiling (default: 8192)",
-    )
-    parser.add_argument(
-        "--profile-output-dir",
-        type=str,
-        default="./profile_traces",
-        help="Output directory for TensorBoard traces (default: ./profile_traces)",
-    )
-    args = parser.parse_args()
-
-    if args.profile:
-        run_profile(
-            args.profile_seq_len,
-            args.num_heads,
-            args.head_dim,
-            args.warmup,
-            args.profile_output_dir,
-        )
-    else:
-        run_benchmark(
-            args.seq_lens,
-            args.num_heads,
-            args.head_dim,
-            args.method,
-        )
@@ -82,18 +82,73 @@ void launch_persistent_topk(const torch::Tensor& logits,
    size_t smem_size = P::kFixedSmemLarge + chunk_size * sizeof(uint32_t);
    if (smem_size < P::kSmemMedium) smem_size = P::kSmemMedium;

+    // Query occupancy for the instantiation that will actually launch;
+    // overestimating it deadlocks the cooperative barrier.
    int occupancy = 1;
-    cudaOccupancyMaxActiveBlocksPerMultiprocessor(
-        &occupancy, P::persistent_topk_kernel<TopK, 4>, P::kThreadsPerBlock,
-        smem_size);
+    cudaError_t occ_err = cudaSuccess;
+    if (vec_size == 4) {
+      occ_err = cudaOccupancyMaxActiveBlocksPerMultiprocessor(
+          &occupancy, P::persistent_topk_kernel<TopK, 4>, P::kThreadsPerBlock,
+          smem_size);
+    } else if (vec_size == 2) {
+      occ_err = cudaOccupancyMaxActiveBlocksPerMultiprocessor(
+          &occupancy, P::persistent_topk_kernel<TopK, 2>, P::kThreadsPerBlock,
+          smem_size);
+    } else {
+      occ_err = cudaOccupancyMaxActiveBlocksPerMultiprocessor(
+          &occupancy, P::persistent_topk_kernel<TopK, 1>, P::kThreadsPerBlock,
+          smem_size);
+    }
+    TORCH_CHECK(occ_err == cudaSuccess,
+                "persistent_topk occupancy query failed: ",
+                cudaGetErrorString(occ_err));
    if (occupancy < 1) occupancy = 1;

-    uint32_t max_resident_ctas = static_cast<uint32_t>(num_sms) * occupancy;
+    // The cooperative spin-wait barrier only runs when at least one row hits
+    // the radix path (seq_len > RADIX_THRESHOLD). Below that, non-CTA-0 CTAs
+    // early-exit, so oversubscription can't deadlock and headroom is wasted.
+    const bool needs_cooperative =
+        static_cast<uint32_t>(max_seq_len) > P::RADIX_THRESHOLD;
+
+    const uint32_t hw_resident_cap =
+        static_cast<uint32_t>(num_sms) * static_cast<uint32_t>(occupancy);
+    uint32_t max_resident_ctas = hw_resident_cap;
+    if (needs_cooperative) {
+      // Reserve one CTA per SM when occupancy allows; fall back to a single
+      // CTA when occupancy == 1 (the most deadlock-prone case — any straggler
+      // kernel that takes the only slot on one SM hangs the barrier). Never
+      // drop below one full group's worth.
+      uint32_t headroom = (occupancy > 1) ? static_cast<uint32_t>(num_sms) : 1u;
+      if (max_resident_ctas >= headroom + ctas_per_group) {
+        max_resident_ctas -= headroom;
+      }
+    }
    uint32_t num_groups = std::min(max_resident_ctas / ctas_per_group,
                                   static_cast<uint32_t>(num_rows));
    if (num_groups == 0) num_groups = 1;
    uint32_t total_ctas = num_groups * ctas_per_group;

+    // If the cooperative launch wouldn't fit, fall back to FilteredTopK
+    // instead of deadlocking. Only relevant when needs_cooperative.
+    if (needs_cooperative && total_ctas > hw_resident_cap) {
+      TORCH_CHECK(max_smem_per_block >= 128 * 1024,
+                  "persistent_topk would oversubscribe and the FilteredTopK "
+                  "fallback requires >=128KB smem per block (have ",
+                  max_smem_per_block, "). total_ctas=", total_ctas,
+                  " > num_sms*occupancy=", hw_resident_cap, " (TopK=", TopK,
+                  ", vec_size=", vec_size, ", ctas_per_group=", ctas_per_group,
+                  ", smem=", smem_size, ").");
+      cudaError_t status =
+          vllm::FilteredTopKRaggedTransform<float, int32_t, TopK>(
+              logits.data_ptr<float>(), output.data_ptr<int32_t>(),
+              lengths.data_ptr<int32_t>(), static_cast<uint32_t>(num_rows),
+              static_cast<uint32_t>(TopK), static_cast<uint32_t>(stride),
+              stream);
+      TORCH_CHECK(status == cudaSuccess,
+                  "FilteredTopK fallback failed: ", cudaGetErrorString(status));
+      return;
+    }
+
    size_t state_bytes = num_groups * sizeof(P::RadixRowState);
    TORCH_CHECK(workspace.size(0) >= static_cast<int64_t>(state_bytes),
                "workspace too small, need ", state_bytes, " bytes");
@@ -478,9 +478,6 @@ FROM ${FINAL_BASE_IMAGE} AS vllm-base

 ARG CUDA_VERSION
 ARG PYTHON_VERSION
-ARG DEADSNAKES_MIRROR_URL
-ARG DEADSNAKES_GPGKEY_URL
-ARG GET_PIP_URL

 ENV DEBIAN_FRONTEND=noninteractive
 WORKDIR /vllm-workspace
@@ -490,43 +487,35 @@ WORKDIR /vllm-workspace
 RUN PYTHON_VERSION_STR=$(echo ${PYTHON_VERSION} | sed 's/\.//g') && \
    echo "export PYTHON_VERSION_STR=${PYTHON_VERSION_STR}" >> /etc/environment

-# Install Python and system dependencies
+# Install Python (via uv / python-build-standalone) and system dependencies.
+# This replaces the deadsnakes PPA, removing the build-time dependency on
+# Launchpad and matching how the build-stage (`base`) installs Python.
+# python-build-standalone bundles dev headers, the venv module, and
+# python3-config, so the python3.X-dev / python3.X-venv apt packages
+# are not needed.
 RUN apt-get update -y \
    && apt-get install -y --no-install-recommends \
-        software-properties-common \
        curl \
        sudo \
        ffmpeg \
        libsm6 \
        libxext6 \
        libgl1 \
-    && if [ ! -z ${DEADSNAKES_MIRROR_URL} ] ; then \
-        if [ ! -z "${DEADSNAKES_GPGKEY_URL}" ] ; then \
-            mkdir -p -m 0755 /etc/apt/keyrings ; \
-            curl -L ${DEADSNAKES_GPGKEY_URL} | gpg --dearmor > /etc/apt/keyrings/deadsnakes.gpg ; \
-            sudo chmod 644 /etc/apt/keyrings/deadsnakes.gpg ; \
-            echo "deb [signed-by=/etc/apt/keyrings/deadsnakes.gpg] ${DEADSNAKES_MIRROR_URL} $(lsb_release -cs) main" > /etc/apt/sources.list.d/deadsnakes.list ; \
-        fi ; \
-    else \
-        for i in 1 2 3; do \
-            add-apt-repository -y ppa:deadsnakes/ppa && break || \
-            { echo "Attempt $i failed, retrying in 5s..."; sleep 5; }; \
-        done ; \
-    fi \
-    && apt-get update -y \
-    && apt-get install -y --no-install-recommends \
-        python${PYTHON_VERSION} \
-        python${PYTHON_VERSION}-dev \
-        python${PYTHON_VERSION}-venv \
        libibverbs-dev \
    && rm -rf /var/lib/apt/lists/* \
-    && update-alternatives --install /usr/bin/python3 python3 /usr/bin/python${PYTHON_VERSION} 1 \
-    && update-alternatives --set python3 /usr/bin/python${PYTHON_VERSION} \
-    && ln -sf /usr/bin/python${PYTHON_VERSION}-config /usr/bin/python3-config \
-    && rm -f /usr/lib/python${PYTHON_VERSION}/EXTERNALLY-MANAGED \
-    && curl -sS ${GET_PIP_URL} | python${PYTHON_VERSION} \
+    && curl -LsSf https://astral.sh/uv/install.sh | sh \
+    && $HOME/.local/bin/uv venv /opt/venv --python ${PYTHON_VERSION} \
+    && rm -f /usr/bin/python3 /usr/bin/python3-config /usr/bin/pip \
+    && ln -s /opt/venv/bin/python3 /usr/bin/python3 \
+    && ln -s /opt/venv/bin/python${PYTHON_VERSION} /usr/bin/python${PYTHON_VERSION} \
+    && ln -s /opt/venv/bin/python3-config /usr/bin/python3-config \
+    && ln -s /opt/venv/bin/pip /usr/bin/pip \
    && python3 --version && python3 -m pip --version

+# Activate virtual environment and add uv to PATH
+ENV PATH="/opt/venv/bin:/root/.local/bin:$PATH"
+ENV VIRTUAL_ENV="/opt/venv"
+
 # Install CUDA development tools for runtime JIT compilation
 # (FlashInfer, DeepGEMM, EP kernels all require compilation at runtime)
 RUN CUDA_VERSION_DASH=$(echo $CUDA_VERSION | cut -d. -f1,2 | tr '.' '-') && \
@@ -538,7 +527,7 @@ RUN CUDA_VERSION_DASH=$(echo $CUDA_VERSION | cut -d. -f1,2 | tr '.' '-') && \
        cuda-nvrtc-${CUDA_VERSION_DASH} \
        cuda-cuobjdump-${CUDA_VERSION_DASH} \
        libcurand-dev-${CUDA_VERSION_DASH} \
-        libcublas-dev-${CUDA_VERSION_DASH} \
+        libcublas-${CUDA_VERSION_DASH} \
        # Required by fastsafetensors (fixes #20384)
        libnuma-dev \
        # numactl CLI for NUMA binding at runtime
@@ -551,9 +540,6 @@ RUN CUDA_VERSION_DASH=$(echo $CUDA_VERSION | cut -d. -f1,2 | tr '.' '-') && \
    apt-get install -y --no-install-recommends --allow-change-held-packages libnccl-dev=${NCCL_VER} libnccl2=${NCCL_VER} && \
    rm -rf /var/lib/apt/lists/*

-# Install uv for faster pip installs
-RUN python3 -m pip install uv
-
 # Environment for uv
 ENV UV_HTTP_TIMEOUT=500
 ENV UV_INDEX_STRATEGY="unsafe-best-match"
@@ -585,7 +571,9 @@ RUN --mount=type=cache,target=/root/.cache/uv \
 ARG FLASHINFER_VERSION=0.6.8.post1
 RUN --mount=type=cache,target=/root/.cache/uv \
    uv pip install --system flashinfer-jit-cache==${FLASHINFER_VERSION} \
-        --extra-index-url https://flashinfer.ai/whl/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.')
+        --extra-index-url https://flashinfer.ai/whl/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.') \
+    && flashinfer show-config \
+    && flashinfer download-cubin

 # ============================================================
 # OPENAI API SERVER DEPENDENCIES
@@ -667,13 +655,6 @@ RUN --mount=type=bind,from=build,src=/tmp/ep_kernels_workspace/dist,target=/vllm
    uv pip install --system ep_kernels/dist/*.whl --verbose \
        --extra-index-url ${PYTORCH_CUDA_INDEX_BASE_URL}/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.')

-# Download FlashInfer precompiled cubins AFTER all pip installs are done.
-# This must run after the vLLM wheel and EP kernels installs above, because
-# those can reinstall/touch flashinfer packages. Downloading cubins earlier
-# (in the flashinfer-jit-cache layer) causes ~2.5 GB of layer duplication
-# when a later pip install overwrites flashinfer package files.
-RUN flashinfer show-config && flashinfer download-cubin
-
 # CUDA image changed from /usr/local/nvidia to /usr/local/cuda in 12.8 but will
 # return to /usr/local/nvidia in 13.0 to allow container providers to mount drivers
 # consistently from the host (see https://github.com/vllm-project/vllm/issues/18859).
@@ -748,7 +729,7 @@ ENV HF_XET_HIGH_PERFORMANCE 1
 ENV HF_HUB_DOWNLOAD_TIMEOUT 60

 # Copy in the v1 package for testing (it isn't distributed yet)
-COPY vllm/v1 /usr/local/lib/python${PYTHON_VERSION}/dist-packages/vllm/v1
+COPY vllm/v1 /opt/venv/lib/python${PYTHON_VERSION}/site-packages/vllm/v1

 # Source code is used in the `python_only_compile.sh` test
 # We hide it inside `src/` so that this source code
@@ -163,7 +163,7 @@ Running with a local file:

 ```bash
 vllm run-batch \
-    -i features/openai_batch/openai_example_batch.jsonl \
+    -i offline_inference/openai_batch/openai_example_batch.jsonl \
    -o results.jsonl \
    --model meta-llama/Meta-Llama-3-8B-Instruct
 ```
@@ -172,7 +172,7 @@ Using remote file:

 ```bash
 vllm run-batch \
-    -i https://raw.githubusercontent.com/vllm-project/vllm/main/examples/features/openai_batch/openai_example_batch.jsonl \
+    -i https://raw.githubusercontent.com/vllm-project/vllm/main/examples/offline_inference/openai_batch/openai_example_batch.jsonl \
    -o results.jsonl \
    --model meta-llama/Meta-Llama-3-8B-Instruct
 ```
@@ -23,7 +23,7 @@ llm = LLM(model="ibm-granite/granite-3.1-8b-instruct", tensor_parallel_size=2)
 !!! note
    With tensor parallelism enabled, each process will read the whole model and split it into chunks, which makes the disk reading time even longer (proportional to the size of tensor parallelism).

-    You can convert the model checkpoint to a sharded checkpoint using [examples/features/sharded_state/load_sharded_state_offline.py](../../examples/features/sharded_state/load_sharded_state_offline.py). The conversion process might take some time, but later you can load the sharded checkpoint much faster. The model loading time should remain constant regardless of the size of tensor parallelism.
+    You can convert the model checkpoint to a sharded checkpoint using [examples/offline_inference/save_sharded_state.py](../../examples/offline_inference/save_sharded_state.py). The conversion process might take some time, but later you can load the sharded checkpoint much faster. The model loading time should remain constant regardless of the size of tensor parallelism.

 ## Quantization

@@ -42,7 +42,7 @@ Traces can be visualized using <https://ui.perfetto.dev/>.

 #### Offline Inference

-Refer to [examples/features/profiling/simple_profiling_offline.py](../../examples/features/profiling/simple_profiling_offline.py) for an example.
+Refer to [examples/offline_inference/simple_profiling.py](../../examples/offline_inference/simple_profiling.py) for an example.

 #### OpenAI Server

@@ -11,7 +11,7 @@ Automatic Prefix Caching (APC in short) caches the KV cache of existing queries,

 Set `enable_prefix_caching=True` in vLLM engine to enable APC. Here is an example:

-[examples/features/automatic_prefix_caching/automatic_prefix_caching_offline.py](../../examples/features/automatic_prefix_caching/automatic_prefix_caching_offline.py)
+[examples/offline_inference/automatic_prefix_caching.py](../../examples/offline_inference/automatic_prefix_caching.py)

 ## Example workloads

@@ -6,12 +6,12 @@ This directory contains examples for extending the context length of models usin

 ## Offline Inference Example

-The [`context_extension.py`](../../examples/features/context_extension/context_extension_offline.py) script demonstrates how to extend the context length of a Qwen model using the YARN method (rope_parameters) and run a simple chat example.
+The [`context_extension.py`](../../examples/offline_inference/context_extension) script demonstrates how to extend the context length of a Qwen model using the YARN method (rope_parameters) and run a simple chat example.

 ### Usage

 ```bash
-python examples/features/context_extension/context_extension_offline.py
+python examples/offline_inference/context_extension.py
 ```

 ## OpenAI Online Method
@@ -47,7 +47,7 @@ the third parameter is the path to the LoRA adapter.
    )
    ```

-Check out [examples/features/lora/multilora_offline.py](../../examples/features/lora/multilora_offline.py) for an example of how to use LoRA adapters with the async engine and how to use more advanced configuration options.
+Check out [examples/offline_inference/multilora_inference.py](../../examples/offline_inference/multilora_inference.py) for an example of how to use LoRA adapters with the async engine and how to use more advanced configuration options.

 ## Serving LoRA Adapters

@@ -68,7 +68,7 @@ You can pass a single image to the `'image'` field of the multi-modal dictionary
        print(generated_text)
    ```

-Full example: [examples/generate/multimodal/vision_language_offline.py](../../examples/generate/multimodal/vision_language_offline.py)
+Full example: [examples/offline_inference/vision_language.py](../../examples/offline_inference/vision_language.py)

 To substitute multiple images inside the same text prompt, you can pass in a list of images instead:

@@ -101,7 +101,7 @@ To substitute multiple images inside the same text prompt, you can pass in a lis
        print(generated_text)
    ```

-Full example: [examples/generate/multimodal/vision_language_multi_image_offline.py](../../examples/generate/multimodal/vision_language_multi_image_offline.py)
+Full example: [examples/offline_inference/vision_language_multi_image.py](../../examples/offline_inference/vision_language_multi_image.py)

 If using the [LLM.chat](../models/generative_models.md#llmchat) method, you can pass images directly in the message content using various formats: image URLs, PIL Image objects, or pre-computed embeddings:

@@ -287,13 +287,13 @@ Instead of NumPy arrays, you can also pass `'torch.Tensor'` instances, as shown
    !!! note
        'process_vision_info' is only applicable to Qwen2.5-VL and similar models.

-Full example: [examples/generate/multimodal/vision_language_offline.py](../../examples/generate/multimodal/vision_language_offline.py)
+Full example: [examples/offline_inference/vision_language.py](../../examples/offline_inference/vision_language.py)

 ### Audio Inputs

 You can pass a tuple `(array, sampling_rate)` to the `'audio'` field of the multi-modal dictionary.

-Full example: [examples/generate/multimodal/audio_language_offline.py](../../examples/generate/multimodal/audio_language_offline.py)
+Full example: [examples/offline_inference/audio_language.py](../../examples/offline_inference/audio_language.py)

 #### Chunking Long Audio for Transcription

@@ -674,7 +674,7 @@ Then, you can use the OpenAI client as follows:
    print("Chat completion output:", chat_response.choices[0].message.content)
    ```

-Full example: [examples/generate/multimodal/openai_chat_completion_client_for_multimodal.py](../../examples/generate/multimodal/openai_chat_completion_client_for_multimodal.py)
+Full example: [examples/online_serving/openai_chat_completion_client_for_multimodal.py](../../examples/online_serving/openai_chat_completion_client_for_multimodal.py)

 !!! tip
    Loading from local file paths is also supported on vLLM: You can specify the allowed local media path via `--allowed-local-media-path` when launching the API server/engine,
@@ -745,7 +745,7 @@ Then, you can use the OpenAI client as follows:
    print("Chat completion output from image url:", result)
    ```

-Full example: [examples/generate/multimodal/openai_chat_completion_client_for_multimodal.py](../../examples/generate/multimodal/openai_chat_completion_client_for_multimodal.py)
+Full example: [examples/online_serving/openai_chat_completion_client_for_multimodal.py](../../examples/online_serving/openai_chat_completion_client_for_multimodal.py)

 !!! note
    By default, the timeout for fetching videos through HTTP URL is `30` seconds.
@@ -958,7 +958,7 @@ Alternatively, you can pass `audio_url`, which is the audio counterpart of `imag
    print("Chat completion output from audio url:", result)
    ```

-Full example: [examples/generate/multimodal/openai_chat_completion_client_for_multimodal.py](../../examples/generate/multimodal/openai_chat_completion_client_for_multimodal.py)
+Full example: [examples/online_serving/openai_chat_completion_client_for_multimodal.py](../../examples/online_serving/openai_chat_completion_client_for_multimodal.py)

 !!! note
    By default, the timeout for fetching audios through HTTP URL is `10` seconds.
@@ -16,7 +16,7 @@ To input multi-modal data, follow this schema in [vllm.inputs.EmbedsPrompt][]:

 You can pass prompt embeddings from Hugging Face Transformers models to the  `'prompt_embeds'` field of the prompt embedding dictionary, as shown in the following examples:

-[examples/features/prompt_embed/prompt_embed_offline.py](../../examples/features/prompt_embed/prompt_embed_offline.py)
+[examples/offline_inference/prompt_embed_inference.py](../../examples/offline_inference/prompt_embed_inference.py)

 ## Online Serving

@@ -41,4 +41,4 @@ vllm serve meta-llama/Llama-3.2-1B-Instruct --runner generate \

 Then, you can use the OpenAI client as follows:

-[examples/features/prompt_embed/prompt_embed_inference_with_openai_client.py](../../examples/features/prompt_embed/prompt_embed_inference_with_openai_client.py)
+[examples/online_serving/prompt_embed_inference_with_openai_client.py](../../examples/online_serving/prompt_embed_inference_with_openai_client.py)
@@ -20,7 +20,6 @@ The following are the supported quantization formats for vLLM:
 - [AMD Quark](quark.md)
 - [Quantized KV Cache](quantized_kvcache.md)
 - [TorchAO](torchao.md)
- [FP8 ViT Encoder Attention](fp8_vit_attn.md)

 ## Supported Hardware

@@ -1,109 +0,0 @@
-# FP8 ViT Encoder Attention
-
-For visual understanding workloads with large images (e.g. QHD, 4K) and relatively
-short text prompts/generation, the ViT encoder attention can become a significant
-bottleneck, especially when the text model is quantized (e.g. NVFP4). vLLM
-supports optional FP8 quantization for the ViT encoder attention via the
-FlashInfer cuDNN backend. Q/K/V are quantized on-the-fly to FP8 before the
-cuDNN attention call.
-
-!!! note
-    - Currently supports Qwen3-VL family models only (`qwen3_vl`, `qwen3_vl_moe`,
-      `qwen3_5`, `qwen3_5_moe`, and other models using Qwen3 ViT).
-    - Dynamic scaling is not compatible with ViT full CUDA graphs.
-    - Performance gains are mostly visible at QHD/4K resolutions or multi-image
-      requests. Smaller images may see no speedup due to quantization overhead
-      (3 quantization kernel launches + un-padding).
-    - FP8 tensor-core speedup is more pronounced on GB300 than GB200.
-
-## Requirements
-
- FlashInfer cuDNN backend with cuDNN >= 9.17.1.
-
-## Usage
-
-Enable FP8 ViT attention by passing `--mm-encoder-attn-dtype fp8` together
-with `--mm-encoder-attn-backend FLASHINFER`:
-
-```bash
-vllm serve $MODEL \
-    --mm-encoder-attn-backend FLASHINFER \
-    --mm-encoder-attn-dtype fp8
-```
-
-By default (no scale file), **dynamic scaling** is used: a 16-entry circular
-buffer of observed Q/K/V amax values drives per-forward scale updates. This
-matches BF16 accuracy without any calibration but adds a small per-forward
-overhead.
-
-## Calibrate-Once, Reuse Workflow (Recommended)
-
-For production, calibrate static scales on a representative dataset once and
-reuse them to avoid the dynamic overhead:
-
-```bash
-# Step 1: calibrate and save scales (runs dynamic scaling for 16 passes,
-# then dumps the learned scales to JSON).
-vllm bench mm-processor \
-    --model $MODEL --mm-encoder-attn-backend FLASHINFER \
-    --mm-encoder-attn-dtype fp8 \
-    --mm-encoder-fp8-scale-save-path /path/to/scales.json \
-    --dataset-name hf --dataset-path lmarena-ai/VisionArena-Chat \
-    --num-prompts 100
-
-# Step 2: serve with static scales (no dynamic overhead).
-vllm serve $MODEL \
-    --mm-encoder-attn-backend FLASHINFER \
-    --mm-encoder-attn-dtype fp8 \
-    --mm-encoder-fp8-scale-path /path/to/scales.json
-```
-
-Saved scales are multiplied by `--mm-encoder-fp8-scale-save-margin` (default
-`1.5`) to leave headroom against activation outliers not present in the
-calibration set. The default has been validated to generalize across datasets
-(e.g. VisionArena-Chat calibration maintains BF16 accuracy on ChartQA).
-
-## Scale File Format
-
-```json
-{
-    "visual.blocks.0.attn.attn": {"q": 224.0, "k": 198.0, "v": 210.0},
-    "visual.blocks.1.attn.attn": {"q": 218.0, "k": 195.0, "v": 207.0}
-}
-```
-
-Keys `q_scale` / `k_scale` / `v_scale` are accepted as aliases.
-
-## Performance
-
-**Core cuDNN attention kernel** (PyTorch profiler, `cudnn_generated_fort_native_sdpa_sm100_flash_fprop`, head_dim=128, seq_len=8192):
-
-| Hardware | BF16 | FP8 | Speedup |
-| -------- | ---- | ---- | ------- |
-| GB200 | 350 us | 312 us | **1.12x** |
-| GB300 | 300 us | 211 us | **1.42x** |
-
-**End-to-end encoder forward time** (Qwen3-VL-30B-A3B-Instruct on GB200, 3 images/request):
-
-| Resolution | BF16 median | FP8 median | Speedup |
-| ---------- | ----------- | ---------- | ------- |
-| HD (720x1280) | 31.77 ms | 36.39 ms | 0.87x |
-| FullHD (1080x1920) | 57.99 ms | 58.73 ms | ~same |
-| QHD (1440x2560) | 131.83 ms | 122.30 ms | **1.08x** |
-| 4K (2160x3840) | 543.44 ms | 460.31 ms | **1.18x** |
-
-Crossover is around FullHD with 3 images/request. At QHD and above, FP8 wins.
-
-## Accuracy
-
-ChartQA, Qwen3-VL-8B-Instruct, 500 samples. FP8 static uses scales calibrated
-on VisionArena-Chat (with default 1.5x margin):
-
-| Metric | BF16 | FP8 dynamic | FP8 static |
-| ------ | ---- | ----------- | ---------- |
-| relaxed_accuracy | 0.780 | 0.776 | 0.780 |
-| anywhere_accuracy | 0.806 | 0.816 | 0.814 |
-| exact_match | 0.584 | 0.582 | 0.578 |
-
-All three configurations match within statistical noise, confirming that
-static scales calibrated on one dataset generalize to another.
@@ -13,7 +13,6 @@ vLLM currently supports the following reasoning models:

 | Model Series | Parser Name | Structured Output Support | Tool Calling |
 | ------------ | ----------- | ---------------- | ----------- |
-| [Cohere Command A Reasoning](https://huggingface.co/CohereLabs/command-a-reasoning-08-2025) | `cohere_command3` | `json`, `regex` | ✅ |
 | [DeepSeek R1 series](https://huggingface.co/collections/deepseek-ai/deepseek-r1-678e1e131c0169c0bc89728d) | `deepseek_r1` | `json`, `regex` | ❌ |
 | [DeepSeek-V3.1](https://huggingface.co/collections/deepseek-ai/deepseek-v31-68a491bed32bd77e7fca048f) | `deepseek_v3` | `json`, `regex` | ❌ |
 | [ERNIE-4.5-VL series](https://huggingface.co/baidu/ERNIE-4.5-VL-28B-A3B-PT) | `ernie45` | `json`, `regex` | ❌ |
@@ -203,7 +202,7 @@ The reasoning content is also available when both tool calling and the reasoning
    print(f"Arguments: {tool_call.arguments}")
    ```

-For more examples, please refer to [examples/reasoning/openai_chat_completion_tool_calls_with_reasoning.py](../../examples/reasoning/openai_chat_completion_tool_calls_with_reasoning.py).
+For more examples, please refer to [examples/online_serving/openai_chat_completion_tool_calls_with_reasoning.py](../../examples/online_serving/openai_chat_completion_tool_calls_with_reasoning.py).

 ## Server-Level Default Chat Template Kwargs

@@ -32,7 +32,7 @@ depend on your model family, traffic pattern, hardware, and sampling settings.
 | Suffix decoding | Low to medium gain | Medium gain | No extra draft model; dynamic speculation depth. |

 For reproducible measurements in your environment, use
-[`examples/features/speculative_decoding/spec_decode_offline.py`](../../../examples/features/speculative_decoding/spec_decode_offline.py)
+[`examples/offline_inference/spec_decode.py`](../../../examples/offline_inference/spec_decode.py)
 or the [benchmark CLI guide](../../benchmarking/cli.md).

 ## `--speculative-config` schema
@@ -1,6 +1,6 @@
 # EAGLE Draft Models

-The following code configures vLLM to use speculative decoding where proposals are generated by an [EAGLE (Extrapolation Algorithm for Greater Language-model Efficiency)](https://arxiv.org/pdf/2401.15077) based draft model. A more detailed example for offline mode, including how to extract request level acceptance rate, can be found in [examples/features/speculative_decoding/spec_decode_offline.py](../../../examples/features/speculative_decoding/spec_decode_offline.py)
+The following code configures vLLM to use speculative decoding where proposals are generated by an [EAGLE (Extrapolation Algorithm for Greater Language-model Efficiency)](https://arxiv.org/pdf/2401.15077) based draft model. A more detailed example for offline mode, including how to extract request level acceptance rate, can be found in [examples/offline_inference/spec_decode.py](../../../examples/offline_inference/spec_decode.py)

 ## Eagle Drafter Example

@@ -165,7 +165,7 @@ As an example, we can use to define a specific format of simplified SQL queries:
    print(completion.choices[0].message.content)
    ```

-See also: [full example](../../examples/features/structured_outputs/README.md)
+See also: [full example](../examples/online_serving/structured_outputs.md)

 ## Reasoning Outputs

@@ -208,7 +208,7 @@ Note that you can use reasoning with any provided structured outputs feature. Th
    print("content: ", completion.choices[0].message.content)
    ```

-See also: [full example](../../examples/features/structured_outputs/README.md)
+See also: [full example](../examples/online_serving/structured_outputs.md)

 !!! note
    When using Qwen3 Coder models with reasoning enabled, structured outputs might become disabled if the reasoning content does not get parsed into the `reasoning` field separately (v0.11.2+).
@@ -304,7 +304,7 @@ Step #2: explanation="Next, let's isolate 'x' by dividing both sides of the equa
 Answer: x = -29/8
 ```

-An example of using `structural_tag` can be found here: [examples/features/structured_outputs](../../examples/features/structured_outputs/README.md)
+An example of using `structural_tag` can be found here: [examples/online_serving/structured_outputs](../../examples/online_serving/structured_outputs)

 ## Offline Inference

@@ -339,4 +339,4 @@ shown below:
    print(outputs[0].outputs[0].text)
    ```

-See also: [full example](../../examples/features/structured_outputs/structured_outputs_offline.py)
+See also: [full example](../examples/online_serving/structured_outputs.md)
@@ -369,16 +369,6 @@ Flags:
 * For non-reasoning: `--tool-call-parser hunyuan_a13b`
 * For reasoning: `--tool-call-parser hunyuan_a13b --reasoning-parser hunyuan_a13b`

-### Cohere Command A Reasoning (`cohere_command3`)
-
-Supported models:
-
-* [`CohereLabs/command-a-reasoning-08-2025`](https://huggingface.co/CohereLabs/command-a-reasoning-08-2025)
-
-Flags: `--tool-call-parser cohere_command3 --reasoning-parser cohere_command3`
-
-Note: the Cohere tool parser requires the `cohere_melody` package, which is not installed by default. Before using this parser please install the [cohere_melody](https://pypi.org/project/cohere-melody/) package.
-
 ### LongCat-Flash-Chat Models (`longcat`)

 Supported models:
@@ -101,7 +101,7 @@ vllm serve /path/to/sharded/model \
    --model-loader-extra-config '{"pattern":"custom-model-rank-{rank}-part-{part}.safetensors"}'
 ```

-To create sharded model files, you can use the script provided in [examples/features/sharded_state/save_sharded_state_offline.py](../../../examples/features/sharded_state/save_sharded_state_offline.py). This script demonstrates how to save a model in the sharded format that is compatible with the Run:ai Model Streamer sharded loader.
+To create sharded model files, you can use the script provided in [examples/offline_inference/save_sharded_state.py](../../../examples/offline_inference/save_sharded_state.py). This script demonstrates how to save a model in the sharded format that is compatible with the Run:ai Model Streamer sharded loader.

 The sharded loader supports all the same tunable parameters as the regular Run:ai Model Streamer, including `concurrency` and `memory_limit`. These can be configured in the same way:

@@ -292,10 +292,10 @@ Pooling models now support token-wise task.

 ### Score task

-`score` task have has been removed in v0.21, use `classify` instead. Only when a classification model outputs num_labels
-equal to 1 can it be used as a scoring model and have its scoring API enabled.
+`score` task is deprecated and will be removed in v0.20. Please use `classify` instead. Only when a
+classification model outputs num_labels equal to 1 can it be used as a scoring model and have its scoring API enabled.

 ### Pooling multitask support

-Pooling multitask support has been removed in v0.21. When the default pooling task is not what you want,
+Pooling multitask support is deprecated and will be removed in v0.20. When the default pooling task is not what you want,
 you need to manually specify it via `PoolerConfig(task=<task>)` offline or `--pooler-config.task <task>` online.
@@ -439,7 +439,6 @@ th {
 | `Mamba2ForCausalLM` | Mamba2 | `mistralai/Mamba-Codestral-7B-v0.1`, etc. | | ✅︎ |
 | `MiMoForCausalLM` | MiMo | `XiaomiMiMo/MiMo-7B-RL`, etc. | ✅︎ | ✅︎ |
 | `MiMoV2FlashForCausalLM` | MiMoV2Flash | `XiaomiMiMo/MiMo-V2-Flash`, etc. | | ✅︎ |
-| `MiMoV2ForCausalLM` | MiMoV2Pro | `XiaomiMiMo/MiMo-V2.5-Pro`, etc. | | ✅︎ |
 | `MiniCPMForCausalLM` | MiniCPM | `openbmb/MiniCPM-2B-sft-bf16`, `openbmb/MiniCPM-2B-dpo-bf16`, `openbmb/MiniCPM-S-1B-sft`, etc. | ✅︎ | ✅︎ |
 | `MiniCPM3ForCausalLM` | MiniCPM3 | `openbmb/MiniCPM3-4B`, etc. | ✅︎ | ✅︎ |
 | `MiniMaxForCausalLM` | MiniMax-Text | `MiniMaxAI/MiniMax-Text-01-hf`, etc. | | |
@@ -591,7 +590,6 @@ These models primarily accept the [`LLM.generate`](./generative_models.md#llmgen
 | `LlavaNextVideoForConditionalGeneration` | LLaVA-NeXT-Video | T + V | `llava-hf/LLaVA-NeXT-Video-7B-hf`, etc. | | ✅︎ |
 | `LlavaOnevisionForConditionalGeneration` | LLaVA-Onevision | T + I<sup>+</sup> + V<sup>+</sup> | `llava-hf/llava-onevision-qwen2-7b-ov-hf`, `llava-hf/llava-onevision-qwen2-0.5b-ov-hf`, etc. | | ✅︎ |
 | `MiDashengLMModel` | MiDashengLM | T + A<sup>+</sup> | `mispeech/midashenglm-7b` | | ✅︎ |
-| `MiMoV2OmniForCausalLM` | MiMo-V2.5-Omni | T + I<sup>E+</sup> + V<sup>E+</sup> + A<sup>+</sup> | `XiaomiMiMo/MiMo-V2.5-Omni` | | ✅︎ |
 | `MiniCPMO` | MiniCPM-O | T + I<sup>E+</sup> + V<sup>E+</sup> + A<sup>E+</sup> | `openbmb/MiniCPM-o-2_6`, etc. | ✅︎ | ✅︎ |
 | `MiniCPMV` | MiniCPM-V | T + I<sup>E+</sup> + V<sup>E+</sup> | `openbmb/MiniCPM-V-2` (see note), `openbmb/MiniCPM-Llama3-V-2_5`, `openbmb/MiniCPM-V-2_6`, `openbmb/MiniCPM-V-4`, `openbmb/MiniCPM-V-4_5`, etc. | ✅︎ | |
 | `MiniMaxVL01ForConditionalGeneration` | MiniMax-VL | T + I<sup>E+</sup> | `MiniMaxAI/MiniMax-VL-01`, etc. | | ✅︎ |
@@ -16,7 +16,7 @@ For MoE models, when any requests are in progress in any rank, we must ensure th

 In all cases, it is beneficial to load-balance requests between DP ranks. For online deployments, this balancing can be optimized by taking into account the state of each DP engine - in particular its currently scheduled and waiting (queued) requests, and KV cache state. Each DP engine has an independent KV cache, and the benefit of prefix caching can be maximized by directing prompts intelligently.

-This document focuses on online deployments (with the API server). DP + EP is also supported for offline usage (via the LLM class), for an example see [examples/features/data_parallel/data_parallel_offline.py](../../examples/features/data_parallel/data_parallel_offline.py).
+This document focuses on online deployments (with the API server). DP + EP is also supported for offline usage (via the LLM class), for an example see [examples/offline_inference/data_parallel.py](../../examples/offline_inference/data_parallel.py).

 There are two distinct modes supported for online deployments - self-contained with internal load balancing, or externally per-rank process deployment and load balancing.

@@ -251,7 +251,7 @@ The following extra parameters are supported:
 Our Responses API is compatible with [OpenAI's Responses API](https://platform.openai.com/docs/api-reference/responses);
 you can use the [official OpenAI Python client](https://github.com/openai/openai-python) to interact with it.

-Code example: [examples/online_serving/openai_responses_client_with_tools.py](../../examples/tool_calling/openai_responses_client_with_tools.py)
+Code example: [examples/online_serving/openai_responses_client_with_tools.py](../../examples/online_serving/openai_responses_client_with_tools.py)

 #### Extra parameters

@@ -279,7 +279,7 @@ you can use the [official OpenAI Python client](https://github.com/openai/openai
 !!! note
    To use the Transcriptions API, please install with extra audio dependencies using `pip install vllm[audio]`.

-Code example: [examples/speech_to_text/openai/openai_transcription_client.py](../../examples/speech_to_text/openai/openai_transcription_client.py)
+Code example: [examples/online_serving/openai_transcription_client.py](../../examples/online_serving/openai_transcription_client.py)

 NOTE: beam search is currently supported in the transcriptions endpoint for encoder-decoder multimodal models, e.g., whisper, but highly inefficient as work for handling the encoder/decoder cache is actively ongoing. This is an active point of ongoing optimization and will be handled properly in the very near future.

@@ -397,7 +397,7 @@ Please mind that the popular `openai/whisper-large-v3-turbo` model does not supp
 !!! note
    To use the Translation API, please install with extra audio dependencies using `pip install vllm[audio]`.

-Code example: [examples/speech_to_text/openai/openai_translation_client.py](../../examples/speech_to_text/openai/openai_translation_client.py)
+Code example: [examples/online_serving/openai_translation_client.py](../../examples/online_serving/openai_translation_client.py)

 #### Extra Parameters

@@ -1,146 +0,0 @@
-# What is Layerwise (Re)loading?
-
-Layerwise reloading is the system used to handle the loading of new weight data into existing weight data destinations without triggering recompilation of the cuda graph and other runtime artifacts. This system is used to enable [QeRL](https://arxiv.org/pdf/2510.11696)-style post training flows, where full-precision trainer weights are quantized and loaded into a target vLLM instance for fast, high-exploration rollouts. The core implementation can be found in [layerwise.py](../../vllm/model_executor/model_loader/reload/layerwise.py).
-
-![Layerwise](../assets/training/layerwise.png)
-
-## Layerwise Reloading for QeRL
-
-In order to load new weights into existing weight data destinations, a weight must undergo the following operations:
-
- Transfer: weights must be transferred from trainer model to target node/device
- Fuse: weight partitions must be fused, for example qkv/gate_up
- Process: this typically means online quantization and kernel-specific padding or striding
- Shard: weights must be sharded according to the selected parallelism strategy
- Copy: weights must be copied into the existing weight data destinations
-
-Layerwise reloading achieves this using the following steps:
-
-1. Weights are **transferred** from the trainer to the target (see [weight_transfer](weight_transfer/README.md))
-2. Weights loaded via `model.load_weights`, during which they are **sharded** and **fused**
-3. Weights are **processed** in an online fashion as soon as all of a layer's weights are loaded
-4. Weights are **copied** into the existing weight data destinations
-
-For more information on implementation, see [Low Level `layerwise` API](#low-level-layerwise-api).
-
-## Layerwise Loading with Online Quantization
-
-Online quantization refers to when a user provides full precision weights and those weights are quantized on-the-fly as they are loaded into the model. The layerwise reloading system handles this by treating online quantization as a **processing** step, which is then handled in an online way both during first-time load and during reload. A typical online quantization method implementation should look like this:
-
-```python
-class Fp8OnlineLinearMethod(Fp8LinearMethod):
-    """Online version of Fp8LinearMethod which loads a full precision checkpoint
-    and quantizes weights during loading."""
-
-    uses_meta_device: bool = True
-
-    def create_weights(self, layer: torch.nn.Module, ...):
-        # weight is materialized and processed during loading
-        layer.weight = ModelWeightParameter(
-            data=torch.empty(..., device="meta"),
-            weight_loader=weight_loader,
-        )
-
-        # set up online processing
-        initialize_online_processing(layer)
-
-    def process_weights_after_loading(self, layer: Module) -> None:
-        if getattr(layer, "_already_called_process_weights_after_loading", False):
-            return
-
-        layer.weight, layer.weight_scale = ops.scaled_fp8_quant(layer.weight)
-
-        # Prevent duplicate processing (e.g., during weight reload)
-        layer._already_called_process_weights_after_loading = True
-```
-
-## Example Usages
-
-### High Level Weight Transfer API
-
-The layerwise reloading system is integrated with the post-training weight transfer system. To use layerwise reloading in conjunction to the weight transfer system, follow the examples found [here](../../examples/rl/). Layerwise reloading is controlled by the `WeightTransferUpdateInfo.is_checkpoint_format` flag and is set to `True` by default.
-
-### Mid Level `reload_weights` API
-
-Layerwise reloading is also exposed via the `reload_weights` API. This interface can be called using the following code:
-
-```python
-from vllm import LLM
-
-llm = LLM("Qwen/Qwen3-0.6B")
-llm.collective_rpc("reload_weights")
-```
-
-This interface also allows specifying a `weights_path` which can be used to select a checkpoint path to load from:
-
-```python
-from vllm import LLM
-
-# fine tuned model checkpoints for testing
-mul_path = "inference-optimization/Qwen3-0.6B-debug-multiply"
-add_path = "inference-optimization/Qwen3-0.6B-debug-add"
-
-llm = LLM("Qwen/Qwen3-0.6B")
-llm.collective_rpc("reload_weights", kwargs={"weights_path": mul_path})
-llm.generate("3 4 = ")  # 12
-
-llm.collective_rpc("reload_weights", kwargs={"weights_path": add_path})
-llm.generate("3 4 = ")  # 7
-```
-
-Finally, a `weights_iterator` can be provided directly. This iterator can be lazy or eagerly defined.
-
-```python
-from vllm import LLM
-
-weights_iterator = [("q_proj", ...), ("k_proj", ...), ...]
-
-llm = LLM("Qwen/Qwen3-0.6B")
-llm.collective_rpc("reload_weights", kwargs={"weights_iterator": weights_iterator})
-```
-
-### Low Level `layerwise` API
-
-[layerwise.py](../../vllm/model_executor/model_loader/reload/layerwise.py) Implements the following functions to execute its lifecycle:
-
-| Function | Purpose | Quantized Reload | Online Quantization |
-| - | - | - | - |
-| `record_metadata_for_reloading` | Record tensor metadata so that layers can be restored on the meta device | Called by `BaseModelLoader` | Called by `BaseModelLoader` |
-| `restore_layer_on_meta` | Restore layer to model format at start of reload | Called by `initialize_layerwise_reload` | Not called. Online quantized weights already start on meta device via `...OnlineLinearMethod.create_weights` |
-| `initialize_online_processing` | Wrap weight loaders with the `online_process_loader` wrapper, which buffers weights until all layer weights have been loaded | Called by `initialize_layerwise_reload` | Called by `...OnlineLinearMethod.create_weights` |
-| `_layerwise_process` | Process layer once all weights are loaded | Called by `online_process_loader` during loading | Called by `online_process_loader` during loading |
-| `_copy_and_restore_kernel_tensors` | Copy processed weights into original tensor locations to affect compiled cuda graphs, etc. | Called by `_layerwise_process` after `process_weights_after_loading` | Not called. There is no compiled cuda graph yet |
-| `finalize_layerwise_processing` | Catch any layers which did not load all weights (for example attention weights or weights with padding) | Called by `BaseModelLoader` | Called by `BaseModelLoader` |
-
-You can plug into this lifecycle directly by calling the `initialize_layerwise_reload`, loading weights, then calling `finalize_layerwise_processing`:
-
-```python
-from vllm import LLM
-from vllm.model_executor.model_loader.reload import initialize_layerwise_reload, finalize_layerwise_processing
-
-llm = LLM("Qwen/Qwen3-0.6B")
-
-# this model path requires `VLLM_ENABLE_V1_MULTIPROCESSING=0` and is not stable
-model = llm.llm_engine.engine_core.engine_core.model_executor.driver_worker.worker.get_model()
-
-# layerwise reload
-initialize_layerwise_reload(model)
-model.load_weights(...)
-finalize_layerwise_processing(model, llm.model_config)
-```
-
-## Troubleshooting Excessive Memory Usage
-
-Layerwise reloading allows users to incrementally load and process weights as they are loaded into the model. This system relies on buffering layer weights on device until all weights of a layer have been loaded. However, without offloading, this approach necessarily causes excessive buffering if weights are loaded out of order.
-
-For this reason, users must take care as to the order of weights when they are reloading into the model. Weight should be loaded "in order", meaning that each layer's weights are fully loaded before beginning to load the next layer's weights. "Out of order" loading can cause layer weights to stay buffered while other layer weights are loading, leading to excessive memory usage. In the example below, q_proj, k_proj, v_proj, and up_proj are all buffered at the same time, using more memory than if up_proj was loaded after q_proj, k_proj and v_proj.
-
-| Correct Loading | Incorrect Loading |
-| - | - |
-| ![Layerwise](../assets/training/layerwise_good_loading.png) | ![Layerwise](../assets/training/layerwise_bad_loading.png) |
-
-Users will see a warning like the one below if weights are loaded out-of-order.
-
-```console
-WARNING [layerwise.py:198] Allocating 28.5 MB of device memory to buffers to load ["QKVParallelLinear", "MergedColumnParallelLinear"] layers. This extra memory usage can be avoided by ordering weights by their parent layer when reloading.
-```
@@ -7,7 +7,7 @@ reproducible results:
  or enable [batch invariance](../features/batch_invariance.md) to make the outputs insensitive to scheduling.
 - In online mode, you can only enable [batch invariance](../features/batch_invariance.md).

-Example: [examples/features/batch_invariance/reproducibility_offline.py](../../examples/features/batch_invariance/reproducibility_offline.py)
+Example: [examples/offline_inference/reproducibility.py](../../examples/offline_inference/reproducibility.py)

 !!! warning

@@ -138,22 +138,14 @@ When `--api-key` is configured, the following `/v1` endpoints require Bearer tok

 - `/v1/models` - List available models
 - `/v1/chat/completions` - Chat completions
- `/v1/chat/completions/batch` - Batch chat completions
- `/v1/chat/completions/render` - Render chat completion requests
 - `/v1/completions` - Text completions
- `/v1/completions/render` - Render completion requests
 - `/v1/embeddings` - Generate embeddings
 - `/v1/audio/transcriptions` - Audio transcription
 - `/v1/audio/translations` - Audio translation
 - `/v1/messages` - Anthropic-compatible messages API
- `/v1/messages/count_tokens` - Count tokens for Anthropic messages
- `/v1/responses` - Create a response
- `/v1/responses/{response_id}` - Retrieve a response
- `/v1/responses/{response_id}/cancel` - Cancel a response
+- `/v1/responses` - Response management
 - `/v1/score` - Scoring API
 - `/v1/rerank` - Reranking API
- `/v1/load_lora_adapter` - Load a LoRA adapter (can alter model behavior; only available when `--enable-lora` is set and `VLLM_ALLOW_RUNTIME_LORA_UPDATING=True`)
- `/v1/unload_lora_adapter` - Unload a LoRA adapter (can alter model behavior; only available when `--enable-lora` is set and `VLLM_ALLOW_RUNTIME_LORA_UPDATING=True`)

 ### Unprotected Endpoints (No API Key Required)

@@ -163,23 +155,16 @@ The following endpoints **do not require authentication** even when `--api-key`

 - `/invocations` - SageMaker-compatible endpoint (routes to the same inference functions as `/v1` endpoints)
 - `/inference/v1/generate` - Generate completions
- `/generative_scoring` - Generative scoring API
 - `/pooling` - Pooling API
 - `/classify` - Classification API
 - `/score` - Scoring API (non-`/v1` variant)
 - `/rerank` - Reranking API (non-`/v1` variant)

-**Operational control endpoints (only when `"generate"` task is supported):**
+**Operational control endpoints (always enabled):**

 - `/pause` - Pause generation (causes denial of service)
 - `/resume` - Resume generation
- `/is_paused` - Check if generation is paused
 - `/scale_elastic_ep` - Trigger scaling operations
- `/is_scaling_elastic_ep` - Check if scaling is in progress
- `/init_weight_transfer_engine` - Initialize weight transfer engine for RLHF
- `/update_weights` - Update model weights (can alter model behavior)
- `/get_world_size` - Get distributed world size
- `/abort_requests` - Abort in-flight requests (only when `--tokens-only` is also set)

 **Utility endpoints:**

@@ -222,9 +207,9 @@ These endpoints are only available when profiling is enabled and should only be

 An attacker who can reach the vLLM HTTP server can:

-1. **Bypass authentication** by using non-`/v1` endpoints like `/invocations`, `/inference/v1/generate`, `/generative_scoring`, `/pooling`, `/classify`, `/score`, or `/rerank` to run arbitrary inference without credentials
-2. **Cause denial of service** by calling `/pause`, `/scale_elastic_ep`, or `/abort_requests` without a token
-3. **Access operational controls** to manipulate server state (e.g., pausing generation, updating model weights via `/update_weights`)
+1. **Bypass authentication** by using non-`/v1` endpoints like `/invocations`, `/inference/v1/generate`, `/pooling`, `/classify`, `/score`, or `/rerank` to run arbitrary inference without credentials
+2. **Cause denial of service** by calling `/pause` or `/scale_elastic_ep` without a token
+3. **Access operational controls** to manipulate server state (e.g., pausing generation)
 4. **If `--enable-tokenizer-info-endpoint` is set:** Access sensitive tokenizer configuration including chat templates, which may reveal prompt engineering strategies or other implementation details
 5. **If `VLLM_SERVER_DEV_MODE=1` is set:** Execute arbitrary RPC commands via `/collective_rpc`, reset caches, put the engine to sleep, and access detailed server configuration

@@ -303,12 +288,6 @@ To disable the Python code interpreter specifically, omit `code_interpreter` fro

 **Consider a custom implementation**: The GPT-OSS Python tool is a reference implementation. For production deployments, consider implementing a custom code execution sandbox with stricter isolation guarantees. See the [GPT-OSS documentation](https://github.com/openai/gpt-oss?tab=readme-ov-file#python) for guidance.

-## Dynamic LoRA Loading
-
-vLLM supports dynamically loading and unloading LoRA adapters at runtime via the `/v1/load_lora_adapter` and `/v1/unload_lora_adapter` API endpoints. This functionality is **not enabled by default** — it requires both `--enable-lora` and the environment variable `VLLM_ALLOW_RUNTIME_LORA_UPDATING=True` to be set.
-
-**Warning:** Dynamic LoRA loading is not a secure operation and should not be enabled in deployments exposed to untrusted clients. If you must enable dynamic LoRA loading, restrict access to the `/v1/load_lora_adapter` and `/v1/unload_lora_adapter` endpoints to trusted administrators only, using a reverse proxy or network-level access controls. Do not expose these endpoints to end users. For details on configuring LoRA adapters, see the [LoRA Adapters documentation](../features/lora.md).
-
 ## Reporting Security Vulnerabilities

 If you believe you have found a security vulnerability in vLLM, please report it following the project's security policy. For more information on how to report security issues and the project's security policy, please see the [vLLM Security Policy](https://github.com/vllm-project/vllm/blob/main/SECURITY.md).
@@ -15,7 +15,7 @@ compares the generation time for two queries that share the same prefix
 but ask different questions.

 Run:
-python examples/features/automatic_prefix_caching/automatic_prefix_caching_offline.py
+python examples/offline_inference/automatic_prefix_caching.py
 """

 import time
@@ -6,7 +6,7 @@ of a Qwen model using the YARN method (rope_parameters)
 and run a simple chat example.

 Usage:
-    python examples/features/context_extension/context_extension_offline.py
+    python examples/offline_inference/context_extension.py
 """

 from vllm import LLM, RequestOutput, SamplingParams
@@ -3,14 +3,14 @@
 """
 Usage:
 Single node:
-    python examples/features/data_parallel/data_parallel_offline.py \
+    python examples/offline_inference/data_parallel.py \
            --model="ibm-research/PowerMoE-3b" \
            -dp=2 \
            -tp=2

 Multi-node:
    Node 0 (assume the node has ip of 10.99.48.128):
-            python examples/features/data_parallel/data_parallel_offline.py \
+            python examples/offline_inference/data_parallel.py \
                    --model="ibm-research/PowerMoE-3b" \
                    -dp=2 \
                    -tp=2 \
@@ -19,7 +19,7 @@ Multi-node:
                    --dp-master-addr=10.99.48.128 \
                    --dp-master-port=13345
    Node 1:
-            python examples/features/data_parallel/data_parallel_offline.py \
+            python examples/offline_inference/data_parallel.py \
                    --model="ibm-research/PowerMoE-3b" \
                    -dp=2 \
                    -tp=2 \
@@ -3,16 +3,16 @@
 """
 Validates the loading of a model saved with the sharded_state format.
 This script demonstrates how to load a model that was previously saved
-using save_sharded_state_offline.py and validates it by running inference.
+using save_sharded_state.py and validates it by running inference.
 Example usage:
 (First need to save a sharded_state mode)

-python save_sharded_state_offline.py \
+python save_sharded_state.py \
    --model /path/to/load \
    --tensor-parallel-size 8 \
    --output /path/to/save/sharded/model

-python load_sharded_state_offline.py \
+python load_sharded_state.py \
    --model /path/to/saved/sharded/model \
    --load-format sharded_state \
    --tensor-parallel-size 8 \
@@ -9,7 +9,7 @@ This directory contains examples demonstrating how to use custom logits processo
 Demonstrates how to instantiate vLLM with a custom logits processor class that operates at the batch level. The example uses a `DummyLogitsProcessor` that masks out all tokens except a specified `target_token` when passed via `SamplingParams.extra_args`.

 ```bash
-python examples/features/logits_processor/custom.py
+python examples/offline_inference/logits_processor/custom.py
 ```

 ### `custom_req.py` — Request-level logits processor wrapper
@@ -17,7 +17,7 @@ python examples/features/logits_processor/custom.py
 Shows how to wrap a request-level logits processor (which operates on individual requests) to be compatible with vLLM's batch-level logits processing interface.

 ```bash
-python examples/features/logits_processor/custom_req.py
+python examples/offline_inference/logits_processor/custom_req.py
 ```

 ### `custom_req_init.py` — Request-level processor with engine config
@@ -25,7 +25,7 @@ python examples/features/logits_processor/custom_req.py
 A special case of wrapping a request-level logits processor where the processor needs access to engine configuration or model metadata during initialization (e.g., vocabulary size, tokenizer info).

 ```bash
-python examples/features/logits_processor/custom_req_init.py
+python examples/offline_inference/logits_processor/custom_req_init.py
 ```

 ## Key Concepts
@@ -8,7 +8,7 @@ This is a guide to performing batch inference using the OpenAI batch file format

 The OpenAI batch file format consists of a series of json objects on new lines.

-[See here for an example file.](https://github.com/vllm-project/vllm/blob/main/examples/features/openai_batch/openai_example_batch.jsonl)
+[See here for an example file.](https://github.com/vllm-project/vllm/blob/main/examples/offline_inference/openai_batch/openai_example_batch.jsonl)

 Each line represents a separate request. See the [OpenAI package reference](https://platform.openai.com/docs/api-reference/batch/requestInput) for more details.

@@ -30,13 +30,13 @@ We currently support `/v1/chat/completions`, `/v1/embeddings`, and `/v1/score` e
 To follow along with this example, you can download the example batch, or create your own batch file in your working directory.

 ```bash
-wget https://raw.githubusercontent.com/vllm-project/vllm/main/examples/features/openai_batch/openai_example_batch.jsonl
+wget https://raw.githubusercontent.com/vllm-project/vllm/main/examples/offline_inference/openai_batch/openai_example_batch.jsonl
 ```

 Once you've created your batch file it should look like this

 ```bash
-cat features/openai_batch/openai_example_batch.jsonl
+cat offline_inference/openai_batch/openai_example_batch.jsonl
 {"custom_id": "request-1", "method": "POST", "url": "/v1/chat/completions", "body": {"model": "meta-llama/Meta-Llama-3-8B-Instruct", "messages": [{"role": "system", "content": "You are a helpful assistant."},{"role": "user", "content": "Hello world!"}],"max_completion_tokens": 1000}}
 {"custom_id": "request-2", "method": "POST", "url": "/v1/chat/completions", "body": {"model": "meta-llama/Meta-Llama-3-8B-Instruct", "messages": [{"role": "system", "content": "You are an unhelpful assistant."},{"role": "user", "content": "Hello world!"}],"max_completion_tokens": 1000}}
 ```
@@ -49,7 +49,7 @@ You can run the batch with the following command, which will write its results t

 ```bash
 python -m vllm.entrypoints.openai.run_batch \
-    -i features/openai_batch/openai_example_batch.jsonl \
+    -i offline_inference/openai_batch/openai_example_batch.jsonl \
    -o results.jsonl \
    --model meta-llama/Meta-Llama-3-8B-Instruct
 ```
@@ -58,7 +58,7 @@ or use command-line:

 ```bash
 vllm run-batch \
-    -i features/openai_batch/openai_example_batch.jsonl \
+    -i offline_inference/openai_batch/openai_example_batch.jsonl \
    -o results.jsonl \
    --model meta-llama/Meta-Llama-3-8B-Instruct
 ```
@@ -77,11 +77,11 @@ cat results.jsonl

 The batch runner supports remote input and output urls that are accessible via http/https.

-For example, to run against our example input file located at `https://raw.githubusercontent.com/vllm-project/vllm/main/examples/features/openai_batch/openai_example_batch.jsonl`, you can run
+For example, to run against our example input file located at `https://raw.githubusercontent.com/vllm-project/vllm/main/examples/offline_inference/openai_batch/openai_example_batch.jsonl`, you can run

 ```bash
 python -m vllm.entrypoints.openai.run_batch \
-    -i https://raw.githubusercontent.com/vllm-project/vllm/main/examples/features/openai_batch/openai_example_batch.jsonl \
+    -i https://raw.githubusercontent.com/vllm-project/vllm/main/examples/offline_inference/openai_batch/openai_example_batch.jsonl \
    -o results.jsonl \
    --model meta-llama/Meta-Llama-3-8B-Instruct
 ```
@@ -90,7 +90,7 @@ or use command-line:

 ```bash
 vllm run-batch \
-    -i https://raw.githubusercontent.com/vllm-project/vllm/main/examples/features/openai_batch/openai_example_batch.jsonl \
+    -i https://raw.githubusercontent.com/vllm-project/vllm/main/examples/offline_inference/openai_batch/openai_example_batch.jsonl \
    -o results.jsonl \
    --model meta-llama/Meta-Llama-3-8B-Instruct
 ```
@@ -113,13 +113,13 @@ To integrate with cloud blob storage, we recommend using presigned urls.
 To follow along with this example, you can download the example batch, or create your own batch file in your working directory.

 ```bash
-wget https://raw.githubusercontent.com/vllm-project/vllm/main/examples/features/openai_batch/openai_example_batch.jsonl
+wget https://raw.githubusercontent.com/vllm-project/vllm/main/examples/offline_inference/openai_batch/openai_example_batch.jsonl
 ```

 Once you've created your batch file it should look like this

 ```bash
-cat features/openai_batch/openai_example_batch.jsonl
+cat offline_inference/openai_batch/openai_example_batch.jsonl
 {"custom_id": "request-1", "method": "POST", "url": "/v1/chat/completions", "body": {"model": "meta-llama/Meta-Llama-3-8B-Instruct", "messages": [{"role": "system", "content": "You are a helpful assistant."},{"role": "user", "content": "Hello world!"}],"max_completion_tokens": 1000}}
 {"custom_id": "request-2", "method": "POST", "url": "/v1/chat/completions", "body": {"model": "meta-llama/Meta-Llama-3-8B-Instruct", "messages": [{"role": "system", "content": "You are an unhelpful assistant."},{"role": "user", "content": "Hello world!"}],"max_completion_tokens": 1000}}
 ```
@@ -127,7 +127,7 @@ cat features/openai_batch/openai_example_batch.jsonl
 Now upload your batch file to your S3 bucket.

 ```bash
-aws s3 cp features/openai_batch/openai_example_batch.jsonl s3://MY_BUCKET/MY_INPUT_FILE.jsonl
+aws s3 cp offline_inference/openai_batch/openai_example_batch.jsonl s3://MY_BUCKET/MY_INPUT_FILE.jsonl
 ```

 ### Step 2: Generate your presigned urls
@@ -15,7 +15,7 @@ Requirements:
 - transformers

 Run:
-    python examples/features/prompt_embed/prompt_embed_offline.py
+    python examples/offline_inference/prompt_embed_inference.py
 """

 import torch
@@ -6,15 +6,15 @@ This folder provides several example scripts on how to inference Qwen2.5-Omni of

 ```bash
 # Audio + image + video
-python examples/generate/multimodal/qwen2_5_omni/only_thinker.py \
+python examples/offline_inference/qwen2_5_omni/only_thinker.py \
    -q mixed_modalities

 # Read vision and audio inputs from a single video file
-python examples/generate/multimodal/qwen2_5_omni/only_thinker.py \
+python examples/offline_inference/qwen2_5_omni/only_thinker.py \
    -q use_audio_in_video

 # Multiple audios
-python examples/generate/multimodal/qwen2_5_omni/only_thinker.py \
+python examples/offline_inference/qwen2_5_omni/only_thinker.py \
    -q multi_audios
 ```

@@ -24,16 +24,16 @@ You can also test Qwen2.5-Omni on a single modality:

 ```bash
 # Process audio inputs
-python examples/generate/multimodal/audio_language_offline.py \
+python examples/offline_inference/audio_language.py \
    --model-type qwen2_5_omni

 # Process image inputs
-python examples/generate/multimodal/vision_language_offline.py \
+python examples/offline_inference/vision_language.py \
    --modality image \
    --model-type qwen2_5_omni

 # Process video inputs
-python examples/generate/multimodal/vision_language_offline.py \
+python examples/offline_inference/vision_language.py \
    --modality video \
    --model-type qwen2_5_omni
 ```
@@ -9,7 +9,7 @@ Validates that:
 3. Results are deterministic across runs (baseline vs reference).

 Usage:
-    python examples/rl/routed_experts_e2e.py \
+    python examples/offline_inference/routed_experts_e2e.py \
        --model Qwen/Qwen3-30B-A3B \
        --tp 4 \
        --max-model-len 4096 \
@@ -7,7 +7,7 @@ read its own shard rather than the entire checkpoint.

 Example usage:

-python save_sharded_state_offline.py \
+python save_sharded_state.py \
    --model /path/to/load \
    --tensor-parallel-size 8 \
    --output /path/to/save
@@ -7,15 +7,15 @@ no internal lb supported in external_launcher mode.

 To run this example:
 ```bash
-$ torchrun --nproc-per-node=2 examples/features/torchrun/torchrun_dp_example_offline.py
+$ torchrun --nproc-per-node=2 examples/offline_inference/torchrun_dp_example.py
 ```

 With custom parallelism settings:
 ```bash
-$ torchrun --nproc-per-node=8 examples/features/torchrun/torchrun_dp_example_offline.py \
+$ torchrun --nproc-per-node=8 examples/offline_inference/torchrun_dp_example.py \
    --tp-size=2 --pp-size=1 --dp-size=4 --enable-ep
 ```
-"""  # noqa: E501
+"""

 import argparse

@@ -4,7 +4,7 @@
 experimental support for tensor-parallel inference with torchrun,
 see https://github.com/vllm-project/vllm/issues/11400 for
 the motivation and use case for this example.
-run the script with `torchrun --nproc-per-node=4 torchrun_example_offline.py`,
+run the script with `torchrun --nproc-per-node=4 torchrun_example.py`,
 the argument `4` should match the product of `tensor_parallel_size` and
 `pipeline_parallel_size` below. see `tests/distributed/test_torchrun_example.py`
 for the unit test.
@@ -1402,7 +1402,7 @@ def run_mantis(questions: list[str], modality: str) -> ModelRequestData:
 # MiniCPM-V
 def run_minicpmv_base(questions: list[str], modality: str, model_name):
    assert modality in ["image", "video", "image+video"]
-    # If you want to use `MiniCPM-o-2_6` with audio inputs, check `audio_language_offline.py` # noqa
+    # If you want to use `MiniCPM-o-2_6` with audio inputs, check `audio_language.py` # noqa

    # 2.0
    # The official repo doesn't work yet, so we need to use a fork for now
@@ -1,135 +1,135 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-"""
-Test pause/resume with Data Parallel (DP) via HTTP API.
-
-This example demonstrates coordinated pause/resume across multiple DP ranks.
-The pause synchronizes across all DP engines via all-reduce.
-
-Prerequisites:
-    Start a vLLM server with data parallelism:
-
-    $ VLLM_SERVER_DEV_MODE=1 vllm serve facebook/opt-125m \
-        --enforce-eager \
-        --data-parallel-size 4 \
-        --tensor-parallel-size 1
-
-    Then run this script:
-
-    $ python data_parallel_pause_resume.py
-
-The test verifies pause works by:
-1. Starting a streaming generation request
-2. Pausing the server mid-generation
-3. Sleeping for PAUSE_DURATION seconds
-4. Resuming the server
-5. Verifying there was a gap in token generation matching the pause duration
-"""
-
-import argparse
-import threading
-import time
-
-import requests
-from openai import OpenAI
-
-BASE_URL = "http://localhost:8000"
-MODEL_NAME = "facebook/opt-125m"
-PAUSE_DURATION = 3.0
-
-
-def pause_generation(base_url: str, mode: str = "keep") -> None:
-    """Pause generation via HTTP endpoint."""
-    url = f"{base_url}/pause"
-    response = requests.post(url, params={"mode": mode}, timeout=60)
-    response.raise_for_status()
-    print("Server paused")
-
-
-def resume_generation(base_url: str) -> None:
-    """Resume generation via HTTP endpoint."""
-    url = f"{base_url}/resume"
-    response = requests.post(url, timeout=60)
-    response.raise_for_status()
-    print("Server resumed")
-
-
-def main():
-    parser = argparse.ArgumentParser()
-    parser.add_argument("--base-url", default=BASE_URL)
-    parser.add_argument("--model", default=MODEL_NAME)
-    args = parser.parse_args()
-
-    client = OpenAI(
-        base_url=f"{args.base_url}/v1",
-        api_key="EMPTY",
-    )
-
-    prompt = "Write a long story about a dragon. Once upon a time"
-    token_times: list[float] = []
-    pause_token_idx = 0
-    pause_triggered = threading.Event()
-
-    def generator_thread():
-        """Stream tokens and record timestamps."""
-        stream = client.completions.create(
-            model=args.model,
-            prompt=prompt,
-            max_tokens=50,
-            stream=True,
-        )
-        for chunk in stream:
-            if chunk.choices[0].text:
-                token_times.append(time.monotonic())
-                token_count = len(token_times)
-                print(f"Token {token_count}: {chunk.choices[0].text!r}")
-
-                # Signal controller after some tokens
-                if token_count >= 5 and not pause_triggered.is_set():
-                    pause_triggered.set()
-
-    def controller_thread():
-        """Pause and resume the server."""
-        nonlocal pause_token_idx
-
-        # Wait for some tokens
-        pause_triggered.wait()
-
-        print(f"\nPausing server (keep mode) at token {len(token_times)}...")
-        pause_generation(args.base_url, mode="keep")
-        pause_token_idx = len(token_times)
-        print(f"Sleeping for {PAUSE_DURATION}s...")
-
-        time.sleep(PAUSE_DURATION)
-
-        print("Resuming server...")
-        resume_generation(args.base_url)
-        print("Resumed!\n")
-
-    # Run both threads
-    gen_thread = threading.Thread(target=generator_thread)
-    ctrl_thread = threading.Thread(target=controller_thread)
-
-    gen_thread.start()
-    ctrl_thread.start()
-
-    gen_thread.join()
-    ctrl_thread.join()
-
-    # Check gap at the pause point
-    if pause_token_idx < len(token_times):
-        pause_gap = token_times[pause_token_idx] - token_times[pause_token_idx - 1]
-        print(
-            f"\nGap after pause (token {pause_token_idx} -> "
-            f"{pause_token_idx + 1}): {pause_gap:.3f}s"
-        )
-        if pause_gap >= PAUSE_DURATION * 0.9:
-            print("Test passed! Pause synchronized across DP ranks.")
-        else:
-            print(f"Test failed! Expected ~{PAUSE_DURATION}s gap, got {pause_gap:.3f}s")
-    else:
-        print("Test failed! No tokens were generated after resuming.")
-
-
-if __name__ == "__main__":
-    main()
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+"""
+Test pause/resume with Data Parallel (DP) via HTTP API.
+
+This example demonstrates coordinated pause/resume across multiple DP ranks.
+The pause synchronizes across all DP engines via all-reduce.
+
+Prerequisites:
+    Start a vLLM server with data parallelism:
+
+    $ VLLM_SERVER_DEV_MODE=1 vllm serve facebook/opt-125m \
+        --enforce-eager \
+        --data-parallel-size 4 \
+        --tensor-parallel-size 1
+
+    Then run this script:
+
+    $ python data_parallel_pause_resume.py
+
+The test verifies pause works by:
+1. Starting a streaming generation request
+2. Pausing the server mid-generation
+3. Sleeping for PAUSE_DURATION seconds
+4. Resuming the server
+5. Verifying there was a gap in token generation matching the pause duration
+"""
+
+import argparse
+import threading
+import time
+
+import requests
+from openai import OpenAI
+
+BASE_URL = "http://localhost:8000"
+MODEL_NAME = "facebook/opt-125m"
+PAUSE_DURATION = 3.0
+
+
+def pause_generation(base_url: str, mode: str = "keep") -> None:
+    """Pause generation via HTTP endpoint."""
+    url = f"{base_url}/pause"
+    response = requests.post(url, params={"mode": mode}, timeout=60)
+    response.raise_for_status()
+    print("Server paused")
+
+
+def resume_generation(base_url: str) -> None:
+    """Resume generation via HTTP endpoint."""
+    url = f"{base_url}/resume"
+    response = requests.post(url, timeout=60)
+    response.raise_for_status()
+    print("Server resumed")
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--base-url", default=BASE_URL)
+    parser.add_argument("--model", default=MODEL_NAME)
+    args = parser.parse_args()
+
+    client = OpenAI(
+        base_url=f"{args.base_url}/v1",
+        api_key="EMPTY",
+    )
+
+    prompt = "Write a long story about a dragon. Once upon a time"
+    token_times: list[float] = []
+    pause_token_idx = 0
+    pause_triggered = threading.Event()
+
+    def generator_thread():
+        """Stream tokens and record timestamps."""
+        stream = client.completions.create(
+            model=args.model,
+            prompt=prompt,
+            max_tokens=50,
+            stream=True,
+        )
+        for chunk in stream:
+            if chunk.choices[0].text:
+                token_times.append(time.monotonic())
+                token_count = len(token_times)
+                print(f"Token {token_count}: {chunk.choices[0].text!r}")
+
+                # Signal controller after some tokens
+                if token_count >= 5 and not pause_triggered.is_set():
+                    pause_triggered.set()
+
+    def controller_thread():
+        """Pause and resume the server."""
+        nonlocal pause_token_idx
+
+        # Wait for some tokens
+        pause_triggered.wait()
+
+        print(f"\nPausing server (keep mode) at token {len(token_times)}...")
+        pause_generation(args.base_url, mode="keep")
+        pause_token_idx = len(token_times)
+        print(f"Sleeping for {PAUSE_DURATION}s...")
+
+        time.sleep(PAUSE_DURATION)
+
+        print("Resuming server...")
+        resume_generation(args.base_url)
+        print("Resumed!\n")
+
+    # Run both threads
+    gen_thread = threading.Thread(target=generator_thread)
+    ctrl_thread = threading.Thread(target=controller_thread)
+
+    gen_thread.start()
+    ctrl_thread.start()
+
+    gen_thread.join()
+    ctrl_thread.join()
+
+    # Check gap at the pause point
+    if pause_token_idx < len(token_times):
+        pause_gap = token_times[pause_token_idx] - token_times[pause_token_idx - 1]
+        print(
+            f"\nGap after pause (token {pause_token_idx} -> "
+            f"{pause_token_idx + 1}): {pause_gap:.3f}s"
+        )
+        if pause_gap >= PAUSE_DURATION * 0.9:
+            print("Test passed! Pause synchronized across DP ranks.")
+        else:
+            print(f"Test failed! Expected ~{PAUSE_DURATION}s gap, got {pause_gap:.3f}s")
+    else:
+        print("Test failed! No tokens were generated after resuming.")
+
+
+if __name__ == "__main__":
+    main()
@@ -12,7 +12,7 @@ from vllm.v1.metrics.loggers import AggregatedLoggingStatLogger
 """
 To run this example, run the following commands simultaneously with
 different CUDA_VISIBLE_DEVICES:
-    python examples/features/data_parallel/multi_instance_data_parallel.py
+    python examples/online_serving/multi_instance_data_parallel.py

    vllm serve ibm-research/PowerMoE-3b -dp 2 -dpr 1 \
        --data-parallel-address 127.0.0.1 --data-parallel-rpc-port 62300 \
@@ -25,6 +25,7 @@ import os
 import pybase64 as base64
 import requests
 from openai import OpenAI
+from utils import get_first_model

 from vllm.utils.argparse_utils import FlexibleArgumentParser

@@ -406,7 +407,7 @@ def parse_args():

 def main(args) -> None:
    chat_type = args.chat_type
-    model = client.models.list().data[0].id
+    model = get_first_model(client)
    example_function_map[chat_type](model, args.max_completion_tokens)


@@ -19,6 +19,7 @@ Environment variables:
 """

 from openai import OpenAI
+from utils import get_first_model


 def example_no_filter():
@@ -29,7 +30,7 @@ def example_no_filter():

    base_url = "http://0.0.0.0:8000/v1"
    client = OpenAI(base_url=base_url, api_key="empty")
-    model = client.models.list().data[0].id
+    model = get_first_model(client)

    response = client.responses.create(
        model=model,
@@ -58,7 +59,7 @@ def example_wildcard():

    base_url = "http://0.0.0.0:8000/v1"
    client = OpenAI(base_url=base_url, api_key="empty")
-    model = client.models.list().data[0].id
+    model = get_first_model(client)

    response = client.responses.create(
        model=model,
@@ -94,7 +95,7 @@ def example_specific_tools():

    base_url = "http://0.0.0.0:8000/v1"
    client = OpenAI(base_url=base_url, api_key="empty")
-    model = client.models.list().data[0].id
+    model = get_first_model(client)

    response = client.responses.create(
        model=model,
@@ -125,7 +126,7 @@ def example_object_format():

    base_url = "http://0.0.0.0:8000/v1"
    client = OpenAI(base_url=base_url, api_key="empty")
-    model = client.models.list().data[0].id
+    model = get_first_model(client)

    response = client.responses.create(
        model=model,
@@ -14,6 +14,7 @@ vllm serve Qwen/Qwen3-1.7B --reasoning-parser qwen3 \
 import json

 from openai import OpenAI
+from utils import get_first_model


 def get_weather(latitude: float, longitude: float) -> str:
@@ -50,7 +51,7 @@ input_messages = [
 def main():
    base_url = "http://0.0.0.0:8000/v1"
    client = OpenAI(base_url=base_url, api_key="empty")
-    model = client.models.list().data[0].id
+    model = get_first_model(client)
    response = client.responses.create(
        model=model, input=input_messages, tools=tools, tool_choice="required"
    )
@@ -15,7 +15,7 @@ vllm serve meta-llama/Llama-3.2-1B-Instruct \
  --enable-prompt-embeds

 Run the client:
-python examples/features/prompt_embed/prompt_embed_inference_with_openai_client.py
+python examples/online_serving/prompt_embed_inference_with_openai_client.py

 Model: meta-llama/Llama-3.2-1B-Instruct
 Note: This model is gated on Hugging Face Hub.
@@ -20,7 +20,7 @@ vllm serve deepseek-ai/DeepSeek-R1-Distill-Qwen-7B \
 If you want to run this script standalone with `uv`, you can use the following:

 ```bash
-uvx --from git+https://github.com/vllm-project/vllm#subdirectory=examples/features/structured_outputs \
+uvx --from git+https://github.com/vllm-project/vllm#subdirectory=examples/online_serving/structured_outputs \
    structured-outputs
 ```

@@ -34,19 +34,19 @@ See [feature docs](https://docs.vllm.ai/en/latest/features/structured_outputs.ht
 Run all constraints, non-streaming:

 ```bash
-uv run structured_outputs_offline.py
+uv run structured_outputs.py
 ```

 Run all constraints, streaming:

 ```bash
-uv run structured_outputs_offline.py --stream
+uv run structured_outputs.py --stream
 ```

 Run certain constraints, for example `structural_tag` and `regex`, streaming:

 ```bash
-uv run structured_outputs_offline.py \
+uv run structured_outputs.py \
    --constraint structural_tag regex \
    --stream
 ```
@@ -54,5 +54,5 @@ uv run structured_outputs_offline.py \
 Run all constraints, with reasoning models and streaming:

 ```bash
-uv run structured_outputs_offline.py --reasoning --stream
+uv run structured_outputs.py --reasoning --stream
 ```
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
khluu	67c74e5980	p Signed-off-by: khluu <khluu000@gmail.com>	2026-05-01 14:47:49 -07:00
khluu	455839ec66	p Signed-off-by: khluu <khluu000@gmail.com>	2026-05-01 14:28:31 -07:00
Yongye Zhuandkhluu	513b5f82dd	[DSV4] Add knob to enable pre-attn gemm (#41443 ) Signed-off-by: Yongye Zhu <zyy1102000@gmail.com> (cherry picked from commit `bcf5cac9fb`)	2026-05-01 12:24:17 -07:00
khluu	135722d58d	Revert "." This reverts commit `a5e0ea09ea`.	2026-05-01 12:23:33 -07:00
Rohan Potdarandkhluu	7758daca7c	[ROCm][Bugfix][GPTOSS]: fix input_ids and expert_map args for quark w4a8 gptoss (#41165 ) Signed-off-by: Rohan138 <rohanpotdar138@gmail.com> (cherry picked from commit `3795d7acf4`)	2026-05-01 12:17:45 -07:00
Isotr0pyandkhluu	5f79634b3b	[Perf] Intergrate Tile Kernels `head_compute_mix_kernel` for Deepseek-V4 (#41255 ) Signed-off-by: Isotr0py <Isotr0py@outlook.com> Co-authored-by: Roger Wang <hey@rogerw.io> (cherry picked from commit `a9484dac7b`)	2026-05-01 12:17:28 -07:00
Bugen Zhaoandkhluu	05adde9955	[Bugfix] Pass reasoning parser kwargs to structured output (#41199 ) Signed-off-by: Bugen Zhao <i@bugenzhao.com> Co-authored-by: wang.yuqi <yuqi.wang@daocloud.io> (cherry picked from commit `a07642667d`)	2026-05-01 12:14:44 -07:00
Jiangyun Zhuandkhluu	9cdcda5fc2	[Bugfix] BailingMoeV2.5: rotate full qk_rope_head_dim in MLA RoPE (#41185 ) Signed-off-by: zjy0516 <riverclouds.zhu@qq.com> (cherry picked from commit `6d7d4da99e`)	2026-05-01 12:13:50 -07:00
JiangWeixiangandkhluu	a5d4a264dd	[Bugfix ] fix bailing_moe_linear (#40859 ) Signed-off-by: ghphotoframe <854746559@qq.com> (cherry picked from commit `ed57f77192`)	2026-05-01 12:13:41 -07:00
Yongye Zhuandkhluu	a5e0ea09ea	. Signed-off-by: Yongye Zhu <zyy1102000@gmail.com> (cherry picked from commit `db0c9d52d4`) Signed-off-by: khluu <khluu000@gmail.com>	2026-04-30 23:16:02 -07:00
Yongye Zhuandkhluu	28a1a85d89	[DSV4] Add BF16 and MXFP8 A2A support for flashinfer a2a one sided (#40960 ) Signed-off-by: Yongye Zhu <zyy1102000@gmail.com> Signed-off-by: Zijing Liu <liuzijing2014@gmail.com> Co-authored-by: Zijing Liu <liuzijing2014@users.noreply.github.com> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> (cherry picked from commit `b4806c8ee1`)	2026-04-30 21:39:14 -07:00
Yongye Zhuandkhluu	82ccd9f3b2	[Bugfix] Fix persistent_topk cooperative deadlock at TopK=1024 (#41189 ) Signed-off-by: Yongye Zhu <zyy1102000@gmail.com> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> (cherry picked from commit `a749a33d8d`)	2026-04-30 21:39:14 -07:00
Thien Tranandkhluu	d9f3961abb	[DSv4] Use `cvt` PTX for FP32->FP4 conversion (#41015 ) Signed-off-by: Thien Tran <gau.nernst@yahoo.com.sg> (cherry picked from commit `296741d025`)	2026-04-30 21:39:14 -07:00
Jee Jee Liandkhluu	4fee1390f4	[Bugfix] Fix repeated DSv4 RoPE cache initialization (#41148 ) Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> (cherry picked from commit `9d8ad5b408`)	2026-04-30 21:39:14 -07:00
Chaunceyandkhluu	bf590b6a24	[Bugfix] DSV32/V4 add missing type conversion for non-streaming tool calls (#41198 ) Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com> (cherry picked from commit `762022cafb`)	2026-04-30 21:39:13 -07:00
Wei Zhaoandkhluu	66df2c03f5	[Bugfix] Fix max_num_batched_token not captured in cuda graph (#40734 ) Signed-off-by: wzhao18 <wzhao18.sz@gmail.com> Signed-off-by: Wei Zhao <51183510+wzhao18@users.noreply.github.com> Co-authored-by: Wei Zhao (Engrg-Hardware 1) <weizha@login-bia02.bia.clusters.nvidia.com> (cherry picked from commit `8b49cf3a37`)	2026-04-30 21:39:13 -07:00
Jiangyun Zhuandkhluu	17896d817b	[Bugfix] fix inductor error for dpsk v4 (#41135 ) Signed-off-by: zjy0516 <riverclouds.zhu@qq.com> (cherry picked from commit `2ae73c758c`)	2026-04-30 21:39:13 -07:00
Wei Zhaoandkhluu	1a131a066c	[Bugfix] Fix Deepseek V4 import error due to AOT compile cache loading (#41090 ) Signed-off-by: wzhao18 <wzhao18.sz@gmail.com> Signed-off-by: Wei Zhao <51183510+wzhao18@users.noreply.github.com> (cherry picked from commit `803b9d7881`)	2026-04-30 21:39:13 -07:00
qizixiandkhluu	6fffb0b8c8	[DSV4] Align aux stream API with DeepseekV4DecoderLayer (#41171 ) Signed-off-by: zixi-qi <zixi@inferact.ai> (cherry picked from commit `6fb3f7b46b`)	2026-04-30 21:39:13 -07:00
Nick Hillandkhluu	a076bd1e67	[Core] Account for `num_gpu_blocks_override` in `max_model_len` checks (#41069 ) Signed-off-by: Nick Hill <nickhill123@gmail.com> (cherry picked from commit `e68fa1b90a`)	2026-04-30 21:39:13 -07:00
Yongye Zhuandkhluu	c244b6fc83	[DSV4] Enable Multi-stream for Pre-Attn GEMM (#41061 ) Signed-off-by: Yongye Zhu <zyy1102000@gmail.com> (cherry picked from commit `5aa371dc8e`)	2026-04-30 21:39:13 -07:00
Jee Jee Liandkhluu	15cc87d2fe	[Model][DSV4] Support base model (#41006 ) Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> (cherry picked from commit `2c8b76c5cb`)	2026-04-30 21:39:13 -07:00
youkaichaoandkhluu	c68fa39ced	Auto-disable expandable_segments around cumem memory pool (#40812 ) Signed-off-by: youkaichao <youkaichao@gmail.com> Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> (cherry picked from commit `2ce95a761b`)	2026-04-30 21:39:13 -07:00
Yongye ZhuandGitHub	05ebca5250	Temporary disable persistent topk (#41442 ) Signed-off-by: Yongye Zhu <zyy1102000@gmail.com>	2026-04-30 20:37:19 -07:00
Zhewen Liandkhluu	88d34c6409	[Docker] Install numactl CLI in CUDA runtime image (#41032 ) Signed-off-by: Zhewen Li <zhewenli@inferact.ai> Co-authored-by: Zhewen Li <zhewenli@inferact.ai> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> (cherry picked from commit `6232fb4b66`)	2026-04-27 12:04:32 -07:00
Yongye Zhuandkhluu	b8160878f0	[DSV4] Add silu clamp limit to shared expert (#40950 ) Signed-off-by: Yongye Zhu <zyy1102000@gmail.com> (cherry picked from commit `706a04d34b`)	2026-04-27 00:40:46 -07:00
Dao007foreverandkhluu	84c276d7ea	[Bugfix] Cap SWA/chunked-local runtime admission to startup pool-sizing bound (#40946 ) Signed-off-by: Dao Le <Dao007forever@gmail.com> Signed-off-by: Nick Hill <nickhill123@gmail.com> Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: Nick Hill <nickhill123@gmail.com> (cherry picked from commit `7b1bc0a3eb`)	2026-04-26 21:43:00 -07:00
khluu	5eb3657578	Revert "[Frontend] Remove frontend pooling multi task support. (#37861 )" This reverts commit `d2e2e856ad`.	2026-04-26 21:42:15 -07:00