use newtype for logprobs count (-1 for all)

Signed-off-by: Bugen Zhao <i@bugenzhao.com>

.buildkite/hardware_tests/intel_xpu_ci/test-intel.yaml

@@ -21,6 +21,10 @@ steps:
     timeout_in_minutes: 30
     optional: true
     device: intel_gpu
+    agent_tags:
+      label: production
+      gpu: 2+
+      mem: 24+
     no_plugin: true
     env:
       REGISTRY: "public.ecr.aws/q9t5s3a7"
@@ -38,6 +42,10 @@ steps:
     timeout_in_minutes: 30
     optional: true
     device: intel_gpu
+    agent_tags:
+      label: production
+      gpu: 1+
+      mem: 16+
     no_plugin: true
     env:
       REGISTRY: "public.ecr.aws/q9t5s3a7"
@@ -55,6 +63,10 @@ steps:
     timeout_in_minutes: 30
     optional: true
     device: intel_gpu
+    agent_tags:
+      label: production
+      gpu: 1+
+      mem: 16+
     no_plugin: true
     env:
       REGISTRY: "public.ecr.aws/q9t5s3a7"

.buildkite/intel_jobs/basic_correctness.yaml

@@ -5,6 +5,10 @@ steps:
 - label: XPU Sleep Mode
   timeout_in_minutes: 30
   device: intel_gpu
+  agent_tags:
+    label: production
+    gpu: 1+
+    mem: 16+
   no_plugin: true
   working_dir: "."
   env:

.buildkite/intel_jobs/engine_intel.yaml

@@ -5,6 +5,10 @@ steps:
 - label: Engine (1 GPU)
   timeout_in_minutes: 30
   device: intel_gpu
+  agent_tags:
+    label: production
+    gpu: 1+
+    mem: 16+
   no_plugin: true
   working_dir: "."
   env:

.buildkite/intel_jobs/expert_parallelism_intel.yaml

@@ -6,6 +6,10 @@ steps:
   key: eplb-algorithm
   timeout_in_minutes: 45
   device: intel_gpu
+  agent_tags:
+    label: production
+    gpu: 1+
+    mem: 16+
   no_plugin: true
   working_dir: "."
   env:

.buildkite/intel_jobs/kernels_intel.yaml

@@ -5,6 +5,10 @@ steps:
 - label: vLLM IR Tests
   timeout_in_minutes: 30
   device: intel_gpu
+  agent_tags:
+    label: production
+    gpu: 1+
+    mem: 16+
   no_plugin: true
   working_dir: "."
   env:

.buildkite/intel_jobs/lora_intel.yaml

@@ -5,6 +5,10 @@ steps:
 - label: LoRA Runtime + Utils
   timeout_in_minutes: 45
   device: intel_gpu
+  agent_tags:
+    label: production
+    gpu: 1+
+    mem: 24+
   no_plugin: true
   working_dir: "."
   env:
@@ -34,6 +38,10 @@ steps:
 - label: LoRA Fused/MoE Kernels
   timeout_in_minutes: 45
   device: intel_gpu
+  agent_tags:
+    label: production
+    gpu: 1+
+    mem: 16+
   no_plugin: true
   working_dir: "."
   env:
@@ -54,6 +62,10 @@ steps:
 - label: LoRA Punica Kernels
   timeout_in_minutes: 45
   device: intel_gpu
+  agent_tags:
+    label: production
+    gpu: 1+
+    mem: 16+
   no_plugin: true
   working_dir: "."
   env:
@@ -74,6 +86,10 @@ steps:
 - label: LoRA Punica FP8/XPU Ops
   timeout_in_minutes: 45
   device: intel_gpu
+  agent_tags:
+    label: production
+    gpu: 1+
+    mem: 16+
   no_plugin: true
   working_dir: "."
   env:
@@ -94,6 +110,10 @@ steps:
 - label: LoRA Models
   timeout_in_minutes: 45
   device: intel_gpu
+  agent_tags:
+    label: production
+    gpu: 2+
+    mem: 24+
   no_plugin: true
   working_dir: "."
   env:
@@ -117,6 +137,10 @@ steps:
 - label: LoRA Multimodal
   timeout_in_minutes: 45
   device: intel_gpu
+  agent_tags:
+    label: production
+    gpu: 1+
+    mem: 16+
   no_plugin: true
   working_dir: "."
   env:

.buildkite/intel_jobs/misc_intel.yaml

@@ -5,6 +5,10 @@ steps:
 - label: V1 Core + KV + Metrics
   timeout_in_minutes: 30
   device: intel_gpu
+  agent_tags:
+    label: production
+    gpu: 1+
+    mem: 16+
   no_plugin: true
   working_dir: "."
   env:
@@ -31,6 +35,10 @@ steps:
 - label: V1 Sample + Logits
   timeout_in_minutes: 30
   device: intel_gpu
+  agent_tags:
+    label: production
+    gpu: 1+
+    mem: 16+
   no_plugin: true
   working_dir: "."
   env:
@@ -71,6 +79,10 @@ steps:
 - label: XPU CPU Offload
   timeout_in_minutes: 60
   device: intel_gpu
+  agent_tags:
+    label: production
+    gpu: 1+
+    mem: 16+
   no_plugin: true
   working_dir: "."
   env:
@@ -95,6 +107,10 @@ steps:
   key: regression
   timeout_in_minutes: 30
   device: intel_gpu
+  agent_tags:
+    label: production
+    gpu: 1+
+    mem: 16+
   no_plugin: true
   working_dir: "."
   env:
@@ -126,6 +142,10 @@ steps:
   timeout_in_minutes: 30
   num_devices: 2
   device: intel_gpu
+  agent_tags:
+    label: production
+    gpu: 2+
+    mem: 16+
   no_plugin: true
   working_dir: "."
   env:
@@ -157,6 +177,10 @@ steps:
   key: async-engine-inputs-utils-worker
   timeout_in_minutes: 30
   device: intel_gpu
+  agent_tags:
+    label: production
+    gpu: 1+
+    mem: 24+
   no_plugin: true
   working_dir: "."
   env:

.buildkite/intel_jobs/model_runner_v2_intel.yaml

@@ -5,6 +5,10 @@ steps:
 - label: Model Runner V2 Core Tests (Intel)
   timeout_in_minutes: 45
   device: intel_gpu
+  agent_tags:
+    label: production
+    gpu: 2+
+    mem: 16+
   no_plugin: true
   working_dir: "."
   env:
@@ -30,6 +34,10 @@ steps:
 - label: Model Runner V2 Examples (Intel)
   timeout_in_minutes: 45
   device: intel_gpu
+  agent_tags:
+    label: production
+    gpu: 1+
+    mem: 24+
   no_plugin: true
   working_dir: "."
   env:

.buildkite/intel_jobs/models_multimodal_intel.yaml

@@ -6,6 +6,10 @@ steps:
   key: multi-modal-models-standard-1-qwen2
   timeout_in_minutes: 45
   device: intel_gpu
+  agent_tags:
+    label: production
+    gpu: 1+
+    mem: 16+
   no_plugin: true
   working_dir: "."
   env:
@@ -27,6 +31,10 @@ steps:
   key: multi-modal-models-standard-2-qwen3-gemma
   timeout_in_minutes: 45
   device: intel_gpu
+  agent_tags:
+    label: production
+    gpu: 1+
+    mem: 16+
   no_plugin: true
   working_dir: "."
   env:
@@ -47,6 +55,10 @@ steps:
   key: multi-modal-models-standard-3-llava-qwen2-vl
   timeout_in_minutes: 45
   device: intel_gpu
+  agent_tags:
+    label: production
+    gpu: 1+
+    mem: 24+
   no_plugin: true
   working_dir: "."
   env:
@@ -68,6 +80,10 @@ steps:
   key: multi-modal-models-standard-4-other-whisper
   timeout_in_minutes: 45
   device: intel_gpu
+  agent_tags:
+    label: production
+    gpu: 1+
+    mem: 16+
   no_plugin: true
   working_dir: "."
   env:
@@ -88,6 +104,10 @@ steps:
   key: multi-modal-processor
   timeout_in_minutes: 45
   device: intel_gpu
+  agent_tags:
+    label: production
+    gpu: 1+
+    mem: 16+
   no_plugin: true
   working_dir: "."
   env:

.buildkite/intel_jobs/test-intel.yaml

@@ -19,6 +19,10 @@ steps:
       - image-build-xpu
     timeout_in_minutes: 30
     device: intel_gpu
+    agent_tags:
+      label: production
+      gpu: 2+
+      mem: 24+
     no_plugin: true
     env:
       REGISTRY: "public.ecr.aws/q9t5s3a7"
@@ -49,6 +53,10 @@ steps:
       - image-build-xpu
     timeout_in_minutes: 30
     device: intel_gpu
+    agent_tags:
+      label: production
+      gpu: 1+
+      mem: 16+
     no_plugin: true
     env:
       REGISTRY: "public.ecr.aws/q9t5s3a7"
@@ -74,6 +82,10 @@ steps:
       - image-build-xpu
     timeout_in_minutes: 30
     device: intel_gpu
+    agent_tags:
+      label: production
+      gpu: 1+
+      mem: 16+
     no_plugin: true
     env:
       REGISTRY: "public.ecr.aws/q9t5s3a7"
@@ -93,6 +105,10 @@ steps:
       - image-build-xpu
     timeout_in_minutes: 30
     device: intel_gpu
+    agent_tags:
+      label: production
+      gpu: 1+
+      mem: 16+
     no_plugin: true
     env:
       REGISTRY: "public.ecr.aws/q9t5s3a7"

.buildkite/scripts/install-kv-connectors.sh

@@ -4,6 +4,11 @@
 
 set -euo pipefail
 
+if python3 -c "import torch; raise SystemExit(0 if torch.version.hip is not None else 1)"; then
+    uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt
+    exit 0
+fi
+
 REQUIREMENTS_FILE="${KV_CONNECTORS_REQUIREMENTS:-/vllm-workspace/requirements/kv_connectors.txt}"
 
 uv pip install --system -r "${REQUIREMENTS_FILE}"

.buildkite/test-amd.yaml

@@ -1594,9 +1594,10 @@ steps:
 #----------------------------------------------------------  mi300 · kernels  ----------------------------------------------------------#
 
 - label: Kernels Attention Test %N # TBD
-  timeout_in_minutes: 180
+  timeout_in_minutes: 55
   mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
   agent_pool: mi300_1
+  optional: true
   parallelism: 2
   working_dir: "/vllm-workspace/tests"
   source_file_dependencies:
@@ -1627,10 +1628,11 @@ steps:
   - pytest -v -s kernels/core --ignore=kernels/core/test_minimax_reduce_rms.py  kernels/test_concat_mla_q.py kernels/test_top_k_per_row.py
 
 - label: Kernels MoE Test %N # TBD
-  timeout_in_minutes: 180
+  timeout_in_minutes: 50
   mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
   agent_pool: mi300_1
-  parallelism: 4
+  optional: true
+  parallelism: 5
   working_dir: "/vllm-workspace/tests"
   source_file_dependencies:
   - csrc/quantization/cutlass_w8a8/moe/
@@ -2120,9 +2122,10 @@ steps:
   - pytest -v -s v1/e2e/spec_decode -k "draft_model or no_sync or batch_inference"
 
 - label: Spec Decode Eagle # TBD
-  timeout_in_minutes: 180
+  timeout_in_minutes: 45
   mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
   agent_pool: mi300_1
+  optional: true
   working_dir: "/vllm-workspace/tests"
   source_file_dependencies:
   - vllm/v1/spec_decode/
@@ -3040,7 +3043,7 @@ steps:
 #----------------------------------------------------------  mi355 · kernels  ----------------------------------------------------------#
 
 - label: Kernels (B200-MI355) # TBD
-  timeout_in_minutes: 180
+  timeout_in_minutes: 15
   mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
   agent_pool: mi355_1
   working_dir: "/vllm-workspace/"
@@ -3064,11 +3067,10 @@ steps:
   - pytest -v -s tests/kernels/attention/test_attention_selector.py
 
 - label: Kernels Attention Test %N # TBD
-  timeout_in_minutes: 180
+  timeout_in_minutes: 60
   mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
   agent_pool: mi355_1
   parallelism: 2
-  optional: true
   working_dir: "/vllm-workspace/tests"
   source_file_dependencies:
   - csrc/attention/
@@ -3082,10 +3084,10 @@ steps:
   - pytest -v -s kernels/attention --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
 
 - label: Kernels MoE Test %N # TBD
-  timeout_in_minutes: 180
+  timeout_in_minutes: 50
   mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
   agent_pool: mi355_1
-  parallelism: 4
+  parallelism: 5
   working_dir: "/vllm-workspace/tests"
   source_file_dependencies:
   - csrc/quantization/cutlass_w8a8/moe/

.buildkite/test_areas/kernels.yaml

@@ -74,6 +74,20 @@ steps:
   commands:
     - pytest -v -s kernels/attention --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
   parallelism: 2
+  mirror:
+    amd:
+      device: mi325_1
+      timeout_in_minutes: 55
+      depends_on:
+      - image-build-amd
+      source_file_dependencies:
+      - csrc/attention/
+      - vllm/v1/attention
+      - vllm/model_executor/layers/attention
+      - tests/kernels/attention
+      - vllm/_aiter_ops.py
+      - vllm/envs.py
+      - vllm/platforms/rocm.py
 
 - label: Kernels Attention DiffKV Test (H100)
   key: kernels-attention-diffkv-test-h100
@@ -128,6 +142,22 @@ steps:
     - pytest -v -s kernels/moe --ignore=kernels/moe/test_modular_oai_triton_moe.py --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
     - pytest -v -s kernels/moe/test_modular_oai_triton_moe.py --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
   parallelism: 5
+  mirror:
+    amd:
+      device: mi325_1
+      timeout_in_minutes: 50
+      source_file_dependencies:
+      - csrc/quantization/cutlass_w8a8/moe/
+      - csrc/moe/
+      - tests/kernels/moe
+      - vllm/model_executor/layers/fused_moe/
+      - vllm/distributed/device_communicators/
+      - vllm/envs.py
+      - vllm/config
+      - vllm/_aiter_ops.py
+      - vllm/platforms/rocm.py
+      depends_on:
+      - image-build-amd
 
 - label: Kernels Mamba Test
   key: kernels-mamba-test

.buildkite/test_areas/lm_eval.yaml

@@ -109,6 +109,7 @@ steps:
       - image-build-amd
       commands:
       - export VLLM_WORKER_MULTIPROC_METHOD=spawn
+      - export PYTORCH_ROCM_ARCH=gfx942 # Limit Quark compilation to save time
       - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-mi3xx.txt
 
 - label: MoE Refactor Integration Test (H100 - TEMPORARY)

.buildkite/test_areas/misc.yaml

@@ -105,6 +105,12 @@ steps:
     # Integration test for streaming correctness (requires special branch).
     - pip install -U git+https://github.com/robertgshaw2-redhat/lm-evaluation-harness.git@streaming-api
     - pytest -v -s entrypoints/openai/correctness/test_lmeval.py::test_lm_eval_accuracy_v1_engine
+  mirror:
+    amd:
+      device: mi325_1
+      timeout_in_minutes: 60
+      depends_on:
+      - image-build-amd
 
 - label: V1 Others (CPU)
   key: v1-others-cpu

.buildkite/test_areas/pytorch.yaml

@@ -107,6 +107,12 @@ steps:
     - tests/compile/passes
   commands:
   - pytest -s -v compile/passes --ignore compile/passes/distributed
+  mirror:
+    amd:
+      device: mi300_1
+      timeout_in_minutes: 180
+      depends_on:
+      - image-build-amd
 
 - label: PyTorch Fullgraph Smoke Test
   key: pytorch-fullgraph-smoke-test

.buildkite/test_areas/spec_decode.yaml

@@ -12,6 +12,20 @@ steps:
     - tests/v1/e2e/spec_decode/
   commands:
     - pytest -v -s v1/e2e/spec_decode -k "eagle_correctness"
+  mirror:
+    amd:
+      device: mi325_1
+      timeout_in_minutes: 45
+      depends_on:
+      - image-build-amd
+      source_file_dependencies:
+      - vllm/v1/spec_decode/
+      - vllm/v1/worker/gpu/spec_decode/
+      - vllm/model_executor/model_loader/
+      - vllm/v1/sample/
+      - vllm/model_executor/layers/
+      - tests/v1/e2e/spec_decode/
+      - vllm/platforms/rocm.py
 
 - label: Spec Decode Eagle Nightly B200
   key: spec-decode-eagle-nightly-b200

docs/contributing/model/basic.md

@@ -136,7 +136,7 @@ The model should also be added to the `MODELS_CONFIG_MAP` dictionary in [vllm/mo
 For case (2), we recommend using as a reference the implementation of [`JambaForCausalLM`](../../../vllm/model_executor/models/jamba.py) (for an example of a model that uses Mamba-1 and attention together) or [`NemotronHForCausalLM`](../../../vllm/model_executor/models/nemotron_h.py) (for an example of a model that uses Mamba-2 and attention together).
 These models should follow the same instructions as case (1), but they should inherit protocol `IsHybrid` (instead of `IsAttentionFree`) and it is *not* necessary to add them to the `MODELS_CONFIG_MAP` (their runtime defaults will be inferred from the protocol).
 
-For case (3), we recommend looking at the implementation of [`MiniMaxText01ForCausalLM`](../../../vllm/model_executor/models/minimax_text_01.py) or [`Lfm2ForCausalLM`](../../../vllm/model_executor/models/lfm2.py) as a reference, which use custom "mamba-like" layers `MiniMaxText01LinearAttention` and `ShortConv` respectively.
+For case (3), we recommend looking at the implementation of [`Lfm2ForCausalLM`](../../../vllm/model_executor/models/lfm2.py) as a reference, which uses a custom "mamba-like" layer `ShortConv`.
 Please follow the same guidelines as case (2) for implementing these models.
 We use "mamba-like" to refer to layers that possess a state that is updated in-place, rather than being appended-to (like KV cache for attention).
 For implementing new custom mamba-like layers, one should inherit from `MambaBase` and implement the methods `get_state_dtype`, `get_state_shape` to calculate the data types and state shapes at runtime, as well as `mamba_type` and `get_attn_backend`.
@@ -144,5 +144,5 @@ It is also necessary to implement the "attention meta-data" class which handles
 Please see [`LinearAttentionMetadata`](../../../vllm/v1/attention/backends/linear_attn.py) or [`ShortConvAttentionMetadata`](../../../vllm/v1/attention/backends/short_conv_attn.py) for examples of this.
 It is also worth noting that we should update `MambaAttentionBackendEnum` in [`registry.py`](../../../vllm/v1/attention/backends/registry.py) when adding a new mamba backend.
 Finally, if one wants to support torch compile and CUDA graphs, it necessary to wrap the call to the mamba-like layer inside a custom op and register it.
-Please see the calls to `direct_register_custom_op` in [vllm/model_executor/models/minimax_text_01.py](../../../vllm/model_executor/models/minimax_text_01.py) or [vllm/model_executor/layers/mamba/short_conv.py](../../../vllm/model_executor/layers/mamba/short_conv.py) for examples of this.
+Please see the calls to `direct_register_custom_op` in [vllm/model_executor/layers/mamba/linear/minimax_linear_attn.py](../../../vllm/model_executor/layers/mamba/linear/minimax_linear_attn.py) or [vllm/model_executor/layers/mamba/short_conv.py](../../../vllm/model_executor/layers/mamba/short_conv.py) for examples of this.
 The new custom op should then be added to the list `_attention_ops` in [vllm/config/compilation.py](../../../vllm/config/compilation.py) to ensure that piecewise CUDA graphs works as intended.

docs/design/attention_backends.md

@@ -170,8 +170,8 @@ Priority is **1 = highest** (tried first).
 | Backend | Version | Dtypes | KV Dtypes | Block Sizes | Head Sizes | Sink | Non-Causal | MM Prefix | DCP | Attention Types | Compute Cap. |
 | ------- | ------- | ------ | --------- | ----------- | ---------- | ---- | ---------- | --------- | --- | --------------- | ------------ |
 | `CPU_ATTN` | | fp16, bf16, fp32 | `auto`, `fp8`, `fp8_e4m3`, `fp8_e5m2` | %16 | 32, 64, 80, 96, 112, 128, 160, 192, 224, 256, 512 | ❌ | ❌ | ❌ | ❌ | All | N/A |
-| `FLASHINFER` | Native† | fp16, bf16 | `auto`, `float16`, `bfloat16`, `fp8`, `fp8_e4m3`, `fp8_e5m2` | 16, 32, 64, 128, 256, 512, 1024 | 64, 128, 256, 512 | ❌ | ❌ | ❌ | ✅ | Decoder | 7.x-9.x |
-| `FLASHINFER` | TRTLLM† | fp16, bf16 | `auto`, `float16`, `bfloat16`, `fp8`, `fp8_e4m3`, `fp8_e5m2`, `nvfp4` | 16, 32, 64, 128, 256, 512, 1024 | 64, 128, 256, 512 | ✅ | ❌ | ❌ | ✅ | Decoder | 10.x |
+| `FLASHINFER` | Native† | fp16, bf16 | `auto`, `float16`, `bfloat16`, `fp8`, `fp8_e4m3`, `fp8_e5m2` | 16, 32, 64, 128, 256, 512, 1024 | 64, 128, 256, 512 | ❌ | ✅ | ❌ | ✅ | Decoder | 7.x-9.x |
+| `FLASHINFER` | TRTLLM† | fp16, bf16 | `auto`, `float16`, `bfloat16`, `fp8`, `fp8_e4m3`, `fp8_e5m2`, `nvfp4` | 16, 32, 64, 128, 256, 512, 1024 | 64, 128, 256, 512 | ✅ | ✅ | ❌ | ✅ | Decoder | 10.x |
 | `FLASH_ATTN` | FA2* | fp16, bf16 | `auto`, `float16`, `bfloat16` | %16 | Any | ❌ | ✅ | ❌ | ✅ | All | ≥8.0 |
 | `FLASH_ATTN` | FA3* | fp16, bf16 | `auto`, `float16`, `bfloat16`, `fp8`, `fp8_e4m3`, `fp8_e5m2` | %16 | Any | ✅ | ✅ | ❌ | ✅ | All | 9.x |
 | `FLASH_ATTN` | FA4* | fp16, bf16 | `auto`, `float16`, `bfloat16` | %16 | Any | ✅ | ✅ | ❌ | ✅ | All | ≥10.0 |

docs/features/kv_offloading_usage.md

@@ -74,6 +74,7 @@ vllm serve <model> \
 | `max_tracker_size` | no | `64000` | single-tier | Max entries in the lookup tracker. |
 | `secondary_tiers` | no | `[]` | multi-tier | List of secondary tier configs (see below). |
 | `offload_prompt_only` | no | `true` | both | If `true`, only prompt (prefill) blocks are offloaded; decode blocks are skipped. |
+| `self_describing_kv_events` | no | `false` | single-tier | Opt-in. When `true` *and* KV cache events are enabled (`--kv-events-config` with `enable_kv_cache_events`), the connector emits self-describing block-granular `BlockStored`/`BlockRemoved` payloads (constituent block hashes, whole-chunk `token_ids`, per-block `block_size`, parent hash, LoRA + group/cache-spec metadata) instead of the placeholder fallback, so external KV-event consumers can index offloaded blocks. Inert unless events are enabled. Currently rejected by `TieringOffloadingSpec`. Full-attention groups only; sliding-window/SSM groups keep the placeholder fallback. In chunk mode (`block_size` > GPU block size), overlapping chunks re-announce shared per-block hashes, so consumers must reference-count (deduplicate) repeated store/remove announcements. |
 | `spec_module_path` | no | — | both | Python import path for a custom `OffloadingSpec` not in the built-in registry. Required only when `spec_name` is not built-in (advanced). |
 
 ## Secondary Tiers

docs/features/tool_calling.md

@@ -321,15 +321,6 @@ For Qwen2.5, the chat template in tokenizer_config.json has already included sup
 
 Flags: `--tool-call-parser hermes`
 
-### MiniMax Models (`minimax_m1`)
-
-Supported models:
-
-* `MiniMaxAi/MiniMax-M1-40k` (use with [examples/tool_chat_template_minimax_m1.jinja](../../examples/tool_chat_template_minimax_m1.jinja))
-* `MiniMaxAi/MiniMax-M1-80k` (use with [examples/tool_chat_template_minimax_m1.jinja](../../examples/tool_chat_template_minimax_m1.jinja))
-
-Flags: `--tool-call-parser minimax --chat-template examples/tool_chat_template_minimax_m1.jinja`
-
 ### DeepSeek-V3 Models (`deepseek_v3`)
 
 Supported models:

docs/models/pooling_models/token_embed.md

@@ -61,7 +61,7 @@ Models of any architecture can be converted into embedding models using `--conve
 | `ColModernVBertForRetrieval` | ColModernVBERT | T / I | `ModernVBERT/colmodernvbert-merged` | | |
 | `ColPaliForRetrieval` | ColPali | T / I | `vidore/colpali-v1.3-hf` | | |
 | `ColQwen3` | Qwen3-VL | T / I | `TomoroAI/tomoro-colqwen3-embed-4b`, `TomoroAI/tomoro-colqwen3-embed-8b` | | |
-| `ColQwen3_5` | ColQwen3.5 | T + I + V | `athrael-soju/colqwen3.5-4.5B-v3` | | |
+| `ColQwen3_5` | ColQwen3.5 | T + I + V | `athrael-soju/colqwen3.5-4.5B-v3`, `vultr/VultronRetrieverPrime-Qwen3.5-8B` | | |
 | `OpsColQwen3Model` | Qwen3-VL | T / I | `OpenSearch-AI/Ops-Colqwen3-4B`, `OpenSearch-AI/Ops-Colqwen3-8B` | | |
 | `Qwen3VLNemotronEmbedModel` | Qwen3-VL | T / I | `nvidia/nemotron-colembed-vl-4b-v2`, `nvidia/nemotron-colembed-vl-8b-v2` | ✅︎ | ✅︎ |
 | `*ForConditionalGeneration`<sup>C</sup>, `*ForCausalLM`<sup>C</sup>, etc. | Generative models | \* | N/A | \* | \* |

docs/models/supported_models.md

@@ -441,7 +441,6 @@ th {
 | `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. | | |
 | `MiniMaxM2ForCausalLM` | MiniMax-M2, MiniMax-M2.1 | `MiniMaxAI/MiniMax-M2`, etc. | ✅︎ | ✅︎ |
 | `MistralForCausalLM` | Ministral-3, Mistral, Mistral-Instruct | `mistralai/Ministral-3-3B-Instruct-2512`, `mistralai/Mistral-7B-v0.1`, `mistralai/Mistral-7B-Instruct-v0.1`, etc. | ✅︎ | ✅︎ |
 | `MistralLarge3ForCausalLM` | Mistral-Large-3-675B-Base-2512, Mistral-Large-3-675B-Instruct-2512 | `mistralai/Mistral-Large-3-675B-Base-2512`, `mistralai/Mistral-Large-3-675B-Instruct-2512`, etc. | ✅︎ | ✅︎ |
@@ -487,8 +486,6 @@ th {
 | `TeleChat2ForCausalLM` | TeleChat2 | `Tele-AI/TeleChat2-3B`, `Tele-AI/TeleChat2-7B`, `Tele-AI/TeleChat2-35B`, etc. | ✅︎ | ✅︎ |
 | `TeleChat3ForCausalLM` | TeleChat3 | `Tele-AI/TeleChat3-36B-Thinking`, `Tele-AI/TeleChat3-Coder-36B-Thinking`, etc. | ✅︎ | ✅︎ |
 | `TeleFLMForCausalLM` | TeleFLM | `CofeAI/FLM-2-52B-Instruct-2407`, `CofeAI/Tele-FLM`, etc. | ✅︎ | ✅︎ |
-| `MiniMaxM1ForCausalLM` | MiniMax-Text | `MiniMaxAI/MiniMax-M1-40k`, `MiniMaxAI/MiniMax-M1-80k`, etc. | | |
-| `MiniMaxText01ForCausalLM` | MiniMax-Text | `MiniMaxAI/MiniMax-Text-01`, etc. | | |
 | `Zamba2ForCausalLM` | Zamba2 | `Zyphra/Zamba2-7B-instruct`, `Zyphra/Zamba2-2.7B-instruct`, `Zyphra/Zamba2-1.2B-instruct`, etc. | | |
 
 !!! note
@@ -595,6 +592,7 @@ These models primarily accept the [`LLM.generate`](./generative_models.md#llmgen
 | `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. | ✅︎ | |
+| `MiniMaxM3SparseForConditionalGeneration` | MiniMax-M3 | T + I<sup>+</sup> + V<sup>+</sup> | `MiniMaxAI/MiniMax-M3`, `MiniMaxAI/MiniMax-M3-MXFP8`, etc. | | |
 | `MiniMaxVL01ForConditionalGeneration` | MiniMax-VL | T + I<sup>E+</sup> | `MiniMaxAI/MiniMax-VL-01`, etc. | | ✅︎ |
 | `Mistral3ForConditionalGeneration` | Mistral3 (HF Transformers) | T + I<sup>+</sup> | `mistralai/Mistral-Small-3.1-24B-Instruct-2503`, etc. | ✅︎ | ✅︎ |
 | `MolmoForCausalLM` | Molmo | T + I<sup>+</sup> | `allenai/Molmo-7B-D-0924`, `allenai/Molmo-7B-O-0924`, etc. | ✅︎ | ✅︎ |

docs/usage/v1_guide.md

@@ -128,7 +128,7 @@ Models that use Mamba-2 and Mamba-1 layers (e.g., `Mamba2ForCausalLM`, `MambaFor
 Hybrid models that combine Mamba-2 and Mamba-1 layers with standard attention layers are also supported (e.g., `BambaForCausalLM`,
 `Zamba2ForCausalLM`, `NemotronHForCausalLM`, `FalconH1ForCausalLM` and `GraniteMoeHybridForCausalLM`, `JambaForCausalLM`, `Plamo2ForCausalLM`).
 
-Hybrid models with mechanisms different to Mamba are also supported (e.g, `MiniMaxText01ForCausalLM`, `MiniMaxM1ForCausalLM`, `Lfm2ForCausalLM`).
+Hybrid models with mechanisms different to Mamba are also supported (e.g, `Lfm2ForCausalLM`).
 
 Please note that prefix caching is not yet supported for any of the above models.
 

examples/generate/multimodal/vision_language_offline.py

@@ -1481,39 +1481,6 @@ def run_minicpmv(questions: list[str], modality: str) -> ModelRequestData:
     return run_minicpmv_base(questions, modality, "openbmb/MiniCPM-V-2_6")
 
 
-def run_minimax_vl_01(questions: list[str], modality: str) -> ModelRequestData:
-    assert modality == "image"
-
-    model_name = "MiniMaxAI/MiniMax-VL-01"
-
-    engine_args = EngineArgs(
-        model=model_name,
-        max_num_seqs=2,
-        limit_mm_per_prompt={modality: 1},
-        trust_remote_code=True,
-        tensor_parallel_size=8,
-    )
-
-    tokenizer = AutoTokenizer.from_pretrained(model_name)
-    messages = [
-        [
-            {
-                "role": "user",
-                "content": [{"type": "image"}, {"type": "text", "text": question}],
-            }
-        ]
-        for question in questions
-    ]
-    prompts = tokenizer.apply_chat_template(
-        messages, add_generation_prompt=True, tokenize=False
-    )
-
-    return ModelRequestData(
-        engine_args=engine_args,
-        prompts=prompts,
-    )
-
-
 # Mistral-3 HF-format
 def run_mistral3(questions: list[str], modality: str) -> ModelRequestData:
     assert modality == "image"
@@ -2485,7 +2452,6 @@ model_example_map = {
     "mantis": run_mantis,
     "minicpmo": run_minicpmo,
     "minicpmv": run_minicpmv,
-    "minimax_vl_01": run_minimax_vl_01,
     "mistral3": run_mistral3,
     "molmo": run_molmo,
     "molmo2": run_molmo2,

examples/pooling/score/colqwen3_5_rerank_online.py

@@ -7,11 +7,27 @@ ColQwen3.5 is a multi-modal ColBERT-style model based on Qwen3.5.
 It produces per-token embeddings and uses MaxSim scoring for retrieval
 and reranking. Supports both text and image inputs.
 
+Works for any ColQwen3.5 checkpoint, e.g. `athrael-soju/colqwen3.5-4.5B-v3`
+or `vultr/VultronRetrieverPrime-Qwen3.5-8B`.
+
 Start the server with:
-    vllm serve athrael-soju/colqwen3.5-4.5B --max-model-len 4096
+    vllm serve athrael-soju/colqwen3.5-4.5B-v3 --max-model-len 4096 \
+        --mm-processor-kwargs '{"min_pixels": 65536, "max_pixels": 1835008}'
 
 Then run this script:
     python colqwen3_5_rerank_online.py
+
+Parity note (matching the native colpali ColQwen3_5Processor pipeline):
+  - Visual-token budget: ColQwen3_5Processor uses max_num_visual_tokens=1792,
+    i.e. max_pixels = 1792 * (patch_size*merge_size)^2 = 1792 * 32^2 = 1835008
+    (with min_pixels = shortest_edge = 65536). Pass these via --mm-processor-kwargs
+    as above; the default budget gives fewer visual tokens and lower retrieval ndcg.
+  - When you build prompts yourself (token_embed), reproduce the processor exactly:
+      image (document): wrap in the instruction template
+        "<|im_start|>user\n<|vision_start|><|image_pad|><|vision_end|>"
+        "Describe the image.<|im_end|><|endoftext|>"
+      query: append the augmentation suffix  <text> + "<|endoftext|>" * 10
+    Omitting these reproduces a silent ~2.5 ndcg@10 drop vs the native pipeline.
 """
 
 import requests

examples/tool_chat_template_minimax_m1.jinja

@@ -1,91 +0,0 @@
-{{ '<begin_of_document>' -}}
-{%- if custom_tools is defined %}
-    {%- set tools = custom_tools %}
-{%- endif %}
-{%- if not tools is defined %}
-    {%- set tools = none %}
-{%- endif %}
-
-{#- Extract system message #}
-{% set ns = namespace(system_prompt='') -%}
-{%- if messages[0]['role'] == 'system' %}
-    {%- if messages[0]['content'] is string %}
-        {%- set ns.system_prompt = messages[0]['content']|trim %}
-    {%- else %}
-        {%- set ns.system_prompt = messages[0]['content'][0]['text']|trim %}
-    {%- endif %}
-    {%- set messages = messages[1:] %}
-{%- else %}
-    {%- if tools is not none %}
-        {%- set ns.system_prompt = "You are a helpful assistant created by Minimax based on MiniMax-M1 model." %}
-    {%- else %}
-        {%- set ns.system_prompt = "You are a helpful assistant created by Minimax based on MiniMax-M1 model." %}
-    {%- endif %}
-{%- endif %}
-
-{#- System message #}
-{%- if ns.system_prompt != '' %}
-{{ '<beginning_of_sentence>system ai_setting=assistant\n' + ns.system_prompt + '<end_of_sentence>\n' -}}
-{%- endif %}
-
-{#- Tools configuration #}
-{%- if tools is not none %}
-{{ '<beginning_of_sentence>system tool_setting=tools\nYou are provided with these tools:\n<tools>\n' -}}
-{%- for tool in tools %}
-{{ tool | tojson ~ '\n' -}}
-{%- endfor %}
-{{ '</tools>\n\nIf you need to call tools, please respond with <tool_calls></tool_calls> XML tags, and provide tool-name and json-object of arguments, following the format below:\n<tool_calls>\n{"name": <tool-name>, "arguments": <args-json-object>}\n...\n</tool_calls><end_of_sentence>\n' -}}
-{%- endif %}
-
-{#- Process messages #}
-{%- for message in messages %}
-    {%- if not (message.role == 'ipython' or message.role == 'tool' or 'tool_calls' in message) %}
-        {%- if message['role'] == 'user' %}
-{{ '<beginning_of_sentence>user name=user\n' -}}
-{%- if message['content'] is string %}
-{{ message['content']|trim -}}
-{%- else %}
-{%- for content in message['content'] %}
-{%- if content['type'] == 'text' %}
-{{ content['text']|trim -}}
-{%- endif %}
-{%- endfor %}
-{%- endif %}
-{{ '<end_of_sentence>\n' -}}
-        {%- elif message['role'] == 'assistant' %}
-{{ '<beginning_of_sentence>ai name=assistant\n' -}}
-{%- if message['content'] is string %}
-{{ message['content']|trim -}}
-{%- else %}
-{%- for content in message['content'] | selectattr('type', 'equalto', 'text') %}
-{{ content['text']|trim -}}
-{%- endfor %}
-{%- endif %}
-{{ '<end_of_sentence>\n' -}}
-        {%- endif %}
-    {%- elif 'tool_calls' in message %}
-{{ '<beginning_of_sentence>ai name=assistant\n<tool_calls>\n' -}}
-{%- for tool_call in message.tool_calls %}
-{{ '{"name": "' + tool_call.function.name + '", "arguments": ' + tool_call.function.arguments | tojson + '}\n' -}}
-{%- endfor %}
-{{ '</tool_calls><end_of_sentence>\n' -}}
-    {%- elif message.role == "tool" or message.role == "ipython" %}
-{{ '<beginning_of_sentence>tool name=tools\n' -}}
-{%- if message.content is string %}
-{{ 'tool result: ' + message.content + '\n\n' -}}
-{%- else %}
-{%- for content in message['content'] %}
-{%- if content['type'] == 'text' %}
-{{ 'tool result: ' + content['text'] + '\n\n' -}}
-{%- elif content.get('name') %}
-{{ 'tool name: ' + content['name'] + '\ntool result: ' + content['text'] + '\n\n' -}}
-{%- endif %}
-{%- endfor %}
-{%- endif %}
-{{ '<end_of_sentence>\n' -}}
-    {%- endif %}
-{%- endfor %}
-
-{%- if add_generation_prompt %}
-{{ '<beginning_of_sentence>ai name=assistant\n' -}}
-{%- endif %}

rust/src/chat/src/renderer/hf/format.rs

@@ -386,7 +386,6 @@ mod tests {
             tool_chat_template_llama3.2_pythonic.jinja         => String
             tool_chat_template_llama4_json.jinja               => OpenAi
             tool_chat_template_llama4_pythonic.jinja           => OpenAi
-            tool_chat_template_minimax_m1.jinja                => OpenAi
             tool_chat_template_mistral.jinja                   => String
             tool_chat_template_mistral3.jinja                  => OpenAi
             tool_chat_template_mistral_parallel.jinja          => String

rust/src/chat/tests/chat.rs

@@ -16,7 +16,8 @@ use vllm_engine_core_client::protocol::logprobs::{
     Logprobs, MaybeWireLogprobs, PositionLogprobs, TokenLogprob,
 };
 use vllm_engine_core_client::protocol::{
-    EngineCoreFinishReason, EngineCoreOutput, EngineCoreOutputs, EngineCoreRequest, StopReason,
+    EngineCoreFinishReason, EngineCoreOutput, EngineCoreOutputs, EngineCoreRequest, LogprobsCount,
+    StopReason,
 };
 use vllm_engine_core_client::test_utils::{IpcNamespace, spawn_mock_engine_task};
 use vllm_engine_core_client::{EngineCoreClient, EngineCoreClientConfig};
@@ -1387,8 +1388,8 @@ async fn chat_stream_and_collect_preserve_prompt_and_sample_logprobs() {
     .await;
 
     let mut request = sample_request("chat-logprobs");
-    request.sampling_params.logprobs = Some(1);
-    request.sampling_params.prompt_logprobs = Some(1);
+    request.sampling_params.logprobs = Some(LogprobsCount::Top(1));
+    request.sampling_params.prompt_logprobs = Some(LogprobsCount::Top(1));
 
     let mut stream = chat.chat(request.clone()).await.unwrap();
     match next_semantic(&mut stream).await.unwrap().unwrap() {

rust/src/chat/tests/templates/vllm_examples/tool_chat_template_minimax_m1.jinja

@@ -1,91 +0,0 @@
-{{ '<begin_of_document>' -}}
-{%- if custom_tools is defined %}
-    {%- set tools = custom_tools %}
-{%- endif %}
-{%- if not tools is defined %}
-    {%- set tools = none %}
-{%- endif %}
-
-{#- Extract system message #}
-{% set ns = namespace(system_prompt='') -%}
-{%- if messages[0]['role'] == 'system' %}
-    {%- if messages[0]['content'] is string %}
-        {%- set ns.system_prompt = messages[0]['content']|trim %}
-    {%- else %}
-        {%- set ns.system_prompt = messages[0]['content'][0]['text']|trim %}
-    {%- endif %}
-    {%- set messages = messages[1:] %}
-{%- else %}
-    {%- if tools is not none %}
-        {%- set ns.system_prompt = "You are a helpful assistant created by Minimax based on MiniMax-M1 model." %}
-    {%- else %}
-        {%- set ns.system_prompt = "You are a helpful assistant created by Minimax based on MiniMax-M1 model." %}
-    {%- endif %}
-{%- endif %}
-
-{#- System message #}
-{%- if ns.system_prompt != '' %}
-{{ '<beginning_of_sentence>system ai_setting=assistant\n' + ns.system_prompt + '<end_of_sentence>\n' -}}
-{%- endif %}
-
-{#- Tools configuration #}
-{%- if tools is not none %}
-{{ '<beginning_of_sentence>system tool_setting=tools\nYou are provided with these tools:\n<tools>\n' -}}
-{%- for tool in tools %}
-{{ tool | tojson ~ '\n' -}}
-{%- endfor %}
-{{ '</tools>\n\nIf you need to call tools, please respond with <tool_calls></tool_calls> XML tags, and provide tool-name and json-object of arguments, following the format below:\n<tool_calls>\n{"name": <tool-name>, "arguments": <args-json-object>}\n...\n</tool_calls><end_of_sentence>\n' -}}
-{%- endif %}
-
-{#- Process messages #}
-{%- for message in messages %}
-    {%- if not (message.role == 'ipython' or message.role == 'tool' or 'tool_calls' in message) %}
-        {%- if message['role'] == 'user' %}
-{{ '<beginning_of_sentence>user name=user\n' -}}
-{%- if message['content'] is string %}
-{{ message['content']|trim -}}
-{%- else %}
-{%- for content in message['content'] %}
-{%- if content['type'] == 'text' %}
-{{ content['text']|trim -}}
-{%- endif %}
-{%- endfor %}
-{%- endif %}
-{{ '<end_of_sentence>\n' -}}
-        {%- elif message['role'] == 'assistant' %}
-{{ '<beginning_of_sentence>ai name=assistant\n' -}}
-{%- if message['content'] is string %}
-{{ message['content']|trim -}}
-{%- else %}
-{%- for content in message['content'] | selectattr('type', 'equalto', 'text') %}
-{{ content['text']|trim -}}
-{%- endfor %}
-{%- endif %}
-{{ '<end_of_sentence>\n' -}}
-        {%- endif %}
-    {%- elif 'tool_calls' in message %}
-{{ '<beginning_of_sentence>ai name=assistant\n<tool_calls>\n' -}}
-{%- for tool_call in message.tool_calls %}
-{{ '{"name": "' + tool_call.function.name + '", "arguments": ' + tool_call.function.arguments | tojson + '}\n' -}}
-{%- endfor %}
-{{ '</tool_calls><end_of_sentence>\n' -}}
-    {%- elif message.role == "tool" or message.role == "ipython" %}
-{{ '<beginning_of_sentence>tool name=tools\n' -}}
-{%- if message.content is string %}
-{{ 'tool result: ' + message.content + '\n\n' -}}
-{%- else %}
-{%- for content in message['content'] %}
-{%- if content['type'] == 'text' %}
-{{ 'tool result: ' + content['text'] + '\n\n' -}}
-{%- elif content.get('name') %}
-{{ 'tool name: ' + content['name'] + '\ntool result: ' + content['text'] + '\n\n' -}}
-{%- endif %}
-{%- endfor %}
-{%- endif %}
-{{ '<end_of_sentence>\n' -}}
-    {%- endif %}
-{%- endfor %}
-
-{%- if add_generation_prompt %}
-{{ '<beginning_of_sentence>ai name=assistant\n' -}}
-{%- endif %}

rust/src/cmd/src/cli.rs

@@ -20,6 +20,7 @@ use serde_with::{DefaultOnNull, OneOrMany, serde_as};
 use thiserror_ext::AsReport as _;
 use uuid::Uuid;
 use vllm_engine_core_client::TransportMode;
+use vllm_engine_core_client::protocol::LogprobsCount;
 use vllm_managed_engine::ManagedEngineConfig;
 use vllm_managed_engine::cli::{ManagedEngineArgs, repartition_managed_engine_args};
 use vllm_server::{
@@ -136,9 +137,9 @@ pub struct SharedRuntimeArgs {
     pub max_model_len: Option<u32>,
     /// Maximum number of log probabilities to return when `logprobs` is
     /// specified in sampling parameters. `-1` means no cap.
-    #[arg(long, value_parser = clap::value_parser!(i32).range(-1..), allow_negative_numbers = true)]
+    #[arg(long, allow_negative_numbers = true)]
     #[serde(default)]
-    pub max_logprobs: Option<i32>,
+    pub max_logprobs: Option<LogprobsCount>,
     /// TCP port for the gRPC Generate service. When not set, no gRPC server is
     /// started.
     #[arg(long)]
@@ -529,7 +530,7 @@ impl ServeArgs {
         self.managed_engine.clone().into_config(
             self.runtime.model.clone(),
             self.runtime.max_model_len,
-            self.runtime.max_logprobs,
+            self.runtime.max_logprobs.map(managed_max_logprobs_to_i32),
             self.runtime.language_model_only,
             self.runtime.disable_log_stats,
             self.runtime.shutdown_timeout,
@@ -555,5 +556,9 @@ fn frontend_ipc_addresses() -> (String, String) {
     )
 }
 
+fn managed_max_logprobs_to_i32(count: LogprobsCount) -> i32 {
+    i32::try_from(count).expect("max_logprobs is parsed through i32")
+}
+
 #[cfg(test)]
 mod tests;

rust/src/cmd/src/cli/tests.rs

@@ -1,5 +1,6 @@
 use expect_test::expect;
 use vllm_engine_core_client::TransportMode;
+use vllm_engine_core_client::protocol::LogprobsCount;
 use vllm_server::{Config, HttpListenerMode, ParserSelection, RendererSelection};
 
 use super::{Cli, Command};
@@ -165,10 +166,10 @@ fn serve_args_forward_max_logprobs_to_frontend_and_managed_engine() {
     let Command::Serve(args) = cli.command else {
         panic!("expected serve args");
     };
-    assert_eq!(args.runtime.max_logprobs, Some(-1));
+    assert_eq!(args.runtime.max_logprobs, Some(LogprobsCount::All));
 
     let frontend_config = args.to_frontend_config("tcp://127.0.0.1:62100".to_string());
-    assert_eq!(frontend_config.max_logprobs, Some(-1));
+    assert_eq!(frontend_config.max_logprobs, Some(LogprobsCount::All));
 
     let engine_config = args.to_managed_engine_config(5555);
     assert_eq!(engine_config.python_args, vec!["--max-logprobs", "-1"]);
@@ -529,7 +530,7 @@ fn frontend_args_json_accepts_supported_non_default_fields() {
     assert_eq!(args.runtime.renderer, RendererSelection::DeepSeekV32);
     assert!(args.runtime.language_model_only);
     assert_eq!(args.runtime.max_model_len, Some(8192));
-    assert_eq!(args.runtime.max_logprobs, Some(-1));
+    assert_eq!(args.runtime.max_logprobs, Some(LogprobsCount::All));
     assert_eq!(args.runtime.shutdown_timeout, 3);
 }
 

rust/src/engine-core-client/examples/external_engine_logprobs.rs

@@ -6,7 +6,7 @@ use futures::StreamExt as _;
 use tokio::time::timeout;
 use tracing_subscriber::EnvFilter;
 use vllm_engine_core_client::protocol::{
-    EngineCoreFinishReason, EngineCoreRequest, EngineCoreSamplingParams,
+    EngineCoreFinishReason, EngineCoreRequest, EngineCoreSamplingParams, LogprobsCount,
 };
 use vllm_engine_core_client::{
     EngineCoreClient, EngineCoreClientConfig, EngineCoreStreamOutput, TransportMode,
@@ -33,10 +33,10 @@ struct Args {
     output_timeout_secs: u64,
     #[arg(long, default_value_t = 1)]
     max_tokens: u32,
-    #[arg(long, default_value_t = 2)]
-    logprobs: i32,
-    #[arg(long, default_value_t = 1)]
-    prompt_logprobs: i32,
+    #[arg(long, default_value_t = LogprobsCount::Top(2), allow_negative_numbers = true)]
+    logprobs: LogprobsCount,
+    #[arg(long, default_value_t = LogprobsCount::Top(1), allow_negative_numbers = true)]
+    prompt_logprobs: LogprobsCount,
     #[arg(long, default_value_t = 96)]
     prompt_repeats: usize,
 }
@@ -64,8 +64,8 @@ fn build_request(
     request_id: String,
     prompt_token_ids: Vec<u32>,
     max_tokens: u32,
-    logprobs: i32,
-    prompt_logprobs: i32,
+    logprobs: LogprobsCount,
+    prompt_logprobs: LogprobsCount,
     client_index: u32,
 ) -> EngineCoreRequest {
     EngineCoreRequest {

rust/src/engine-core-client/src/protocol/logprobs_count.rs

@@ -0,0 +1,135 @@
+use std::fmt;
+use std::str::FromStr;
+
+use serde::{Deserialize, Deserializer, Serialize, Serializer};
+
+/// Number of log probabilities requested for a token position.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum LogprobsCount {
+    /// Return the full model vocabulary.
+    All,
+    /// Return the top-N tokens by probability.
+    Top(u32),
+}
+
+impl LogprobsCount {
+    /// Expands the count to the actual number of logprobs to return, given the vocabulary size.
+    pub fn expanded(self, vocab_size: usize) -> usize {
+        match self {
+            Self::All => vocab_size,
+            Self::Top(count) => count as usize,
+        }
+    }
+}
+
+impl TryFrom<i32> for LogprobsCount {
+    type Error = String;
+
+    fn try_from(value: i32) -> Result<Self, Self::Error> {
+        match value {
+            -1 => Ok(Self::All),
+            value if value < -1 => Err(format!("must be non-negative or -1, got {value}")),
+            value => Ok(Self::Top(value as u32)),
+        }
+    }
+}
+
+impl TryFrom<LogprobsCount> for i32 {
+    type Error = String;
+
+    fn try_from(value: LogprobsCount) -> Result<Self, Self::Error> {
+        match value {
+            LogprobsCount::All => Ok(-1),
+            LogprobsCount::Top(count) => {
+                i32::try_from(count).map_err(|_| format!("must fit within i32, got {count}"))
+            }
+        }
+    }
+}
+
+impl FromStr for LogprobsCount {
+    type Err = String;
+
+    fn from_str(s: &str) -> Result<Self, Self::Err> {
+        let value = s
+            .parse::<i32>()
+            .map_err(|e| format!("must be an i32 integer, got {s:?}: {e}"))?;
+        Self::try_from(value)
+    }
+}
+
+impl fmt::Display for LogprobsCount {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        match self {
+            Self::All => (-1).fmt(f),
+            Self::Top(count) => count.fmt(f),
+        }
+    }
+}
+
+impl Serialize for LogprobsCount {
+    fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
+    where
+        S: Serializer,
+    {
+        let value: i32 = (*self).try_into().map_err(serde::ser::Error::custom)?;
+        value.serialize(serializer)
+    }
+}
+
+impl<'de> Deserialize<'de> for LogprobsCount {
+    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
+    where
+        D: Deserializer<'de>,
+    {
+        let value = i32::deserialize(deserializer)?;
+        Self::try_from(value).map_err(serde::de::Error::custom)
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use rmpv::Value;
+
+    use super::*;
+    use crate::protocol::{decode_msgpack, encode_msgpack};
+
+    #[test]
+    fn logprobs_count_serializes_as_wire_integer() {
+        assert_eq!(serde_json::to_value(LogprobsCount::All).unwrap(), -1);
+        assert_eq!(serde_json::to_value(LogprobsCount::Top(3)).unwrap(), 3);
+    }
+
+    #[test]
+    fn logprobs_count_deserializes_wire_integer() {
+        assert_eq!(
+            serde_json::from_value::<LogprobsCount>(serde_json::json!(-1)).unwrap(),
+            LogprobsCount::All
+        );
+        assert_eq!(
+            serde_json::from_value::<LogprobsCount>(serde_json::json!(3)).unwrap(),
+            LogprobsCount::Top(3)
+        );
+        assert!(serde_json::from_value::<LogprobsCount>(serde_json::json!(-2)).is_err());
+        assert!(
+            serde_json::from_value::<LogprobsCount>(serde_json::json!(i64::from(i32::MAX) + 1))
+                .is_err()
+        );
+    }
+
+    #[test]
+    fn logprobs_count_decodes_msgpack_signed_and_unsigned() {
+        let mut encoded = Vec::new();
+        rmpv::encode::write_value(&mut encoded, &Value::from(-1)).unwrap();
+        assert_eq!(
+            decode_msgpack::<LogprobsCount>(&encoded).unwrap(),
+            LogprobsCount::All
+        );
+
+        let encoded = encode_msgpack(&LogprobsCount::Top(7)).unwrap();
+        assert_eq!(
+            decode_msgpack::<LogprobsCount>(&encoded).unwrap(),
+            LogprobsCount::Top(7)
+        );
+    }
+}

rust/src/engine-core-client/src/protocol/mod.rs

@@ -56,6 +56,7 @@ mod classified_outputs;
 pub mod dtype;
 pub mod handshake;
 pub mod logprobs;
+mod logprobs_count;
 pub mod lora;
 pub mod multimodal;
 pub mod stats;
@@ -66,6 +67,7 @@ pub use classified_outputs::{
 };
 pub use dtype::ModelDtype;
 pub use logprobs::decode_engine_core_outputs;
+pub use logprobs_count::LogprobsCount;
 
 /// Request types are encoded as single-byte protocol constants so they can be
 /// sent over the ZMQ socket without an extra encoding step.
@@ -277,14 +279,20 @@ pub struct EngineCoreSamplingParams {
     pub max_tokens: u32,
     /// Minimum number of tokens to generate before EOS or stop-token handling.
     pub min_tokens: u32,
+    /// Maximum number of reasoning ("thinking") tokens to emit before the
+    /// reasoning section is force-closed. `None` means unlimited; the
+    /// user-facing `-1` sentinel is normalized to `None` by the frontend before
+    /// reaching this DTO, so only non-negative values are sent. Enforced
+    /// engine-side (and only when a reasoning parser is configured).
+    pub thinking_token_budget: Option<u64>,
     /// Number of log probabilities to return per generated token.
     ///
-    /// `None` disables sample logprobs. `-1` requests the full vocabulary.
-    pub logprobs: Option<i32>,
+    /// `None` disables sample logprobs.
+    pub logprobs: Option<LogprobsCount>,
     /// Number of log probabilities to return per prompt token.
     ///
-    /// `None` disables prompt logprobs. `-1` requests the full vocabulary.
-    pub prompt_logprobs: Option<i32>,
+    /// `None` disables prompt logprobs.
+    pub prompt_logprobs: Option<LogprobsCount>,
     /// Minimum probability threshold for token sampling.
     pub min_p: f32,
     /// Frequency penalty applied by the sampler.
@@ -345,6 +353,7 @@ impl EngineCoreSamplingParams {
             seed: None,
             max_tokens: 65536,
             min_tokens: 0,
+            thinking_token_budget: None,
             logprobs: None,
             prompt_logprobs: None,
             min_p: 0.0,

rust/src/engine-core-client/src/tests/client.rs

@@ -150,6 +150,7 @@ fn sample_request_with_id(request_id: &str) -> EngineCoreRequest {
             top_k: 8,
             max_tokens: 32,
             min_tokens: 1,
+            thinking_token_budget: Some(256),
             stop_token_ids: vec![151643],
             eos_token_id: Some(151645),
             all_stop_token_ids: BTreeSet::from([151643, 151645]),
@@ -2502,6 +2503,7 @@ fn python_msgpack_fixtures_match_rust_encoding() {
             seed: None,
             max_tokens: 16,
             min_tokens: 0,
+            thinking_token_budget: None,
             logprobs: None,
             prompt_logprobs: None,
             min_p: 0.0,

rust/src/engine-core-client/src/tests/python_compat.py

@@ -39,6 +39,7 @@ class EngineCoreSamplingParams(msgspec.Struct, dict=True, omit_defaults=True):
     seed: int | None = None
     max_tokens: int = 16
     min_tokens: int = 0
+    thinking_token_budget: int | None = None
     min_p: float = 0.0
     frequency_penalty: float = 0.0
     presence_penalty: float = 0.0
@@ -122,6 +123,7 @@ request = EngineCoreRequest(
         seed=None,
         max_tokens=32,
         min_tokens=1,
+        thinking_token_budget=256,
         min_p=0.0,
         frequency_penalty=0.0,
         presence_penalty=0.0,

rust/src/server/src/config.rs

@@ -2,12 +2,13 @@ use std::collections::HashMap;
 use std::fmt;
 use std::time::Duration;
 
-use anyhow::{Result, bail};
+use anyhow::Result;
 use axum::http::{HeaderName, HeaderValue, Method};
 use educe::Educe;
 use serde::Serialize;
 use serde_json::Value;
 use vllm_chat::{ChatTemplateContentFormatOption, ParserSelection, RendererSelection};
+use vllm_engine_core_client::protocol::LogprobsCount;
 use vllm_engine_core_client::{CoordinatorMode as EngineCoreCoordinatorMode, TransportMode};
 
 /// How the HTTP server obtains its listening socket.
@@ -133,7 +134,7 @@ pub struct Config {
     pub chat_template_content_format: ChatTemplateContentFormatOption,
     /// Optional maximum number of top log probabilities accepted by the
     /// frontend. `None` delegates to the text layer default.
-    pub max_logprobs: Option<i32>,
+    pub max_logprobs: Option<LogprobsCount>,
     /// HTTP/API-server behavior switches.
     pub api_server_options: ApiServerOptions,
     /// CORS settings applied to every HTTP response.
@@ -158,15 +159,6 @@ impl Config {
     pub fn validate(&self) -> Result<()> {
         vllm_chat::validate_parser_overrides(&self.tool_call_parser, &self.reasoning_parser)?;
         self.cors.validate()?;
-        if let Some(max_logprobs) = self.max_logprobs
-            && max_logprobs < -1
-        {
-            bail!(
-                "max_logprobs must be non-negative or -1, got {}",
-                max_logprobs
-            );
-        }
-
         Ok(())
     }
 

rust/src/server/src/error.rs

@@ -103,6 +103,7 @@ fn is_request_validation_error(error: &vllm_text::Error) -> bool {
             | vllm_text::Error::EmptyPromptTokenIds { .. }
             | vllm_text::Error::Logprobs(_)
             | vllm_text::Error::OutOfVocab(_)
+            | vllm_text::Error::InvalidThinkingTokenBudget
             // An empty tokenized prompt detected later, at request prepare
             // time, surfaces through the transparent Llm wrapper.
             | vllm_text::Error::Llm(vllm_llm::Error::EmptyPromptTokenIds { .. })
@@ -127,6 +128,18 @@ mod tests {
         assert!(response.error.message.contains("9000"));
     }
 
+    #[test]
+    fn invalid_thinking_token_budget_maps_to_invalid_request() {
+        let api_error = text_submit_error(
+            "failed to submit completion request",
+            vllm_text::Error::InvalidThinkingTokenBudget,
+        );
+        assert_eq!(api_error.status_code(), StatusCode::BAD_REQUEST);
+        let response = api_error.to_error_response();
+        assert_eq!(response.error.error_type, "invalid_request_error");
+        assert!(response.error.message.contains("thinking_token_budget"));
+    }
+
     #[test]
     fn chat_wrapped_prompt_too_long_maps_to_invalid_request() {
         let error = vllm_chat::Error::Text(vllm_text::Error::PromptTooLong {

rust/src/server/src/grpc/convert.rs

@@ -3,7 +3,7 @@
 
 use tonic::Status;
 use uuid::Uuid;
-use vllm_engine_core_client::protocol::{StopReason, StructuredOutputsParams};
+use vllm_engine_core_client::protocol::{LogprobsCount, StopReason, StructuredOutputsParams};
 use vllm_text::{
     DecodedLogprobs, DecodedPromptLogprobs, FinishReason, Finished, Prompt, SamplingParams,
     TextDecodeOptions, TextRequest,
@@ -202,18 +202,22 @@ fn build_sampling_params(
 /// Map the proto `CandidateTokens` selector to a `(logprobs_count,
 /// logprob_token_ids)` pair.
 ///
-/// - `top_n(k)` → `(k, None)` — return top-k candidates by probability
-/// - `all` → `(-1, None)` — return the full vocabulary
+/// - `top_n(k)` → `(Top(k), None)` — return top-k candidates by probability
+/// - `all` → `(All, None)` — return the full vocabulary
 /// - `token_ids(n)` → `(1, Some(vec of n token ids))` — return logprobs for specific tokens (the
 ///   count `n` is stored in the proto as the number of token IDs that follow, but the actual IDs
 ///   are carried via `logprob_token_ids` on `SamplingParams`)
-/// - absent → `(1, None)` — just the sampled/scored token
-fn candidate_logprob_spec(candidates: Option<&pb::CandidateTokens>) -> (i32, Option<Vec<u32>>) {
+/// - absent → `(Top(1), None)` — just the sampled/scored token
+fn candidate_logprob_spec(
+    candidates: Option<&pb::CandidateTokens>,
+) -> (LogprobsCount, Option<Vec<u32>>) {
     match candidates.and_then(|c| c.select.as_ref()) {
-        Some(pb::candidate_tokens::Select::TopN(n)) => (*n as i32, None),
-        Some(pb::candidate_tokens::Select::All(true)) => (-1, None),
-        Some(pb::candidate_tokens::Select::TokenIds(ids)) => (1, Some(ids.ids.clone())),
-        _ => (1, None),
+        Some(pb::candidate_tokens::Select::TopN(n)) => (LogprobsCount::Top(*n), None),
+        Some(pb::candidate_tokens::Select::All(true)) => (LogprobsCount::All, None),
+        Some(pb::candidate_tokens::Select::TokenIds(ids)) => {
+            (LogprobsCount::Top(1), Some(ids.ids.clone()))
+        }
+        _ => (LogprobsCount::Top(1), None),
     }
 }
 

rust/src/server/src/routes/inference/generate/convert.rs

@@ -87,6 +87,7 @@ pub(super) fn prepare_generate_request(
 #[cfg(test)]
 mod tests {
     use serde_json::json;
+    use vllm_engine_core_client::protocol::LogprobsCount;
     use vllm_text::Prompt;
 
     use super::prepare_generate_request;
@@ -132,10 +133,13 @@ mod tests {
             Prompt::TokenIds(vec![11, 22, 33])
         );
         assert_eq!(prepared.text_request.sampling_params.max_tokens, Some(7));
-        assert_eq!(prepared.text_request.sampling_params.logprobs, Some(2));
+        assert_eq!(
+            prepared.text_request.sampling_params.logprobs,
+            Some(LogprobsCount::Top(2))
+        );
         assert_eq!(
             prepared.text_request.sampling_params.prompt_logprobs,
-            Some(1)
+            Some(LogprobsCount::Top(1))
         );
         assert!(prepared.text_request.sampling_params.ignore_eos);
         assert_eq!(prepared.text_request.priority, -3);
@@ -150,6 +154,33 @@ mod tests {
         );
     }
 
+    #[test]
+    fn prepare_generate_request_forwards_thinking_token_budget() {
+        let request: GenerateRequest = serde_json::from_value(json!({
+            "model": "Qwen/Qwen1.5-0.5B-Chat",
+            "token_ids": [11, 22, 33],
+            "sampling_params": {
+                "thinking_token_budget": 64
+            }
+        }))
+        .expect("parse request");
+
+        let prepared = prepare_generate_request(
+            request,
+            &served(&["Qwen/Qwen1.5-0.5B-Chat"]),
+            ResolvedRequestContext::default(),
+        )
+        .expect("prepare");
+
+        // The raw inference route shares `vllm_text::SamplingParams`, so the
+        // field is carried through to lowering exactly like the OpenAI routes
+        // (normalization/validation then happens in `lower_sampling_params`).
+        assert_eq!(
+            prepared.text_request.sampling_params.thinking_token_budget,
+            Some(64)
+        );
+    }
+
     #[test]
     fn prepare_generate_request_gates_continuous_usage_on_include_usage() {
         let request: GenerateRequest = serde_json::from_value(json!({

rust/src/server/src/routes/inference/generate/validate.rs

@@ -34,16 +34,6 @@ pub(super) fn validate_request_compat(
         );
     }
 
-    if let Some(prompt_logprobs) = request.sampling_params.prompt_logprobs
-        && prompt_logprobs < 0
-        && prompt_logprobs != -1
-    {
-        bail_invalid_request!(
-            param = "sampling_params",
-            "`prompt_logprobs` must be a non-negative value or -1."
-        );
-    }
-
     Ok(())
 }
 

rust/src/server/src/routes/openai/chat_completions/convert.rs

@@ -4,6 +4,7 @@ use vllm_chat::{
     ChatMessage as VllmChatMessage, ChatOptions, ChatRequest, ChatTool, ChatToolChoice,
     GenerationPromptMode, SamplingParams,
 };
+use vllm_engine_core_client::protocol::LogprobsCount;
 
 use super::types::ChatCompletionRequest;
 use super::validate;
@@ -94,7 +95,7 @@ pub(super) fn prepare_chat_request(
 
     // Auto-enable prompt logprobs for non-streaming echo, matching Python vLLM's
     // behavior.
-    let top_logprobs = request.top_logprobs.unwrap_or(0);
+    let top_logprobs = request.top_logprobs.unwrap_or(LogprobsCount::Top(0));
     let prompt_logprobs = request
         .prompt_logprobs
         .or((request.echo && !request.stream).then_some(top_logprobs));
@@ -115,6 +116,7 @@ pub(super) fn prepare_chat_request(
             seed: request.seed,
             max_tokens: request.max_completion_tokens,
             min_tokens: request.min_tokens,
+            thinking_token_budget: request.thinking_token_budget,
             logprobs: request.logprobs.then_some(top_logprobs),
             prompt_logprobs,
             min_p: request.min_p,
@@ -377,6 +379,7 @@ mod tests {
         ChatTool as VllmChatTool, ChatToolChoice, GenerationPromptMode,
         SamplingParams as VllmSamplingParams,
     };
+    use vllm_engine_core_client::protocol::LogprobsCount;
     use vllm_text::output::TextDecodeOptions;
 
     use super::prepare_chat_request;
@@ -613,6 +616,31 @@ mod tests {
         assert_eq!(prepared.chat_request.sampling_params, expected);
     }
 
+    #[test]
+    fn prepare_chat_request_passes_through_thinking_token_budget() {
+        let prepare = |budget: Option<i64>| {
+            prepare_chat_request(
+                ChatCompletionRequest {
+                    thinking_token_budget: budget,
+                    ..base_request()
+                },
+                &served(&["Qwen/Qwen1.5-0.5B-Chat"]),
+                ResolvedRequestContext::default(),
+            )
+            .expect("request is valid")
+            .chat_request
+            .sampling_params
+            .thinking_token_budget
+        };
+
+        // The convert layer forwards the raw value verbatim (including the `-1`
+        // "unlimited" sentinel); normalization/validation happens during
+        // lowering (see `vllm_text::lower`).
+        assert_eq!(prepare(Some(64)), Some(64));
+        assert_eq!(prepare(Some(-1)), Some(-1));
+        assert_eq!(prepare(None), None);
+    }
+
     #[test]
     fn prepare_chat_request_accepts_developer_messages() {
         let request = ChatCompletionRequest {
@@ -941,7 +969,7 @@ mod tests {
         let request = ChatCompletionRequest {
             stream: false,
             logprobs: true,
-            prompt_logprobs: Some(2),
+            prompt_logprobs: Some(LogprobsCount::Top(2)),
             ..base_request()
         };
 
@@ -954,10 +982,13 @@ mod tests {
 
         assert!(prepared.options.requested_logprobs);
         assert!(prepared.options.include_prompt_logprobs);
-        assert_eq!(prepared.chat_request.sampling_params.logprobs, Some(0));
+        assert_eq!(
+            prepared.chat_request.sampling_params.logprobs,
+            Some(LogprobsCount::Top(0))
+        );
         assert_eq!(
             prepared.chat_request.sampling_params.prompt_logprobs,
-            Some(2)
+            Some(LogprobsCount::Top(2))
         );
     }
 
@@ -965,7 +996,7 @@ mod tests {
     fn prepare_chat_request_keeps_prompt_logprobs_independent_from_echo() {
         let request = ChatCompletionRequest {
             logprobs: true,
-            top_logprobs: Some(3),
+            top_logprobs: Some(LogprobsCount::Top(3)),
             echo: true,
             ..base_request()
         };
@@ -977,7 +1008,10 @@ mod tests {
         )
         .expect("request is valid");
 
-        assert_eq!(prepared.chat_request.sampling_params.logprobs, Some(3));
+        assert_eq!(
+            prepared.chat_request.sampling_params.logprobs,
+            Some(LogprobsCount::Top(3))
+        );
         assert_eq!(prepared.chat_request.sampling_params.prompt_logprobs, None);
         assert!(!prepared.options.include_prompt_logprobs);
     }

rust/src/server/src/routes/openai/chat_completions/types.rs

@@ -6,6 +6,7 @@ use serde_json::Value;
 use serde_with::SerializeDisplay;
 use validator::Validate;
 use vllm_chat::ReasoningEffort;
+use vllm_engine_core_client::protocol::LogprobsCount;
 
 use crate::routes::openai::utils::structured_outputs::ResponseFormat;
 use crate::routes::openai::utils::types::{
@@ -44,10 +45,8 @@ pub struct ChatCompletionRequest {
     #[serde(default)]
     pub logprobs: bool,
 
-    /// An integer specifying the number of most likely tokens to return
-    /// -1 means return all
-    #[validate(range(min = -1))]
-    pub top_logprobs: Option<i32>,
+    /// Number of most likely tokens to return. `-1` means return full vocab.
+    pub top_logprobs: Option<LogprobsCount>,
 
     /// Deprecated: Replaced by max_completion_tokens
     #[deprecated(note = "Use max_completion_tokens instead")]
@@ -155,8 +154,8 @@ pub struct ChatCompletionRequest {
     /// Truncate prompt tokens to this length
     pub truncate_prompt_tokens: Option<i64>,
 
-    /// Number of prompt logprobs to return
-    pub prompt_logprobs: Option<i32>,
+    /// Number of prompt logprobs to return. `-1` means return full vocab.
+    pub prompt_logprobs: Option<LogprobsCount>,
 
     /// Restrict output to these token IDs only
     pub allowed_token_ids: Option<Vec<u32>>,
@@ -165,8 +164,10 @@ pub struct ChatCompletionRequest {
     pub bad_words: Option<Vec<String>>,
 
     // -------- Extra vLLM Parameters --------
-    /// Token budget for reasoning/thinking
-    pub thinking_token_budget: Option<u32>,
+    /// Token budget for reasoning/thinking. Accepts a non-negative integer, or
+    /// `-1` for unlimited (mirroring the Python frontend, which normalizes `-1`
+    /// to "no budget").
+    pub thinking_token_budget: Option<i64>,
 
     /// Whether to include reasoning content in the response
     #[serde(default = "default_true")]

rust/src/server/src/routes/openai/chat_completions/validate.rs

@@ -1,6 +1,7 @@
 use super::types::ChatCompletionRequest;
 use crate::error::{ApiError, bail_invalid_request};
 use crate::routes::openai::utils::types::{ChatMessage, Tool, ToolChoice, ToolChoiceValue};
+use vllm_engine_core_client::protocol::LogprobsCount;
 
 /// Enforce the minimal compatibility contract for the Rust OpenAI server.
 pub(super) fn validate_request_compat(
@@ -30,14 +31,12 @@ pub(super) fn validate_request_compat(
     }
 
     if let Some(prompt_logprobs) = request.prompt_logprobs {
-        if prompt_logprobs < 0 && prompt_logprobs != -1 {
-            bail_invalid_request!(
-                param = "prompt_logprobs",
-                "prompt_logprobs must be a non-negative value or -1."
-            );
-        }
-
-        if request.stream && (prompt_logprobs > 0 || prompt_logprobs == -1) {
+        if request.stream
+            && matches!(
+                prompt_logprobs,
+                LogprobsCount::All | LogprobsCount::Top(1..)
+            )
+        {
             bail_invalid_request!(
                 param = "prompt_logprobs",
                 "prompt_logprobs are not available when stream=true."
@@ -108,11 +107,6 @@ pub(super) fn validate_request_compat(
         "truncate_prompt_tokens",
         "truncate_prompt_tokens is not supported.",
     )?;
-    reject_non_default(
-        request.thinking_token_budget.as_ref(),
-        "thinking_token_budget",
-        "thinking_token_budget is not supported.",
-    )?;
     reject_non_default(
         request.media_io_kwargs.as_ref(),
         "media_io_kwargs",
@@ -159,6 +153,7 @@ mod tests {
 
     use serde_json::json;
     use vllm_chat::ReasoningEffort;
+    use vllm_engine_core_client::protocol::LogprobsCount;
 
     use super::validate_request_compat;
     use crate::routes::openai::chat_completions::types::ChatCompletionRequest;
@@ -304,7 +299,7 @@ mod tests {
     #[test]
     fn validate_request_compat_rejects_top_logprobs_without_logprobs() {
         let request = ChatCompletionRequest {
-            top_logprobs: Some(0),
+            top_logprobs: Some(LogprobsCount::Top(0)),
             ..base_request()
         };
         assert!(validate_request_compat(&request, &served(&["Qwen/Qwen1.5-0.5B-Chat"])).is_err());
@@ -313,26 +308,26 @@ mod tests {
     #[test]
     fn validate_request_compat_rejects_streaming_prompt_logprobs_requests() {
         let request = ChatCompletionRequest {
-            prompt_logprobs: Some(1),
+            prompt_logprobs: Some(LogprobsCount::Top(1)),
             ..base_request()
         };
         assert!(validate_request_compat(&request, &served(&["Qwen/Qwen1.5-0.5B-Chat"])).is_err());
 
         let request = ChatCompletionRequest {
-            prompt_logprobs: Some(-1),
+            prompt_logprobs: Some(LogprobsCount::All),
             ..base_request()
         };
         assert!(validate_request_compat(&request, &served(&["Qwen/Qwen1.5-0.5B-Chat"])).is_err());
     }
 
     #[test]
-    fn validate_request_compat_rejects_invalid_prompt_logprobs_value() {
-        let request = ChatCompletionRequest {
-            stream: false,
-            prompt_logprobs: Some(-2),
-            ..base_request()
-        };
-        assert!(validate_request_compat(&request, &served(&["Qwen/Qwen1.5-0.5B-Chat"])).is_err());
+    fn chat_request_deserialization_rejects_invalid_prompt_logprobs_value() {
+        let result = serde_json::from_value::<ChatCompletionRequest>(json!({
+            "model": "Qwen/Qwen1.5-0.5B-Chat",
+            "messages": [{"role": "user", "content": "hello"}],
+            "prompt_logprobs": -2
+        }));
+        assert!(result.is_err());
     }
 
     #[test]

rust/src/server/src/routes/openai/completions/convert.rs

@@ -1,3 +1,4 @@
+use vllm_engine_core_client::protocol::LogprobsCount;
 use vllm_text::{SamplingParams, TextDecodeOptions, TextRequest};
 
 use super::types::CompletionRequest;
@@ -61,15 +62,7 @@ pub(super) fn prepare_completion_request(
         .map(|request| request.lora_name.clone())
         .unwrap_or_else(|| lora_resolution.model_names.first().cloned().unwrap_or_default());
 
-    let logprobs = match request.logprobs {
-        Some(logprobs) => Some(i32::try_from(logprobs).map_err(|_| {
-            ApiError::invalid_request(
-                "`logprobs` must fit within a signed 32-bit integer.".to_string(),
-                Some("logprobs"),
-            )
-        })?),
-        None => None,
-    };
+    let logprobs = request.logprobs.map(LogprobsCount::Top);
     let prompt_only = request.echo && request.max_tokens == Some(0);
     let prompt_logprobs =
         request.prompt_logprobs.or(if request.echo && (!request.stream || prompt_only) {
@@ -108,6 +101,7 @@ pub(super) fn prepare_completion_request(
             seed: request.seed,
             max_tokens,
             min_tokens: request.min_tokens,
+            thinking_token_budget: request.thinking_token_budget,
             logprobs,
             prompt_logprobs,
             min_p: request.min_p,
@@ -162,6 +156,7 @@ pub(super) fn prepare_completion_request(
 mod tests {
     use axum::http::HeaderMap;
     use serde_json::json;
+    use vllm_engine_core_client::protocol::LogprobsCount;
     use vllm_text::Prompt;
 
     use super::prepare_completion_request;
@@ -246,7 +241,10 @@ mod tests {
             Prompt::TokenIds(vec![11, 22, 33])
         );
         assert_eq!(prepared.text_request.sampling_params.max_tokens, Some(7));
-        assert_eq!(prepared.text_request.sampling_params.logprobs, Some(2));
+        assert_eq!(
+            prepared.text_request.sampling_params.logprobs,
+            Some(LogprobsCount::Top(2))
+        );
         assert_eq!(prepared.text_request.sampling_params.top_p, Some(0.9));
         assert_eq!(prepared.text_request.sampling_params.top_k, Some(42));
         assert_eq!(prepared.text_request.sampling_params.min_p, Some(0.1));
@@ -266,6 +264,34 @@ mod tests {
         assert!(!prepared.text_request.decode_options.skip_special_tokens);
     }
 
+    #[test]
+    fn prepare_completion_request_passes_through_thinking_token_budget() {
+        let prepare = |budget: serde_json::Value| {
+            let request: CompletionRequest = serde_json::from_value(json!({
+                "model": "Qwen/Qwen1.5-0.5B-Chat",
+                "prompt": "hello",
+                "thinking_token_budget": budget,
+            }))
+            .expect("parse request");
+            prepare_completion_request(
+                request,
+                &served(&["Qwen/Qwen1.5-0.5B-Chat"]),
+                ResolvedRequestContext::default(),
+            )
+            .expect("prepare")
+            .text_request
+            .sampling_params
+            .thinking_token_budget
+        };
+
+        // The convert layer forwards the raw value verbatim (including the `-1`
+        // "unlimited" sentinel); normalization/validation happens during
+        // lowering (see `vllm_text::lower`).
+        assert_eq!(prepare(json!(64)), Some(64));
+        assert_eq!(prepare(json!(-1)), Some(-1));
+        assert_eq!(prepare(json!(null)), None);
+    }
+
     #[test]
     fn prepare_completion_request_maps_stream_usage_and_token_format_options() {
         let request: CompletionRequest = serde_json::from_value(json!({
@@ -381,10 +407,13 @@ mod tests {
         .expect("prepare");
 
         assert!(prepared.options.prompt_only);
-        assert_eq!(prepared.text_request.sampling_params.logprobs, Some(3));
+        assert_eq!(
+            prepared.text_request.sampling_params.logprobs,
+            Some(LogprobsCount::Top(3))
+        );
         assert_eq!(
             prepared.text_request.sampling_params.prompt_logprobs,
-            Some(3)
+            Some(LogprobsCount::Top(3))
         );
     }
 
@@ -406,10 +435,13 @@ mod tests {
         )
         .expect("prepare");
 
-        assert_eq!(prepared.text_request.sampling_params.logprobs, Some(3));
+        assert_eq!(
+            prepared.text_request.sampling_params.logprobs,
+            Some(LogprobsCount::Top(3))
+        );
         assert_eq!(
             prepared.text_request.sampling_params.prompt_logprobs,
-            Some(3)
+            Some(LogprobsCount::Top(3))
         );
     }
 
@@ -450,10 +482,13 @@ mod tests {
             ResolvedRequestContext::default(),
         )
         .expect("prepare");
-        assert_eq!(prepared.text_request.sampling_params.logprobs, Some(1));
+        assert_eq!(
+            prepared.text_request.sampling_params.logprobs,
+            Some(LogprobsCount::Top(1))
+        );
         assert_eq!(
             prepared.text_request.sampling_params.prompt_logprobs,
-            Some(2)
+            Some(LogprobsCount::Top(2))
         );
     }
 

rust/src/server/src/routes/openai/completions/types.rs

@@ -3,6 +3,7 @@ use std::collections::HashMap;
 use serde::{Deserialize, Serialize};
 use serde_json::{Map, Value};
 use validator::Validate;
+use vllm_engine_core_client::protocol::LogprobsCount;
 use vllm_text::Prompt;
 
 use crate::routes::openai::utils::types::{
@@ -131,8 +132,8 @@ pub struct CompletionRequest {
     /// Restrict output to these token IDs only
     pub allowed_token_ids: Option<Vec<u32>>,
 
-    /// Number of prompt logprobs to return
-    pub prompt_logprobs: Option<i32>,
+    /// Number of prompt logprobs to return. `-1` means return full vocab.
+    pub prompt_logprobs: Option<LogprobsCount>,
 
     // -------- Extra vLLM Parameters --------
     /// Whether to add special tokens (e.g. BOS) to the prompt
@@ -146,6 +147,11 @@ pub struct CompletionRequest {
     /// Additional kwargs for structured outputs
     pub structured_outputs: Option<Value>,
 
+    /// Token budget for reasoning/thinking. Accepts a non-negative integer, or
+    /// `-1` for unlimited (mirroring the Python frontend, which normalizes `-1`
+    /// to "no budget").
+    pub thinking_token_budget: Option<i64>,
+
     /// Request scheduling priority (lower means earlier; default 0)
     pub priority: Option<i32>,
 

rust/src/server/src/routes/openai/completions/validate.rs

@@ -1,3 +1,4 @@
+use vllm_engine_core_client::protocol::LogprobsCount;
 use vllm_text::Prompt;
 
 use super::types::CompletionRequest;
@@ -44,29 +45,18 @@ pub(super) fn validate_request_compat(
         bail_invalid_request!(param = "suffix", "suffix is not supported.");
     }
 
-    if let Some(logprobs) = request.logprobs
-        && logprobs > i32::MAX as u32
-    {
-        bail_invalid_request!(
-            param = "logprobs",
-            "`logprobs` must fit within a signed 32-bit integer."
-        );
-    }
-
     if let Some(prompt_logprobs) = request.prompt_logprobs {
-        if request.stream && (prompt_logprobs > 0 || prompt_logprobs == -1) {
+        if request.stream
+            && matches!(
+                prompt_logprobs,
+                LogprobsCount::All | LogprobsCount::Top(1..)
+            )
+        {
             bail_invalid_request!(
                 param = "prompt_logprobs",
                 "`prompt_logprobs` are not available when `stream=true`."
             );
         }
-
-        if prompt_logprobs < 0 && prompt_logprobs != -1 {
-            bail_invalid_request!(
-                param = "prompt_logprobs",
-                "`prompt_logprobs` must be a non-negative value or -1."
-            );
-        }
     }
 
     if request.use_beam_search {
@@ -101,6 +91,7 @@ pub(super) fn validate_request_compat(
 #[cfg(test)]
 mod tests {
     use serde_json::json;
+    use vllm_engine_core_client::protocol::LogprobsCount;
 
     use super::validate_request_compat;
     use crate::routes::openai::completions::types::CompletionRequest;
@@ -150,7 +141,7 @@ mod tests {
     #[test]
     fn validate_request_compat_rejects_streaming_prompt_logprobs() {
         let request = CompletionRequest {
-            prompt_logprobs: Some(1),
+            prompt_logprobs: Some(LogprobsCount::Top(1)),
             ..base_request()
         };
         assert!(
@@ -162,7 +153,7 @@ mod tests {
     fn validate_request_compat_accepts_non_stream_prompt_logprobs() {
         let request = CompletionRequest {
             stream: false,
-            prompt_logprobs: Some(-1),
+            prompt_logprobs: Some(LogprobsCount::All),
             ..base_request()
         };
         assert!(

rust/src/text/src/backend/mod.rs

@@ -2,6 +2,7 @@ pub mod hf;
 
 use std::sync::Arc;
 
+use vllm_engine_core_client::protocol::LogprobsCount;
 use vllm_tokenizer::DynTokenizer;
 
 use crate::error::Result;
@@ -26,9 +27,7 @@ pub struct SamplingLimits {
     /// Runtime context window size reported by the engine startup handshake.
     pub max_model_len: u32,
     /// Maximum number of top log probabilities accepted by this frontend.
-    ///
-    /// `-1` means allowing requests up to the model vocabulary size.
-    pub max_logprobs: i32,
+    pub max_logprobs: LogprobsCount,
 
     /// Model vocabulary size from the model config, used to bound generated
     /// token IDs and logits-domain sampling controls.
@@ -41,7 +40,7 @@ pub struct SamplingLimits {
 impl SamplingLimits {
     /// Original Python definition:
     /// <https://github.com/vllm-project/vllm/blob/b5adb027ad03c29b46181752ba3b1cb84eff1dd4/vllm/config/model.py#L216-L220>
-    pub const DEFAULT_MAX_LOGPROBS: i32 = 20;
+    pub const DEFAULT_MAX_LOGPROBS: LogprobsCount = LogprobsCount::Top(20);
     /// Original Python definition:
     /// <https://github.com/vllm-project/vllm/blob/b5adb027ad03c29b46181752ba3b1cb84eff1dd4/vllm/sampling_params.py#L30-L32>
     pub const MAX_LOGPROB_TOKEN_IDS: usize = 128;

rust/src/text/src/error.rs

@@ -20,6 +20,8 @@ pub enum Error {
     Logprobs(#[from] LogprobsError),
     #[error(transparent)]
     OutOfVocab(#[from] OutOfVocabError),
+    #[error("`thinking_token_budget` must be a non-negative integer or -1 for unlimited.")]
+    InvalidThinkingTokenBudget,
     #[error("text request stream `{request_id}` closed before terminal output")]
     StreamClosedBeforeTerminalOutput { request_id: String },
     #[error(transparent)]

rust/src/text/src/lib.rs

@@ -19,6 +19,7 @@ pub use output::{
 pub use request::{Prompt, SamplingParams, TextRequest};
 use trait_set::trait_set;
 use vllm_engine_core_client::EngineCoreClient;
+use vllm_engine_core_client::protocol::LogprobsCount;
 pub use vllm_llm::FinishReason;
 use vllm_llm::{GenerateOutputStream, Llm};
 use vllm_tokenizer::DynTokenizer;
@@ -48,7 +49,7 @@ pub struct TextLlm {
     /// Runtime context window size reported by the engine startup handshake.
     max_model_len: u32,
     /// Maximum number of top log probabilities accepted by this text facade.
-    max_logprobs: i32,
+    max_logprobs: LogprobsCount,
 }
 
 impl TextLlm {
@@ -68,7 +69,7 @@ impl TextLlm {
     }
 
     /// Override the maximum accepted logprobs count.
-    pub fn with_max_logprobs(mut self, max_logprobs: Option<i32>) -> Self {
+    pub fn with_max_logprobs(mut self, max_logprobs: Option<LogprobsCount>) -> Self {
         if let Some(max_logprobs) = max_logprobs {
             self.max_logprobs = max_logprobs;
         }

rust/src/text/src/lower.rs

@@ -87,6 +87,7 @@ pub fn lower_sampling_params(
         seed,
         max_tokens,
         min_tokens,
+        thinking_token_budget,
         logprobs,
         prompt_logprobs,
         min_p,
@@ -128,6 +129,7 @@ pub fn lower_sampling_params(
         prompt_len,
     )?;
     let min_tokens = min_tokens.unwrap_or(0);
+    let thinking_token_budget = normalize_thinking_token_budget(thinking_token_budget)?;
     let frequency_penalty = frequency_penalty.unwrap_or(0.0);
     let presence_penalty = presence_penalty.unwrap_or(0.0);
 
@@ -149,6 +151,7 @@ pub fn lower_sampling_params(
         seed,
         max_tokens,
         min_tokens,
+        thinking_token_budget,
         logprobs,
         prompt_logprobs,
         min_p,
@@ -170,6 +173,21 @@ pub fn lower_sampling_params(
     Ok(params)
 }
 
+/// Normalize the user-facing `thinking_token_budget` into the engine value.
+///
+/// Mirrors Python's `validate_thinking_token_budget`
+/// (<https://github.com/vllm-project/vllm/blob/ecf9d83520eb217401b47d8a5451a27c5231b8c2/vllm/sampling_params.py#L35-L55>):
+/// `None` and the `-1` "unlimited" sentinel both map to `None`; any other
+/// negative value is rejected; non-negative values pass through unchanged. Like
+/// Python's `int`, no upper bound is imposed.
+fn normalize_thinking_token_budget(value: Option<i64>) -> Result<Option<u64>> {
+    match value {
+        None | Some(-1) => Ok(None),
+        Some(budget) if budget >= 0 => Ok(Some(budget as u64)),
+        Some(_) => Err(Error::InvalidThinkingTokenBudget),
+    }
+}
+
 /// Convert bad-word strings into token-ID sequences, following the Python vLLM
 /// logic in `SamplingParams.update_from_tokenizer()`.
 ///
@@ -251,6 +269,7 @@ mod tests {
     use std::collections::{BTreeSet, HashMap};
 
     use serial_test::file_serial;
+    use vllm_engine_core_client::protocol::LogprobsCount;
 
     use super::*;
     use crate::backend::hf::HfTextBackend;
@@ -366,6 +385,36 @@ mod tests {
         )
     }
 
+    #[test]
+    fn lower_sampling_params_normalizes_thinking_token_budget() {
+        let lower = |budget: Option<i64>| {
+            lower_sampling_params_with_limits(
+                SamplingParams {
+                    thinking_token_budget: budget,
+                    ..SamplingParams::default()
+                },
+                sample_sampling_limits(),
+            )
+        };
+
+        // Non-negative budgets (including 0) pass through unchanged.
+        assert_eq!(lower(Some(256)).unwrap().thinking_token_budget, Some(256));
+        assert_eq!(lower(Some(0)).unwrap().thinking_token_budget, Some(0));
+        // `None` and the `-1` "unlimited" sentinel both disable the budget.
+        assert_eq!(lower(None).unwrap().thinking_token_budget, None);
+        assert_eq!(lower(Some(-1)).unwrap().thinking_token_budget, None);
+        // No upper bound is imposed, matching Python's `int`.
+        assert_eq!(
+            lower(Some(i64::from(u32::MAX) + 1)).unwrap().thinking_token_budget,
+            Some(u64::from(u32::MAX) + 1)
+        );
+        // Other negatives are rejected.
+        assert!(matches!(
+            lower(Some(-2)),
+            Err(Error::InvalidThinkingTokenBudget)
+        ));
+    }
+
     #[test]
     fn lower_text_request_applies_python_style_eos_hints() {
         let prepared = lower_text_request(
@@ -386,6 +435,7 @@ mod tests {
                 seed: None,
                 max_tokens: 999997,
                 min_tokens: 0,
+                thinking_token_budget: None,
                 logprobs: None,
                 prompt_logprobs: None,
                 min_p: 0.0,
@@ -437,6 +487,7 @@ mod tests {
                 seed: None,
                 max_tokens: 999997,
                 min_tokens: 0,
+                thinking_token_budget: None,
                 logprobs: None,
                 prompt_logprobs: None,
                 min_p: 0.0,
@@ -567,6 +618,7 @@ mod tests {
                 seed: None,
                 max_tokens: 40957,
                 min_tokens: 0,
+                thinking_token_budget: None,
                 logprobs: None,
                 prompt_logprobs: None,
                 min_p: 0.0,
@@ -628,6 +680,7 @@ mod tests {
                 seed: None,
                 max_tokens: 999997,
                 min_tokens: 0,
+                thinking_token_budget: None,
                 logprobs: None,
                 prompt_logprobs: None,
                 min_p: 0.0,
@@ -697,6 +750,7 @@ mod tests {
                 seed: None,
                 max_tokens: 32,
                 min_tokens: 2,
+                thinking_token_budget: None,
                 logprobs: None,
                 prompt_logprobs: None,
                 min_p: 0.1,
@@ -721,8 +775,8 @@ mod tests {
     #[test]
     fn lower_sampling_params_passes_logprobs_fields_through() {
         let sampling_params = SamplingParams {
-            logprobs: Some(3),
-            prompt_logprobs: Some(-1),
+            logprobs: Some(LogprobsCount::Top(3)),
+            prompt_logprobs: Some(LogprobsCount::All),
             ..Default::default()
         };
 
@@ -739,7 +793,7 @@ mod tests {
                 default_max_tokens: None,
             },
             SamplingLimits {
-                max_logprobs: -1,
+                max_logprobs: LogprobsCount::All,
                 ..sample_sampling_limits()
             },
             3,
@@ -747,15 +801,15 @@ mod tests {
         )
         .unwrap();
 
-        assert_eq!(params.logprobs, Some(3));
-        assert_eq!(params.prompt_logprobs, Some(-1));
+        assert_eq!(params.logprobs, Some(LogprobsCount::Top(3)));
+        assert_eq!(params.prompt_logprobs, Some(LogprobsCount::All));
     }
 
     #[test]
     fn lower_sampling_params_rejects_full_vocab_logprobs_over_default_cap() {
         let error = lower_sampling_params_with_limits(
             SamplingParams {
-                logprobs: Some(-1),
+                logprobs: Some(LogprobsCount::All),
                 ..Default::default()
             },
             sample_sampling_limits(),
@@ -776,24 +830,24 @@ mod tests {
     fn lower_sampling_params_expands_full_vocab_logprobs_from_model_vocab() {
         let params = lower_sampling_params_with_limits(
             SamplingParams {
-                logprobs: Some(-1),
+                logprobs: Some(LogprobsCount::All),
                 ..Default::default()
             },
             SamplingLimits {
-                max_logprobs: 1500,
+                max_logprobs: LogprobsCount::Top(1500),
                 ..sample_sampling_limits()
             },
         )
         .unwrap();
 
-        assert_eq!(params.logprobs, Some(-1));
+        assert_eq!(params.logprobs, Some(LogprobsCount::All));
     }
 
     #[test]
     fn lower_sampling_params_rejects_invalid_logprob_token_ids() {
         let error = lower_sampling_params_with_limits(
             SamplingParams {
-                logprobs: Some(1),
+                logprobs: Some(LogprobsCount::Top(1)),
                 logprob_token_ids: Some(vec![1000]),
                 ..Default::default()
             },
@@ -929,6 +983,7 @@ mod tests {
                 seed: None,
                 max_tokens: 128,
                 min_tokens: 0,
+                thinking_token_budget: None,
                 logprobs: None,
                 prompt_logprobs: None,
                 min_p: 0.1,

rust/src/text/src/lower/logprobs.rs

@@ -1,15 +1,14 @@
 //! Python-compatible validation for logprobs sampling params.
 //!
-//! `-1` is expanded only for bounds checks. The original request values are
+//! `All` is expanded only for bounds checks. The original request values are
 //! passed through to engine-core.
 
 use crate::backend::SamplingLimits;
 use thiserror::Error;
+use vllm_engine_core_client::protocol::LogprobsCount;
 
 #[derive(Debug, Error)]
 pub enum LogprobsError {
-    #[error("{parameter} must be non-negative or -1, got {value}")]
-    InvalidCount { parameter: &'static str, value: i32 },
     #[error(
         "requested {parameter} of {requested}, which is greater than max allowed: {max_allowed}"
     )]
@@ -30,19 +29,21 @@ pub enum LogprobsError {
         "when both logprobs and logprob_token_ids are set, logprobs must equal \
          len(logprob_token_ids). Got logprobs={logprobs}, len(logprob_token_ids)={num_token_ids}."
     )]
-    TokenIdsMismatch { logprobs: i32, num_token_ids: usize },
+    TokenIdsMismatch {
+        logprobs: LogprobsCount,
+        num_token_ids: usize,
+    },
 }
 
 /// Validate logprobs count sampling parameters.
 pub(super) fn validate_logprobs(
-    logprobs: Option<i32>,
-    prompt_logprobs: Option<i32>,
+    logprobs: Option<LogprobsCount>,
+    prompt_logprobs: Option<LogprobsCount>,
     logprob_token_ids: Option<&[u32]>,
     sampling_limits: SamplingLimits,
 ) -> Result<(), LogprobsError> {
     let vocab_size = sampling_limits.model_vocab_size;
-    let max_logprobs =
-        normalize_logprobs_count(sampling_limits.max_logprobs, vocab_size, "max_logprobs")?;
+    let max_logprobs = sampling_limits.max_logprobs.expanded(vocab_size);
 
     validate_logprobs_count(logprobs, max_logprobs, vocab_size, "logprobs")?;
     validate_logprobs_count(prompt_logprobs, max_logprobs, vocab_size, "prompt_logprobs")?;
@@ -50,7 +51,7 @@ pub(super) fn validate_logprobs(
 }
 
 fn validate_logprobs_count(
-    requested: Option<i32>,
+    requested: Option<LogprobsCount>,
     max_logprobs: usize,
     vocab_size: usize,
     parameter: &'static str,
@@ -59,7 +60,7 @@ fn validate_logprobs_count(
         return Ok(());
     };
 
-    let requested = normalize_logprobs_count(requested, vocab_size, parameter)?;
+    let requested = requested.expanded(vocab_size);
     if requested > max_logprobs {
         return Err(LogprobsError::TooManyCount {
             parameter,
@@ -72,7 +73,7 @@ fn validate_logprobs_count(
 }
 
 pub(super) fn validate_logprob_token_ids(
-    logprobs: Option<i32>,
+    logprobs: Option<LogprobsCount>,
     logprob_token_ids: Option<&[u32]>,
 ) -> Result<(), LogprobsError> {
     let Some(logprob_token_ids) = logprob_token_ids else {
@@ -88,7 +89,7 @@ pub(super) fn validate_logprob_token_ids(
     }
 
     if let Some(logprobs) = logprobs
-        && logprobs != n as i32
+        && logprobs != LogprobsCount::Top(n as u32)
     {
         return Err(LogprobsError::TokenIdsMismatch {
             logprobs,
@@ -98,15 +99,3 @@ pub(super) fn validate_logprob_token_ids(
 
     Ok(())
 }
-
-fn normalize_logprobs_count(
-    value: i32,
-    vocab_size: usize,
-    parameter: &'static str,
-) -> Result<usize, LogprobsError> {
-    match value {
-        -1 => Ok(vocab_size),
-        value if value < 0 => Err(LogprobsError::InvalidCount { parameter, value }),
-        value => Ok(value as usize),
-    }
-}

rust/src/text/src/output/decoded.rs

@@ -309,7 +309,7 @@ fn matches_stop_string(stops: &[String], output: &str, new_bytes: usize) -> Opti
         .find_map(|(ss_idx, (ss, len, start_off))| {
             output[start_off..]
                 .windows(len)
-                .rposition(|w| w == ss)
+                .position(|w| w == ss)
                 .map(|pos| (ss_idx, start_off + pos))
         })
 }
@@ -562,6 +562,13 @@ mod tests {
         assert_eq!(result, Some((0, 4)));
     }
 
+    #[test]
+    fn stop_string_matches_leftmost_with_multiple_new_bytes() {
+        let stops = vec!["\n".to_string()];
+        let result = matches_stop_string(&stops, "Answer\n\n", 2);
+        assert_eq!(result, Some((0, 6)));
+    }
+
     #[test]
     fn stop_string_matches_at_beginning() {
         let stops = vec!["say".to_string()];

rust/src/text/src/request.rs

@@ -3,9 +3,9 @@ use std::collections::HashMap;
 use enum_as_inner::EnumAsInner;
 use serde::{Deserialize, Serialize};
 use serde_json::Value;
-use vllm_engine_core_client::protocol::StructuredOutputsParams;
 use vllm_engine_core_client::protocol::lora::LoraRequest;
 use vllm_engine_core_client::protocol::multimodal::MmFeatures;
+use vllm_engine_core_client::protocol::{LogprobsCount, StructuredOutputsParams};
 
 use crate::error::{Error, Result};
 use crate::output::TextDecodeOptions;
@@ -56,14 +56,20 @@ pub struct SamplingParams {
     pub max_tokens: Option<u32>,
     /// Minimum number of tokens to generate before EOS or stop-token handling.
     pub min_tokens: Option<u32>,
+    /// Maximum number of reasoning ("thinking") tokens to emit before the
+    /// reasoning section is force-closed. `None` or the user-facing `-1`
+    /// "unlimited" sentinel both disable the budget. The raw value is carried
+    /// here; `-1` is normalized to `None` (and other negatives rejected) during
+    /// lowering (see `lower_sampling_params`).
+    pub thinking_token_budget: Option<i64>,
     /// Number of log probabilities to return per generated token.
     ///
-    /// `None` disables sample logprobs. `-1` requests the full vocabulary.
-    pub logprobs: Option<i32>,
+    /// `None` disables sample logprobs.
+    pub logprobs: Option<LogprobsCount>,
     /// Number of log probabilities to return per prompt token.
     ///
-    /// `None` disables prompt logprobs. `-1` requests the full vocabulary.
-    pub prompt_logprobs: Option<i32>,
+    /// `None` disables prompt logprobs.
+    pub prompt_logprobs: Option<LogprobsCount>,
     /// Minimum probability threshold for token sampling. `None` means no
     /// explicit user override.
     pub min_p: Option<f32>,
@@ -116,6 +122,7 @@ impl Default for SamplingParams {
             seed: None,
             max_tokens: None,
             min_tokens: None,
+            thinking_token_budget: None,
             logprobs: None,
             prompt_logprobs: None,
             min_p: None,

tests/compile/passes/test_double_aiter_rms_quant_fusion.py

@@ -22,7 +22,7 @@ import torch
 
 import vllm.config
 from tests.compile.backend import TestBackend
-from vllm._aiter_ops import is_aiter_found_and_supported, rocm_aiter_ops
+from vllm._aiter_ops import rocm_aiter_ops
 from vllm.compilation.passes.utility.noop_elimination import NoOpEliminationPass
 from vllm.compilation.passes.utility.post_cleanup import PostCleanupPass
 from vllm.config import (
@@ -83,9 +83,8 @@ class _ViewDoubleQuantModel(torch.nn.Module):
     [_NoViewDoubleQuantModel, _ViewDoubleQuantModel],
     ids=["no_view", "with_view"],
 )
-@pytest.mark.skipif(
-    not is_aiter_found_and_supported(),
-    reason="Only test on ROCm with AITER installed and supported",
+@pytest.mark.skip(
+    reason="Skipping for now because pytorch compiler removes one the two quant ops"
 )
 def test_double_aiter_rms_fp8_group_quant_fusion(
     model_cls: type[torch.nn.Module],

tests/distributed/test_pipeline_parallel.py

@@ -175,7 +175,7 @@ MULTIMODAL_MODELS = {
     "facebook/chameleon-7b": PPTestSettings.fast(),
     "adept/fuyu-8b": PPTestSettings.fast(),
     "zai-org/glm-4v-9b": PPTestSettings.fast(),
-    "OpenGVLab/InternVL2-1B": PPTestSettings.fast(),
+    "OpenGVLab/InternVL3-1B": PPTestSettings.fast(),
     "llava-hf/llava-1.5-7b-hf": PPTestSettings.fast(),
     "llava-hf/llava-v1.6-mistral-7b-hf": PPTestSettings.fast(),
     "llava-hf/LLaVA-NeXT-Video-7B-hf": PPTestSettings.fast(),
@@ -203,7 +203,7 @@ TEST_MODELS = [
     "intfloat/e5-mistral-7b-instruct",
     "BAAI/bge-multilingual-gemma2",
     # [MULTIMODAL GENERATION]
-    "OpenGVLab/InternVL2-1B",
+    "OpenGVLab/InternVL3-1B",
     "microsoft/Phi-3.5-vision-instruct",
     "fixie-ai/ultravox-v0_5-llama-3_2-1b",
     # [LANGUAGE GENERATION - HYBRID ARCH]

tests/entrypoints/anthropic/test_anthropic_messages_conversion.py

@@ -8,6 +8,8 @@ AnthropicServingMessages._convert_anthropic_to_openai_request().
 Also covers extended-thinking edge cases such as ``redacted_thinking``
 blocks echoed back by Anthropic clients, and streaming conversion in
 ``message_stream_converter``.
+
+Also covers cache usage computation in ``_build_anthropic_usage``.
 """
 
 import json
@@ -18,7 +20,11 @@ import pytest
 from vllm.entrypoints.anthropic.protocol import (
     AnthropicMessagesRequest,
 )
-from vllm.entrypoints.anthropic.serving import AnthropicServingMessages
+from vllm.entrypoints.anthropic.serving import (
+    AnthropicServingMessages,
+    _build_anthropic_usage,
+    _get_cached_tokens,
+)
 from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionResponseStreamChoice,
     ChatCompletionStreamResponse,
@@ -27,6 +33,7 @@ from vllm.entrypoints.openai.engine.protocol import (
     DeltaFunctionCall,
     DeltaMessage,
     DeltaToolCall,
+    PromptTokenUsageInfo,
     UsageInfo,
 )
 
@@ -653,6 +660,108 @@ class TestThinkingBlockConversion:
         assert asst.get("content") == "Hi!"
 
 
+# ======================================================================
+# Cache usage computation
+# ======================================================================
+
+
+class TestGetCachedTokens:
+    """Tests for _get_cached_tokens helper."""
+
+    def test_none_usage(self):
+        assert _get_cached_tokens(None) is None
+
+    def test_no_prompt_tokens_details(self):
+        usage = UsageInfo(prompt_tokens=100, completion_tokens=10)
+        assert _get_cached_tokens(usage) is None
+
+    def test_cached_tokens_present(self):
+        usage = UsageInfo(
+            prompt_tokens=100,
+            completion_tokens=10,
+            prompt_tokens_details=PromptTokenUsageInfo(cached_tokens=80),
+        )
+        assert _get_cached_tokens(usage) == 80
+
+    def test_cached_tokens_zero(self):
+        """Zero cached tokens should return 0, not None."""
+        usage = UsageInfo(
+            prompt_tokens=100,
+            completion_tokens=10,
+            prompt_tokens_details=PromptTokenUsageInfo(cached_tokens=0),
+        )
+        assert _get_cached_tokens(usage) == 0
+
+    def test_cached_tokens_none_in_details(self):
+        usage = UsageInfo(
+            prompt_tokens=100,
+            completion_tokens=10,
+            prompt_tokens_details=PromptTokenUsageInfo(cached_tokens=None),
+        )
+        assert _get_cached_tokens(usage) is None
+
+
+class TestBuildAnthropicUsage:
+    """Tests for _build_anthropic_usage helper.
+
+    Anthropic defines: total_input = input_tokens + cache_read + cache_creation
+    vLLM's prompt_tokens is the total.
+    """
+
+    def test_no_cache_info(self):
+        """When cache info is unavailable, return raw prompt_tokens."""
+        result = _build_anthropic_usage(100, 10, None)
+        assert result.input_tokens == 100
+        assert result.output_tokens == 10
+        assert result.cache_read_input_tokens is None
+        assert result.cache_creation_input_tokens is None
+
+    def test_cache_hit(self):
+        """When cache is hit, input_tokens excludes cached tokens."""
+        usage = UsageInfo(
+            prompt_tokens=100,
+            completion_tokens=10,
+            prompt_tokens_details=PromptTokenUsageInfo(cached_tokens=80),
+        )
+        result = _build_anthropic_usage(100, 10, usage)
+        assert result.input_tokens == 20  # 100 - 80
+        assert result.output_tokens == 10
+        assert result.cache_read_input_tokens == 80
+        assert result.cache_creation_input_tokens == 0
+
+    def test_zero_cached_tokens(self):
+        """Zero cached tokens should still set cache_creation to 0."""
+        usage = UsageInfo(
+            prompt_tokens=100,
+            completion_tokens=10,
+            prompt_tokens_details=PromptTokenUsageInfo(cached_tokens=0),
+        )
+        result = _build_anthropic_usage(100, 10, usage)
+        assert result.input_tokens == 100  # 100 - 0
+        assert result.cache_read_input_tokens == 0
+        assert result.cache_creation_input_tokens == 0
+
+    def test_all_tokens_cached(self):
+        """When all tokens are cached, input_tokens should be 0."""
+        usage = UsageInfo(
+            prompt_tokens=100,
+            completion_tokens=10,
+            prompt_tokens_details=PromptTokenUsageInfo(cached_tokens=100),
+        )
+        result = _build_anthropic_usage(100, 10, usage)
+        assert result.input_tokens == 0
+        assert result.cache_read_input_tokens == 100
+        assert result.cache_creation_input_tokens == 0
+
+    def test_no_prompt_tokens_details(self):
+        """UsageInfo without prompt_tokens_details returns no cache info."""
+        usage = UsageInfo(prompt_tokens=100, completion_tokens=10)
+        result = _build_anthropic_usage(100, 10, usage)
+        assert result.input_tokens == 100
+        assert result.cache_read_input_tokens is None
+        assert result.cache_creation_input_tokens is None
+
+
 class TestInlineSystemMessageInMessagesArray:
     """Verify that ``role: system`` messages embedded inside the ``messages``
     array are preserved in their original position.
@@ -1098,6 +1207,135 @@ class TestMessageStartIncludesTypeAndRole:
         assert message["role"] == "assistant"
 
 
+class TestStreamingCacheUsageSemantics:
+    """Locks in the documented streaming behavior of cache usage fields.
+
+    vLLM's OpenAI chat completion streaming only attaches
+    ``prompt_tokens_details`` to the terminal usage chunk. The Anthropic layer
+    mirrors that contract: cache fields are omitted on ``message_start`` (key
+    absence signals "unknown") and populated on ``message_delta`` (the final
+    cumulative count). This is intentionally consistent with vLLM's OpenAI
+    behavior, even though Anthropic's upstream API populates cache fields on
+    ``message_start``; closing that gap requires plumbing cache info into the
+    first chunk at the OpenAI layer, which is out of scope here.
+    """
+
+    @pytest.mark.asyncio
+    async def test_streaming_cache_fields_absent_then_populated(self):
+        """First chunk lacks prompt_tokens_details (vLLM contract);
+        message_start omits cache fields. The final chunk carries
+        prompt_tokens_details, so message_delta carries resolved values."""
+
+        async def sse_input():
+            yield _make_stream_chunk(
+                delta=DeltaMessage(role="assistant", content="hi"),
+                usage=UsageInfo(prompt_tokens=100, total_tokens=100),
+            )
+            yield _make_stream_chunk(finish_reason="stop")
+            yield _make_stream_chunk(
+                choices=[],
+                usage=UsageInfo(
+                    prompt_tokens=100,
+                    completion_tokens=5,
+                    total_tokens=105,
+                    prompt_tokens_details=PromptTokenUsageInfo(cached_tokens=80),
+                ),
+            )
+            yield "data: [DONE]"
+
+        converter = _make_stream_converter()
+        output = []
+        async for event in converter.message_stream_converter(sse_input()):
+            output.append(event)
+        events = _parse_sse_events(output)
+
+        # message_start: cache fields unknown → omitted from JSON entirely.
+        start_usage = events[0][1]["message"]["usage"]
+        assert events[0][0] == "message_start"
+        assert start_usage["input_tokens"] == 100
+        assert "cache_read_input_tokens" not in start_usage
+        assert "cache_creation_input_tokens" not in start_usage
+
+        # message_delta: authoritative usage with cache fields populated.
+        delta_usage = next(
+            data["usage"] for ev, data in events if ev == "message_delta"
+        )
+        assert delta_usage["input_tokens"] == 20  # 100 - 80
+        assert delta_usage["cache_read_input_tokens"] == 80
+        assert delta_usage["cache_creation_input_tokens"] == 0
+
+    @pytest.mark.asyncio
+    async def test_streaming_no_cache_hit(self):
+        """When the final chunk reports cached_tokens=0, message_delta carries
+        cache fields = 0 (cache miss); message_start still omits them."""
+
+        async def sse_input():
+            yield _make_stream_chunk(
+                delta=DeltaMessage(role="assistant"),
+                usage=UsageInfo(prompt_tokens=50, total_tokens=50),
+            )
+            yield _make_stream_chunk(finish_reason="stop")
+            yield _make_stream_chunk(
+                choices=[],
+                usage=UsageInfo(
+                    prompt_tokens=50,
+                    completion_tokens=5,
+                    total_tokens=55,
+                    prompt_tokens_details=PromptTokenUsageInfo(cached_tokens=0),
+                ),
+            )
+            yield "data: [DONE]"
+
+        converter = _make_stream_converter()
+        output = []
+        async for event in converter.message_stream_converter(sse_input()):
+            output.append(event)
+        events = _parse_sse_events(output)
+
+        start_usage = events[0][1]["message"]["usage"]
+        delta_usage = next(
+            data["usage"] for ev, data in events if ev == "message_delta"
+        )
+        assert start_usage["input_tokens"] == 50
+        assert "cache_read_input_tokens" not in start_usage
+        assert "cache_creation_input_tokens" not in start_usage
+        assert delta_usage["input_tokens"] == 50  # 50 - 0
+        assert delta_usage["cache_read_input_tokens"] == 0
+        assert delta_usage["cache_creation_input_tokens"] == 0
+
+    @pytest.mark.asyncio
+    async def test_streaming_no_prompt_tokens_details_at_all(self):
+        """If --enable-prompt-tokens-details is off, no chunk carries cache
+        info; both message_start and message_delta omit cache fields."""
+
+        async def sse_input():
+            yield _make_stream_chunk(
+                delta=DeltaMessage(role="assistant"),
+                usage=UsageInfo(prompt_tokens=30, total_tokens=30),
+            )
+            yield _make_stream_chunk(finish_reason="stop")
+            yield _make_stream_chunk(
+                choices=[],
+                usage=UsageInfo(prompt_tokens=30, completion_tokens=2, total_tokens=32),
+            )
+            yield "data: [DONE]"
+
+        converter = _make_stream_converter()
+        output = []
+        async for event in converter.message_stream_converter(sse_input()):
+            output.append(event)
+        events = _parse_sse_events(output)
+
+        start_usage = events[0][1]["message"]["usage"]
+        delta_usage = next(
+            data["usage"] for ev, data in events if ev == "message_delta"
+        )
+        assert "cache_read_input_tokens" not in start_usage
+        assert "cache_creation_input_tokens" not in start_usage
+        assert "cache_read_input_tokens" not in delta_usage
+        assert "cache_creation_input_tokens" not in delta_usage
+
+
 # ======================================================================
 # Auto-detection of system-first template requirement
 # ======================================================================

tests/entrypoints/pooling/classify/test_online_vision.py

@@ -25,7 +25,7 @@ def server():
         "--runner",
         "pooling",
         "--max-model-len",
-        "5000",
+        "16384",
         "--enforce-eager",
         "--limit-mm-per-prompt",
         json.dumps({"video": MAXIMUM_VIDEOS}),
@@ -143,4 +143,4 @@ def test_chat_video_url_request(server: RemoteOpenAIServer, model_name: str):
     assert output.model == model_name
     assert len(output.data) == 1
     assert len(output.data[0].probs) == 2
-    assert output.usage.prompt_tokens == 4807
+    assert output.usage.prompt_tokens == 8993

tests/entrypoints/serve/render/test_derender.py

@@ -8,6 +8,7 @@ import pytest
 import pytest_asyncio
 
 from tests.utils import RemoteLaunchRenderServer
+from vllm.tokenizers import get_tokenizer
 
 MODEL_NAME = "hmellor/tiny-random-LlamaForCausalLM"
 
@@ -486,3 +487,438 @@ async def test_derender_completion_kv_transfer_params_passthrough(client):
     )
     assert response.status_code == 200
     assert response.json()["kv_transfer_params"] == kv
+
+
+# ---------------------------------------------------------------------------
+# E2E: render -> derender roundtrip with parser (reasoning + tool calls)
+# ---------------------------------------------------------------------------
+
+PARSER_MODEL = "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B"
+
+_E2E_TOOLS = [
+    {
+        "type": "function",
+        "function": {
+            "name": "get_weather",
+            "description": "Get weather for a city",
+            "parameters": {
+                "type": "object",
+                "properties": {"city": {"type": "string"}},
+            },
+        },
+    }
+]
+
+
+@pytest.fixture(scope="module")
+def parser_server():
+    args = [
+        "--enable-auto-tool-choice",
+        "--tool-call-parser",
+        "hermes",
+        "--reasoning-parser",
+        "deepseek_r1",
+    ]
+    with RemoteLaunchRenderServer(PARSER_MODEL, args) as remote_server:
+        yield remote_server
+
+
+@pytest_asyncio.fixture
+async def parser_client(parser_server):
+    async with httpx.AsyncClient(
+        base_url=parser_server.url_for(""), timeout=60.0
+    ) as http_client:
+        yield http_client
+
+
+@pytest.fixture(scope="module")
+def parser_tokenizer():
+    return get_tokenizer(PARSER_MODEL)
+
+
+def _encode(tokenizer, text: str) -> list[int]:
+    return tokenizer.encode(text, add_special_tokens=False)
+
+
+def _decoded(tokenizer, token_ids: list[int]) -> str:
+    return tokenizer.decode(token_ids, skip_special_tokens=True)
+
+
+def _require_markers_survive(tokenizer, text: str, *markers: str) -> list[int]:
+    """Encode text and skip the test if any marker is lost in roundtrip."""
+    ids = _encode(tokenizer, text)
+    decoded = tokenizer.decode(ids, skip_special_tokens=False)
+    for m in markers:
+        if m not in decoded:
+            pytest.skip(f"Marker {m!r} lost in encode->decode roundtrip")
+    return ids
+
+
+async def _e2e_render_chat(
+    client: httpx.AsyncClient,
+    model: str,
+    messages: list[dict],
+) -> dict:
+    resp = await client.post(
+        "/v1/chat/completions/render",
+        json={"model": model, "messages": messages},
+    )
+    assert resp.status_code == 200, resp.text
+    return resp.json()
+
+
+def _e2e_generate_response(
+    token_ids: list[int],
+    request_id: str = "chatcmpl-e2e-test",
+) -> dict:
+    return {
+        "request_id": request_id,
+        "choices": [
+            {
+                "index": 0,
+                "token_ids": token_ids,
+                "finish_reason": "stop",
+            }
+        ],
+    }
+
+
+@pytest.mark.asyncio
+async def test_e2e_plain_roundtrip(parser_client, parser_tokenizer):
+    """Plain text without reasoning markers roundtrips correctly."""
+    messages = [{"role": "user", "content": "What is 2+2?"}]
+    gen_req = await _e2e_render_chat(parser_client, PARSER_MODEL, messages)
+
+    answer = "The answer is four."
+    output_ids = _encode(parser_tokenizer, answer)
+    expected = _decoded(parser_tokenizer, output_ids)
+
+    resp = await parser_client.post(
+        "/v1/chat/completions/derender",
+        json={
+            "model": PARSER_MODEL,
+            "generate_response": _e2e_generate_response(output_ids),
+            "prompt_tokens": len(gen_req["token_ids"]),
+        },
+    )
+    assert resp.status_code == 200, resp.text
+    content = resp.json()["choices"][0]["message"]["content"]
+    assert content == expected
+
+
+@pytest.mark.asyncio
+async def test_e2e_token_identity(parser_client, parser_tokenizer):
+    """encode(derender(token_ids)) == token_ids (RL invariant)."""
+    messages = [{"role": "user", "content": "Hi"}]
+    gen_req = await _e2e_render_chat(parser_client, PARSER_MODEL, messages)
+
+    answer = "Hello! How can I help?"
+    output_ids = _encode(parser_tokenizer, answer)
+
+    resp = await parser_client.post(
+        "/v1/chat/completions/derender",
+        json={
+            "model": PARSER_MODEL,
+            "generate_response": _e2e_generate_response(output_ids),
+            "prompt_tokens": len(gen_req["token_ids"]),
+        },
+    )
+    assert resp.status_code == 200
+    content = resp.json()["choices"][0]["message"]["content"]
+    re_encoded = _encode(parser_tokenizer, content)
+    assert output_ids == re_encoded
+
+
+@pytest.mark.asyncio
+async def test_e2e_non_ascii_roundtrip(parser_client, parser_tokenizer):
+    """CJK + emoji roundtrip without U+FFFD."""
+    messages = [{"role": "user", "content": "Reply in Chinese"}]
+    gen_req = await _e2e_render_chat(parser_client, PARSER_MODEL, messages)
+
+    answer = "你好世界 😀"
+    output_ids = _encode(parser_tokenizer, answer)
+
+    resp = await parser_client.post(
+        "/v1/chat/completions/derender",
+        json={
+            "model": PARSER_MODEL,
+            "generate_response": _e2e_generate_response(output_ids),
+            "prompt_tokens": len(gen_req["token_ids"]),
+        },
+    )
+    assert resp.status_code == 200
+    content = resp.json()["choices"][0]["message"]["content"]
+    assert "�" not in content
+
+
+@pytest.mark.asyncio
+async def test_e2e_parsed_reasoning(parser_client, parser_tokenizer):
+    """<think>...</think> splits into reasoning + content."""
+    messages = [{"role": "user", "content": "What is 2+3?"}]
+    gen_req = await _e2e_render_chat(parser_client, PARSER_MODEL, messages)
+
+    reasoning_text = "The user wants 2 plus 3. That is 5."
+    answer_text = "The answer is 5."
+    output_text = f"<think>{reasoning_text}</think>{answer_text}"
+    output_ids = _require_markers_survive(parser_tokenizer, output_text, "</think>")
+
+    resp = await parser_client.post(
+        "/v1/chat/completions/derender",
+        json={
+            "model": PARSER_MODEL,
+            "generate_response": _e2e_generate_response(output_ids),
+            "prompt_tokens": len(gen_req["token_ids"]),
+            "chat_request": {
+                "model": PARSER_MODEL,
+                "messages": messages,
+                "include_reasoning": True,
+            },
+        },
+    )
+    assert resp.status_code == 200, resp.text
+    msg = resp.json()["choices"][0]["message"]
+    assert msg["reasoning"] is not None
+    assert reasoning_text in msg["reasoning"]
+    assert answer_text in msg["content"]
+    assert "<think>" not in msg["content"]
+
+
+@pytest.mark.asyncio
+async def test_e2e_parsed_tool_call(parser_client, parser_tokenizer):
+    """<tool_call> extracted into tool_calls field."""
+    messages = [{"role": "user", "content": "Weather in Paris?"}]
+    gen_req = await _e2e_render_chat(parser_client, PARSER_MODEL, messages)
+
+    output_text = (
+        "<think>Let me check the weather.</think>"
+        '<tool_call>\n{"name": "get_weather", '
+        '"arguments": {"city": "Paris"}}\n</tool_call>'
+    )
+    output_ids = _require_markers_survive(
+        parser_tokenizer,
+        output_text,
+        "</think>",
+        "<tool_call>",
+        "</tool_call>",
+    )
+
+    resp = await parser_client.post(
+        "/v1/chat/completions/derender",
+        json={
+            "model": PARSER_MODEL,
+            "generate_response": _e2e_generate_response(output_ids),
+            "prompt_tokens": len(gen_req["token_ids"]),
+            "chat_request": {
+                "model": PARSER_MODEL,
+                "messages": messages,
+                "tools": _E2E_TOOLS,
+                "tool_choice": "auto",
+            },
+        },
+    )
+    assert resp.status_code == 200, resp.text
+    choice = resp.json()["choices"][0]
+    assert choice["message"]["tool_calls"]
+    assert choice["message"]["tool_calls"][0]["function"]["name"] == "get_weather"
+
+
+@pytest.mark.asyncio
+async def test_e2e_parsed_reasoning_and_tool_call(parser_client, parser_tokenizer):
+    """Reasoning + tool call in the same output."""
+    messages = [{"role": "user", "content": "Weather in Paris?"}]
+    gen_req = await _e2e_render_chat(parser_client, PARSER_MODEL, messages)
+
+    reasoning_text = "I should look up the weather."
+    tool_text = (
+        '<tool_call>\n{"name": "get_weather", '
+        '"arguments": {"city": "Paris"}}\n</tool_call>'
+    )
+    output_text = f"<think>{reasoning_text}</think>{tool_text}"
+    output_ids = _require_markers_survive(
+        parser_tokenizer, output_text, "</think>", "<tool_call>"
+    )
+
+    resp = await parser_client.post(
+        "/v1/chat/completions/derender",
+        json={
+            "model": PARSER_MODEL,
+            "generate_response": _e2e_generate_response(output_ids),
+            "prompt_tokens": len(gen_req["token_ids"]),
+            "chat_request": {
+                "model": PARSER_MODEL,
+                "messages": messages,
+                "tools": _E2E_TOOLS,
+                "tool_choice": "auto",
+                "include_reasoning": True,
+            },
+        },
+    )
+    assert resp.status_code == 200, resp.text
+    choice = resp.json()["choices"][0]
+    assert choice["message"]["reasoning"] is not None
+    assert reasoning_text in choice["message"]["reasoning"]
+    assert choice["message"]["tool_calls"]
+
+
+@pytest.mark.asyncio
+async def test_e2e_no_chat_request_fallback(parser_client, parser_tokenizer):
+    """Without chat_request, derender falls back to plain detokenization."""
+    messages = [{"role": "user", "content": "Hello"}]
+    gen_req = await _e2e_render_chat(parser_client, PARSER_MODEL, messages)
+
+    answer = "Hi there!"
+    output_ids = _encode(parser_tokenizer, answer)
+
+    resp = await parser_client.post(
+        "/v1/chat/completions/derender",
+        json={
+            "model": PARSER_MODEL,
+            "generate_response": _e2e_generate_response(output_ids),
+            "prompt_tokens": len(gen_req["token_ids"]),
+        },
+    )
+    assert resp.status_code == 200
+    content = resp.json()["choices"][0]["message"]["content"]
+    assert "Hi" in content
+
+
+# ---------------------------------------------------------------------------
+# E2E: HarmonyParser + GPT-OSS
+# ---------------------------------------------------------------------------
+
+HARMONY_MODEL = "openai/gpt-oss-20b"
+
+
+def _ensure_harmony_vocab():
+    """Pre-cache the o200k_base BPE file needed by openai-harmony.
+
+    The Rust tiktoken-rs backend downloads from Azure Blob Storage, which
+    may be unreachable in some environments.  When the cache is cold we
+    fetch the file ourselves and place it in ``/tmp/tiktoken-rs-cache/``
+    using the SHA-1(URL) filename that tiktoken-rs expects.
+    """
+    import hashlib
+    import urllib.request
+    from pathlib import Path
+
+    url = "https://openaipublic.blob.core.windows.net/encodings/o200k_base.tiktoken"
+    cache_dir = Path("/tmp/tiktoken-rs-cache")
+    cache_key = hashlib.sha1(url.encode()).hexdigest()
+    cache_file = cache_dir / cache_key
+    if not cache_file.exists():
+        cache_dir.mkdir(parents=True, exist_ok=True)
+        urllib.request.urlretrieve(url, cache_file)
+
+
+@pytest.fixture(scope="module")
+def harmony_server():
+    _ensure_harmony_vocab()
+    args = [
+        "--trust-remote-code",
+        "--enable-auto-tool-choice",
+        "--tool-call-parser",
+        "openai",
+        "--reasoning-parser",
+        "openai_gptoss",
+    ]
+    with RemoteLaunchRenderServer(HARMONY_MODEL, args) as remote_server:
+        yield remote_server
+
+
+@pytest_asyncio.fixture
+async def harmony_client(harmony_server):
+    async with httpx.AsyncClient(
+        base_url=harmony_server.url_for(""), timeout=60.0
+    ) as http_client:
+        yield http_client
+
+
+@pytest.fixture(scope="module")
+def harmony_tokenizer():
+    return get_tokenizer(HARMONY_MODEL, trust_remote_code=True)
+
+
+def _harmony_extract_assistant_ids(
+    tokenizer, assistant_msg: dict, user_content: str = "test"
+) -> list[int]:
+    """Extract assistant token IDs via apply_chat_template diff."""
+    prompt = [{"role": "user", "content": user_content}]
+    full = prompt + [assistant_msg]
+    text_prompt = tokenizer.apply_chat_template(
+        prompt, add_generation_prompt=True, tokenize=False
+    )
+    text_full = tokenizer.apply_chat_template(
+        full, add_generation_prompt=False, tokenize=False
+    )
+    prompt_ids = tokenizer.encode(text_prompt)
+    full_ids = tokenizer.encode(text_full)
+    assistant_ids = list(full_ids[len(prompt_ids) :])
+    if not assistant_ids:
+        pytest.skip("Could not extract assistant tokens for Harmony")
+    return assistant_ids
+
+
+@pytest.mark.asyncio
+async def test_e2e_harmony_plain_roundtrip(harmony_client, harmony_tokenizer):
+    """GPT-OSS content-only roundtrip."""
+    messages = [{"role": "user", "content": "What is 2+2?"}]
+    gen_req = await _e2e_render_chat(harmony_client, HARMONY_MODEL, messages)
+
+    assistant_msg = {"role": "assistant", "content": "Four."}
+    output_ids = _harmony_extract_assistant_ids(harmony_tokenizer, assistant_msg)
+
+    resp = await harmony_client.post(
+        "/v1/chat/completions/derender",
+        json={
+            "model": HARMONY_MODEL,
+            "generate_response": _e2e_generate_response(output_ids),
+            "prompt_tokens": len(gen_req["token_ids"]),
+            "chat_request": {
+                "model": HARMONY_MODEL,
+                "messages": messages,
+            },
+        },
+    )
+    assert resp.status_code == 200, resp.text
+    content = resp.json()["choices"][0]["message"]["content"]
+    assert content is not None and len(content) > 0
+    assert "Four" in content
+
+
+@pytest.mark.asyncio
+async def test_e2e_harmony_reasoning(harmony_client, harmony_tokenizer):
+    """GPT-OSS reasoning: analysis channel extracted."""
+    messages = [{"role": "user", "content": "Add 2 and 3."}]
+    gen_req = await _e2e_render_chat(harmony_client, HARMONY_MODEL, messages)
+
+    reasoning_text = "The user wants 2 plus 3."
+    answer_text = "The answer is 5."
+    assistant_msg = {
+        "role": "assistant",
+        "thinking": reasoning_text,
+        "content": answer_text,
+    }
+    output_ids = _harmony_extract_assistant_ids(harmony_tokenizer, assistant_msg)
+
+    decoded = harmony_tokenizer.decode(output_ids)
+    if reasoning_text not in decoded:
+        pytest.skip("Harmony template did not render thinking")
+
+    resp = await harmony_client.post(
+        "/v1/chat/completions/derender",
+        json={
+            "model": HARMONY_MODEL,
+            "generate_response": _e2e_generate_response(output_ids),
+            "prompt_tokens": len(gen_req["token_ids"]),
+            "chat_request": {
+                "model": HARMONY_MODEL,
+                "messages": messages,
+                "include_reasoning": True,
+            },
+        },
+    )
+    assert resp.status_code == 200, resp.text
+    msg = resp.json()["choices"][0]["message"]
+    assert msg["reasoning"] is not None
+    assert reasoning_text in msg["reasoning"]
+    assert answer_text in (msg["content"] or "")

tests/entrypoints/serve/tokenize/test_serving_tokenization.py

@@ -15,7 +15,7 @@ from vllm.entrypoints.serve.tokenize.protocol import (
     TokenizeChatRequest,
     TokenizeCompletionRequest,
 )
-from vllm.entrypoints.serve.tokenize.serving import OpenAIServingTokenization
+from vllm.entrypoints.serve.tokenize.serving import ServingTokenization
 from vllm.v1.engine.async_llm import AsyncLLM
 
 MODEL_NAME = "openai-community/gpt2"
@@ -58,7 +58,7 @@ class MockModelConfig:
         return self.diff_sampling_param or {}
 
 
-def _build_serving_tokenization(engine: AsyncLLM) -> OpenAIServingTokenization:
+def _build_serving_tokenization(engine: AsyncLLM) -> ServingTokenization:
     models = OpenAIServingModels(
         engine_client=engine,
         base_model_paths=BASE_MODEL_PATHS,
@@ -71,8 +71,7 @@ def _build_serving_tokenization(engine: AsyncLLM) -> OpenAIServingTokenization:
         chat_template=None,
         chat_template_content_format="auto",
     )
-    return OpenAIServingTokenization(
-        engine,
+    return ServingTokenization(
         models,
         openai_serving_render=serving_render,
         request_logger=None,

tests/kernels/attention/test_attention_selector.py

@@ -15,16 +15,14 @@ from vllm.config import (
 from vllm.platforms import current_platform
 from vllm.platforms.cpu import CpuPlatform
 
-# CudaPlatform and RocmPlatform import their respective compiled C extensions
-# at module level, raising ModuleNotFoundError on incompatible builds.
-try:
+if current_platform.is_cuda():
     from vllm.platforms.cuda import CudaPlatform
-except (ImportError, ModuleNotFoundError):
+else:
     CudaPlatform = None
 
-try:
+if current_platform.is_rocm():
     from vllm.platforms.rocm import RocmPlatform
-except (ImportError, ModuleNotFoundError):
+else:
     RocmPlatform = None
 
 from vllm.v1.attention.backends.registry import AttentionBackendEnum
@@ -434,9 +432,15 @@ def test_per_head_quant_scales_backend_selection(
     [
         ("FLASH_ATTN", True, True),  # FlashAttn supports non-causal
         ("FLASH_ATTN", False, True),  # FlashAttn also works with causal
-        ("FLASHINFER", True, False),  # FlashInfer does not support non-causal
-        ("FLASHINFER", False, True),  # FlashInfer works with causal
-    ],
+    ]
+    + (
+        [
+            ("FLASHINFER", True, True),  # FlashInfer supports non-causal
+            ("FLASHINFER", False, True),  # FlashInfer works with causal
+        ]
+        if CudaPlatform is not None
+        else []
+    ),
 )
 def test_non_causal_backend_selection(
     backend_name: str, use_non_causal: bool, should_succeed: bool
@@ -459,11 +463,12 @@ def test_non_causal_backend_selection(
         attention_config=attention_config, cache_config=cache_config
     )
 
-    if CudaPlatform is None:
-        pytest.skip("CudaPlatform not available")
+    platform = CudaPlatform or RocmPlatform
+    if platform is None:
+        pytest.skip("CudaPlatform and RocmPlatform are not available")
     with (
         set_current_vllm_config(vllm_config),
-        patch("vllm.platforms.current_platform", CudaPlatform()),
+        patch("vllm.platforms.current_platform", platform()),
     ):
         if should_succeed:
             backend = get_attn_backend(

tests/kernels/attention/test_prefix_prefill.py

@@ -5,10 +5,12 @@ import math
 import random
 import time
 from collections.abc import Callable
+from contextlib import nullcontext
 
 import pytest
 import torch
 import torch.nn.functional as F
+from torch.nn.attention import SDPBackend, sdpa_kernel
 
 from vllm.platforms import current_platform
 from vllm.utils.torch_utils import STR_DTYPE_TO_TORCH_DTYPE, set_random_seed
@@ -557,15 +559,21 @@ def test_contexted_kv_attention_alibi(
             query_len, seq_len, alibi_slopes, device, dtype
         )
 
-        # Compute attention
-        out = F.scaled_dot_product_attention(
-            q_sdpa,
-            k_sdpa,
-            v_sdpa,
-            attn_mask=alibi_mask,
-            dropout_p=0.0,
-            scale=scale,
-        )
+        # Compute attention. On ROCm we force use of the Math SDPA backend rather than
+        # the Flash or Mem-Efficient backends for increased numerical accuracy
+        if current_platform.is_rocm():
+            sdpa_context = sdpa_kernel(SDPBackend.MATH)
+        else:
+            sdpa_context = nullcontext()
+        with sdpa_context:
+            out = F.scaled_dot_product_attention(
+                q_sdpa,
+                k_sdpa,
+                v_sdpa,
+                attn_mask=alibi_mask,
+                dropout_p=0.0,
+                scale=scale,
+            )
 
         # Reshape output back to [query_len, num_heads, head_size]
         out = out.view(num_heads, query_len, head_size).permute(1, 0, 2)

tests/kernels/attention/test_rocm_triton_attn_dsv4.py

@@ -90,7 +90,9 @@ def _ref_sparse_prefill_ragged(
     return out.to(torch.bfloat16)
 
 
-def _pack_fp8_ds_mla_cache(kv: torch.Tensor, block_size: int) -> torch.Tensor:
+def _pack_fp8_ds_mla_cache(
+    kv: torch.Tensor, block_size: int, is_extra: bool = False
+) -> torch.Tensor:
     assert kv.shape[-1] == HEAD_DIM
     num_tokens = kv.shape[0]
     num_blocks = (num_tokens + block_size - 1) // block_size
@@ -101,7 +103,9 @@ def _pack_fp8_ds_mla_cache(kv: torch.Tensor, block_size: int) -> torch.Tensor:
     )
     cache_flat = cache.view(torch.uint8).flatten()
     kv_nope_fp8 = (
-        kv[:, :NOPE_HEAD_DIM].to(current_platform.fp8_dtype()).view(torch.uint8)
+        kv[:, :NOPE_HEAD_DIM]
+        .to(torch.float8_e4m3fn if is_extra else current_platform.fp8_dtype())
+        .view(torch.uint8)
     )
     kv_rope_u8 = kv[:, NOPE_HEAD_DIM:].contiguous().view(torch.uint8)
 
@@ -120,7 +124,7 @@ def _pack_fp8_ds_mla_cache(kv: torch.Tensor, block_size: int) -> torch.Tensor:
 
 
 def _read_fp8_ds_mla_cache(
-    cache: torch.Tensor, slot: int, block_size: int
+    cache: torch.Tensor, slot: int, block_size: int, is_extra: bool = False
 ) -> torch.Tensor:
     cache_flat = cache.view(torch.uint8).flatten()
     block_idx = slot // block_size
@@ -129,7 +133,9 @@ def _read_fp8_ds_mla_cache(
     token_base = block_base + pos * 576
 
     nope_u8 = cache_flat[token_base : token_base + NOPE_HEAD_DIM]
-    nope = nope_u8.view(current_platform.fp8_dtype()).to(torch.float32)
+    nope = nope_u8.view(
+        torch.float8_e4m3fn if is_extra else current_platform.fp8_dtype()
+    ).to(torch.float32)
     rope_u8 = cache_flat[
         token_base + NOPE_HEAD_DIM : token_base + NOPE_HEAD_DIM + ROPE_HEAD_DIM * 2
     ]
@@ -157,7 +163,9 @@ def _ref_sparse_decode_ragged(
         ]
         if extra_cache is not None and extra_rows is not None:
             row_kv.extend(
-                _read_fp8_ds_mla_cache(extra_cache, int(slot), block_size)
+                _read_fp8_ds_mla_cache(
+                    extra_cache, int(slot), block_size, is_extra=True
+                )
                 for slot in extra_rows[query_idx]
             )
 
@@ -326,7 +334,7 @@ def test_sparse_attn_decode_ragged_kernel() -> None:
     main_kv = torch.randn(6, HEAD_DIM, dtype=torch.bfloat16, device=device) * 0.125
     extra_kv = torch.randn(5, HEAD_DIM, dtype=torch.bfloat16, device=device) * 0.125
     main_cache = _pack_fp8_ds_mla_cache(main_kv, block_size)
-    extra_cache = _pack_fp8_ds_mla_cache(extra_kv, block_size)
+    extra_cache = _pack_fp8_ds_mla_cache(extra_kv, block_size, is_extra=True)
     main_indices = torch.tensor([0, 2, 4, 1], dtype=torch.int32, device=device)
     main_indptr = torch.tensor([0, 2, 4], dtype=torch.int32, device=device)
     extra_indices = torch.tensor([1, 3, 0], dtype=torch.int32, device=device)
@@ -477,7 +485,7 @@ def test_sparse_attn_decode_split_k_kernel(
         rows = [[1, 3, 0, 5, 2, 4], [3, 0, 6]]
         extra_kv = torch.randn(7, HEAD_DIM, dtype=torch.bfloat16, device=device) * 0.125
         extra_rows = rows
-        extra_cache = _pack_fp8_ds_mla_cache(extra_kv, block_size)
+        extra_cache = _pack_fp8_ds_mla_cache(extra_kv, block_size, is_extra=True)
         extra_indices, extra_indptr = _ragged_from_rows(rows, device)
 
     attn_sink = (

tests/kernels/attention/test_triton_unified_attention.py

@@ -18,11 +18,7 @@ HEAD_SIZES = [128, 256]
 BLOCK_SIZES = [16]
 
 DTYPES = [torch.bfloat16]
-QDTYPES = (
-    [None, torch.float8_e4m3fn]
-    if not current_platform.is_rocm()
-    else [None, torch.float8_e4m3fnuz]
-)
+QDTYPES = [None, current_platform.fp8_dtype()]
 FP8_DTYPE = current_platform.fp8_dtype()
 
 # one value large enough to test overflow in index calculation.

tests/kernels/core/test_minimax_reduce_rms.py

@@ -10,8 +10,12 @@ from torch.multiprocessing import spawn
 from tests.kernels.utils import opcheck
 from tests.utils import ensure_current_vllm_config, init_test_distributed_environment
 from vllm.distributed import cleanup_dist_env_and_memory
-from vllm.model_executor.layers.minimax_rms_norm import MiniMaxText01RMSNormTP
+from vllm.model_executor.layers.minimax_rms_norm import (
+    MiniMaxText01RMSNormTP,
+    rms_norm_tp,
+)
 from vllm.platforms import current_platform
+from vllm.triton_utils import HAS_TRITON
 from vllm.utils.network_utils import get_open_port
 from vllm.utils.torch_utils import set_random_seed
 
@@ -54,8 +58,19 @@ def _worker_forward_qk(
     torch.manual_seed(seed + 1000 + local_rank)
     qkv = torch.randn(num_tokens, hq + hk + hk, dtype=dtype, device="cuda")
 
-    q_ref, k_ref, v_ref = qkv.clone().split([hq, hk, hk], dim=-1)
-    ref_q, ref_k = MiniMaxText01RMSNormTP.forward_qk(q_norm, k_norm, q_ref, k_ref)
+    # Reference: eager all-reduce path. ``forward_qk`` no longer all-reduces
+    # the variance (it is the tp==1 / already-reduced building block), so the
+    # multi-rank reference must use the eager path that performs the global
+    # variance all-reduce, matching the fused kernel below.
+    ref_q, ref_k = rms_norm_tp._minimax_qk_norm_tp_eager(
+        qkv.clone(),
+        q_norm.weight,
+        k_norm.weight,
+        hq,
+        hk,
+        world_size,
+        eps,
+    )
 
     # Set up Lamport workspace.
     from vllm.distributed.parallel_state import get_tp_group
@@ -150,3 +165,44 @@ def test_minimax_reduce_rms_qk(
         nprocs=world_size,
         join=True,
     )
+
+
+@pytest.mark.skipif(
+    not current_platform.is_cuda() or not HAS_TRITON,
+    reason="CUDA and Triton required",
+)
+@pytest.mark.parametrize("num_tokens", [1, 7, 128, 333, 2049])
+@pytest.mark.parametrize("hidden_dims", [(3072, 512), (768, 256), (3000, 500)])
+@pytest.mark.parametrize("dtype", [torch.bfloat16, torch.float16])
+@pytest.mark.parametrize("tp_world", [1, 4, 8])
+@pytest.mark.parametrize("eps", [1e-6])
+@pytest.mark.parametrize("seed", [42])
+def test_minimax_qk_norm_triton_fallback(
+    monkeypatch, num_tokens, hidden_dims, dtype, tp_world, eps, seed
+):
+    """Single-GPU check: Triton fallback kernels vs the pure-torch reference.
+
+    The all-reduce is a TP communication barrier, so it is monkeypatched to
+    identity here; both the Triton path and the reference see the same
+    (patched) reduction. This validates the kernel math and the folded
+    ``/ tp_world`` scaling without needing multiple ranks -- ``hidden_dims``
+    are the per-rank q/k segment widths.
+    """
+    monkeypatch.setattr(rms_norm_tp, "_all_reduce_variance", lambda v: v)
+
+    q_size, kv_size = hidden_dims
+    device = "cuda"
+    torch.manual_seed(seed)
+    qkv = torch.randn(num_tokens, q_size + 2 * kv_size, dtype=dtype, device=device)
+    q_weight = torch.randn(q_size, dtype=dtype, device=device)
+    k_weight = torch.randn(kv_size, dtype=dtype, device=device)
+
+    q_triton, k_triton = rms_norm_tp._minimax_qk_norm_tp_fallback(
+        qkv, q_weight, k_weight, q_size, kv_size, 0, tp_world, eps
+    )
+    q_ref, k_ref = rms_norm_tp._minimax_qk_norm_tp_eager(
+        qkv, q_weight, k_weight, q_size, kv_size, tp_world, eps
+    )
+
+    torch.testing.assert_close(q_triton, q_ref, atol=3e-2, rtol=3e-2)
+    torch.testing.assert_close(k_triton, k_ref, atol=3e-2, rtol=3e-2)

tests/kernels/moe/test_ocp_mx_moe.py

@@ -9,6 +9,7 @@ import pytest
 import torch
 from packaging import version
 
+from vllm._aiter_ops import is_aiter_found
 from vllm.platforms import current_platform
 from vllm.utils.flashinfer import has_flashinfer
 
@@ -31,17 +32,15 @@ HOPPER_MXFP4_BF16_AVAILABLE = (
 # ROCm platform and dependencies
 ROCM_AVAILABLE = current_platform.is_rocm()
 ROCM_TRITON_KERNELS_AVAILABLE = False
-ROCM_AITER_AVAILABLE = False
+ROCM_AITER_AVAILABLE = is_aiter_found()
 ROCM_GFX950 = False
 
 if ROCM_AVAILABLE:
-    from vllm._aiter_ops import rocm_aiter_ops
     from vllm.platforms.rocm import on_gfx950
     from vllm.utils.import_utils import has_triton_kernels
 
     ROCM_TRITON_KERNELS_AVAILABLE = has_triton_kernels()
     ROCM_GFX950 = on_gfx950()
-    ROCM_AITER_AVAILABLE = rocm_aiter_ops.is_enabled()
 
     if ROCM_AITER_AVAILABLE:
         from aiter.ops.triton.moe.quant_moe import upcast_from_mxfp
@@ -83,7 +82,7 @@ def enable_pickle(monkeypatch):
     [
         ModelCase("fxmarty/qwen_1.5-moe-a2.7b-mxfp4", tp=2),
         ModelCase("fxmarty/deepseek_r1_3_layers_mxfp4", tp=8),
-        ModelCase("fxmarty/Llama-4-Scout-17B-16E-Instruct-2-layers-mxfp4", tp=1),
+        ModelCase("mawong-amd/Llama-4-Scout-17B-16E-Instruct-2-layers-mxfp4", tp=1),
         ModelCase("fxmarty/Llama-3.1-70B-Instruct-2-layers-mxfp6", tp=1),
         ModelCase("fxmarty/Llama-3.1-70B-Instruct-2-layers-mxfp6", tp=4),
     ],
@@ -102,6 +101,7 @@ def test_mxfp4_loading_and_execution_moe(vllm_runner, model_case: ModelCase):
         tensor_parallel_size=model_case.tp,
         load_format="dummy",
         compilation_config={"cudagraph_capture_sizes": [16]},
+        gpu_memory_utilization=0.8,  # mxfp6 models use more scratch space
     ) as llm:
         # Disabled as check_model is broken: https://github.com/vllm-project/vllm/pull/18465#issuecomment-3329880562
         # def check_model(model):
@@ -1267,7 +1267,7 @@ def test_rocm_mxfp4_moe_oracle(
 
     This test validates that the oracle functions work end-to-end:
     - select_mxfp4_moe_backend() selects a valid backend
-    - convert_to_mxfp4_moe_kernel_format() converts weights without error
+    - convert_gpt_oss_weight_to_mxfp4_moe_kernel_format() converts weights without error
     - make_mxfp4_moe_quant_config() builds a valid quant config
     - make_mxfp4_moe_kernel() creates a kernel that runs without error
     - The kernel output is within accuracy tolerance of reference
@@ -1287,7 +1287,7 @@ def test_rocm_mxfp4_moe_oracle(
     from vllm.model_executor.layers.fused_moe.oracle.mxfp4 import (
         Mxfp4MoeBackend,
         backend_to_kernel_cls,
-        convert_to_mxfp4_moe_kernel_format,
+        convert_gpt_oss_weight_to_mxfp4_moe_kernel_format,
         make_mxfp4_moe_kernel,
         make_mxfp4_moe_quant_config,
     )
@@ -1387,7 +1387,7 @@ def test_rocm_mxfp4_moe_oracle(
 
     # Convert weights using oracle
     w13_conv, w2_conv, w13_scale_conv, w2_scale_conv, w13_bias_conv, w2_bias_conv = (
-        convert_to_mxfp4_moe_kernel_format(
+        convert_gpt_oss_weight_to_mxfp4_moe_kernel_format(
             mxfp4_backend=backend,
             layer=layer,  # type: ignore[arg-type]
             w13_weight=w13_quant,
@@ -1423,7 +1423,7 @@ def test_rocm_mxfp4_moe_oracle(
             mxfp4_backend=backend,
             experts_cls=experts_cls,
             routing_tables=None,
-            shared_experts=None,
+            layer=None,
         )
 
         # Create inputs

tests/kernels/test_mhc_kernels.py

@@ -8,8 +8,8 @@ from vllm.model_executor.kernels.mhc.tilelang import (
     _tilelang_hc_prenorm_gemm,
     _torch_hc_prenorm_gemm,
 )
+from vllm.model_executor.layers.mhc import HAS_TILELANG_MHC
 from vllm.platforms import current_platform
-from vllm.utils.import_utils import has_tilelang
 from vllm.utils.torch_utils import set_random_seed
 
 DEVICE = current_platform.device_type
@@ -97,8 +97,8 @@ def hc_head_ref(
 
 
 @pytest.mark.skipif(
-    not (current_platform.is_cuda_alike() and has_tilelang()),
-    reason="CUDA or ROCm and tilelang required",
+    not HAS_TILELANG_MHC,
+    reason="TileLang MHC support required",
 )
 @pytest.mark.parametrize("num_tokens", [1, 4, 8, 128])
 @pytest.mark.parametrize("hidden_size", [4096, 7168])
@@ -150,8 +150,8 @@ def test_mhc_pre_tilelang(num_tokens, hidden_size, hc_mult):
 
 
 @pytest.mark.skipif(
-    not (current_platform.is_cuda_alike() and has_tilelang()),
-    reason="CUDA or ROCm and tilelang required",
+    not HAS_TILELANG_MHC,
+    reason="TileLang MHC support required",
 )
 @pytest.mark.parametrize(
     ("num_tokens", "hidden_size"),
@@ -190,8 +190,8 @@ def test_hc_prenorm_gemm_tilelang(num_tokens, hidden_size):
 
 
 @pytest.mark.skipif(
-    not (current_platform.is_cuda_alike() and has_tilelang()),
-    reason="CUDA or ROCm and tilelang required",
+    not HAS_TILELANG_MHC,
+    reason="TileLang MHC support required",
 )
 @pytest.mark.parametrize("num_tokens", [1, 4, 8, 128])
 @pytest.mark.parametrize("hidden_size", [4096, 7168])
@@ -217,8 +217,8 @@ def test_mhc_post_tilelang(num_tokens, hidden_size, hc_mult):
 
 
 @pytest.mark.skipif(
-    not (current_platform.is_cuda_alike() and has_tilelang()),
-    reason="CUDA or ROCm and tilelang required",
+    not HAS_TILELANG_MHC,
+    reason="TileLang MHC support required",
 )
 @pytest.mark.parametrize("num_tokens", [1, 4, 8, 128])
 @pytest.mark.parametrize("hidden_size", [4096, 7168])
@@ -324,8 +324,8 @@ def test_hc_head_triton(num_tokens, hidden_size, hc_mult):
 
 
 @pytest.mark.skipif(
-    not (current_platform.is_cuda_alike() and has_tilelang()),
-    reason="CUDA or ROCm and tilelang required",
+    not HAS_TILELANG_MHC,
+    reason="TileLang MHC support required",
 )
 @pytest.mark.parametrize("num_tokens", [1, 4, 8, 128])
 @pytest.mark.parametrize("hidden_size", [4096, 7168])

tests/models/multimodal/generation/test_common.py

@@ -810,29 +810,6 @@ VLM_TEST_SETTINGS = {
         hf_output_post_proc=model_utils.minicpmv_trunc_hf_output,
         patch_hf_runner=model_utils.minicpmv_26_patch_hf_runner,
     ),
-    "minimax_vl_01": VLMTestInfo(
-        models=["MiniMaxAI/MiniMax-VL-01"],
-        prompt_formatter=lambda img_prompt: f"<beginning_of_sentence>user: {img_prompt} assistant:<end_of_sentence>",  # noqa: E501
-        img_idx_to_prompt=lambda _: "<image>",
-        test_type=(VLMTestType.IMAGE, VLMTestType.MULTI_IMAGE),
-        max_model_len=8192,
-        max_num_seqs=4,
-        dtype="bfloat16",
-        hf_output_post_proc=model_utils.minimax_vl_01_hf_output,
-        patch_hf_runner=model_utils.minimax_vl_01_patch_hf_runner,
-        auto_cls=AutoModelForImageTextToText,
-        marks=[
-            large_gpu_mark(min_gb=80),
-            # TODO: [ROCm] Fix pickle issue with ROCm spawn and tp>1
-            pytest.mark.skipif(
-                current_platform.is_rocm(),
-                reason=(
-                    "ROCm: Model too large for single GPU; "
-                    "multi-GPU blocked by HF _LazyConfigMapping pickle issue with spawn"
-                ),
-            ),
-        ],
-    ),
     "molmo": VLMTestInfo(
         models=["allenai/Molmo-7B-D-0924"],
         test_type=(VLMTestType.IMAGE, VLMTestType.MULTI_IMAGE),

tests/models/multimodal/generation/vlm_utils/model_utils.py

@@ -245,13 +245,6 @@ def minicpmv_trunc_hf_output(hf_output: RunnerOutput, model: str) -> RunnerOutpu
     return output_ids, output_str, out_logprobs
 
 
-def minimax_vl_01_hf_output(hf_output: RunnerOutput, model: str) -> RunnerOutput:
-    output_ids, output_str, out_logprobs = hf_output
-    if output_str.endswith("<end_of_sentence>"):
-        output_str = output_str.split("<end_of_sentence>")[0]
-    return output_ids, output_str, out_logprobs
-
-
 def ultravox_trunc_hf_output(hf_output: RunnerOutput, model: str) -> RunnerOutput:
     output_ids, output_str, out_logprobs = hf_output
 
@@ -1023,17 +1016,6 @@ def minicpmv_26_patch_hf_runner(hf_model: HfRunner) -> HfRunner:
     return hf_model
 
 
-def minimax_vl_01_patch_hf_runner(hf_model: HfRunner) -> HfRunner:
-    orig_generate = hf_model.model.generate
-
-    def _generate(self, *args, image_sizes=None, **kwargs):
-        return orig_generate(*args, decode_text=False, **kwargs)
-
-    hf_model.model.generate = types.MethodType(_generate, hf_model.model)
-
-    return hf_model
-
-
 def molmo_patch_hf_runner(hf_model: HfRunner) -> HfRunner:
     """Patches and returns an instance of the HfRunner to use for Molmo."""
     hf_processor = hf_model.processor

tests/models/multimodal/pooling/test_colqwen3_5.py

@@ -152,3 +152,21 @@ def test_colqwen3_5_relevance_ordering(
     dtype: str,
 ) -> None:
     _run_relevance_test(vllm_runner, model, dtype=dtype)
+
+
+def test_colqwen3_5_config_enables_bidirectional_attention() -> None:
+    """ColQwen3.5 retrieval must be served BIDIRECTIONAL (is_causal=False) so the
+    full_attention layers build with AttentionType.ENCODER_ONLY. This guards the
+    silent-causal regression (no GPU / model load needed)."""
+    from types import SimpleNamespace
+
+    from vllm.model_executor.models.config import (
+        MODELS_CONFIG_MAP,
+        ColQwen3_5Config,
+    )
+
+    assert MODELS_CONFIG_MAP["ColQwen3_5"] is ColQwen3_5Config
+
+    model_config = SimpleNamespace(hf_config=SimpleNamespace())
+    ColQwen3_5Config.verify_and_update_model_config(model_config)
+    assert model_config.hf_config.is_causal is False

tests/models/multimodal/processing/test_minimax_vl_01.py

@@ -1,113 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-
-import pytest
-from PIL import Image
-
-from vllm.multimodal import MULTIMODAL_REGISTRY
-from vllm.multimodal.parse import ImageSize
-from vllm.multimodal.processing import BaseMultiModalProcessor
-
-from ....conftest import ImageTestAssets
-from ...utils import build_model_context
-
-
-@pytest.mark.parametrize("model_id", ["MiniMaxAI/MiniMax-VL-01"])
-@pytest.mark.parametrize("num_imgs", [1, 2])
-def test_processor_override(
-    image_assets: ImageTestAssets,
-    model_id: str,
-    num_imgs: int,
-):
-    ctx = build_model_context(
-        model_id,
-        mm_processor_kwargs=None,
-        limit_mm_per_prompt={"image": num_imgs},
-    )
-    processor = MULTIMODAL_REGISTRY.create_processor(ctx.model_config)
-    prompt = "<image>" * num_imgs
-    image = Image.new("RGB", size=(364, 364))
-    mm_data = {"image": [image] * num_imgs}
-
-    processed_inputs = processor(
-        prompt,
-        mm_items=processor.info.parse_mm_data(mm_data),
-        hf_processor_mm_kwargs={},
-    )
-    image_placeholders = processed_inputs["mm_placeholders"]["image"]
-
-    assert len(image_placeholders) == num_imgs
-
-
-def _validate_image_prompt_replacements_one(
-    processor: BaseMultiModalProcessor,
-    num_imgs: int,
-    failed_size_excs: list[tuple[ImageSize, Exception]],
-    image_size: ImageSize,
-) -> None:
-    prompt = "<image>" * num_imgs
-    image = Image.new("RGB", size=image_size)
-    mm_data = {"image": [image] * num_imgs}
-
-    try:
-        processed_inputs = processor(
-            prompt,
-            mm_items=processor.info.parse_mm_data(mm_data),
-            hf_processor_mm_kwargs={},
-        )
-
-        image_placeholders = processed_inputs["mm_placeholders"]["image"]
-        assert len(image_placeholders) == num_imgs
-
-    except Exception as exc:
-        failed_size_excs.append((image_size, exc))
-
-
-def _test_image_prompt_replacements(
-    processor,
-    *,
-    num_imgs: int,
-    image_sizes: list[ImageSize],
-) -> None:
-    failed_size_excs = list[tuple[ImageSize, Exception]]()
-
-    for size in image_sizes:
-        _validate_image_prompt_replacements_one(
-            processor, num_imgs, failed_size_excs, size
-        )
-
-    if failed_size_excs:
-        msg = "Found failing image sizes:" + "\n========\n".join(
-            f"[{size}]\n{exc}" for size, exc in failed_size_excs
-        )
-        raise AssertionError(msg)
-
-
-@pytest.mark.parametrize("model_id", ["MiniMaxAI/MiniMax-VL-01"])
-@pytest.mark.parametrize("num_imgs", [1, 2])
-def test_processor_prompt_replacements_regression(model_id, num_imgs):
-    ctx = build_model_context(
-        model_id,
-        mm_processor_kwargs=None,
-        limit_mm_per_prompt={"image": num_imgs},
-    )
-    processor = MULTIMODAL_REGISTRY.create_processor(ctx.model_config)
-
-    image_ratios = [
-        (171, 152),
-        (184, 161),
-        (198, 176),
-        (333, 296),
-        (369, 328),
-        (488, 183),
-        (2560, 1669),
-    ]
-    image_sizes = [
-        size for w, h in image_ratios for size in [ImageSize(w, h), ImageSize(h, w)]
-    ]
-
-    _test_image_prompt_replacements(
-        processor,
-        num_imgs=num_imgs,
-        image_sizes=image_sizes,
-    )

tests/models/multimodal/processing/test_qwen3_vl.py

@@ -138,3 +138,49 @@ def test_processor_multi_video(
         assert video_phs[i].offset >= prev_end, (
             f"Placeholder {i} overlaps with placeholder {i - 1}"
         )
+
+
+@pytest.mark.parametrize("model_id", [MODEL_ID])
+@pytest.mark.parametrize(
+    "hf_mm_kwargs",
+    [{"num_frames": [8, 16]}, {"fps": [2.0, 4.0]}],
+)
+def test_processor_multi_video_list_kwargs(
+    model_id: str,
+    hf_mm_kwargs: dict[str, Any],
+) -> None:
+    """Regression test: a multi-video request with list-valued per-video
+    ``mm_processor_kwargs`` (one ``fps``/``num_frames`` per video) must not
+    crash.
+
+    Before the fix, ``_call_hf_processor`` copied the whole kwargs to every
+    video without slicing, so ``_get_video_second_idx`` received the list
+    where a scalar was expected and raised ``TypeError``.
+    """
+    ctx = build_model_context(
+        model_id,
+        limit_mm_per_prompt={"image": 0, "video": 2},
+    )
+    processor = MULTIMODAL_REGISTRY.create_processor(ctx.model_config)
+
+    prompt = (
+        "<|vision_start|><|video_pad|><|vision_end|>"
+        "<|vision_start|><|video_pad|><|vision_end|>"
+    )
+    mm_data = {
+        "video": [
+            _build_video_mm_data(num_frames=16)["video"][0],
+            _build_video_mm_data(num_frames=32)["video"][0],
+        ]
+    }
+
+    processed = processor(
+        prompt,
+        mm_items=processor.info.parse_mm_data(mm_data),
+        hf_processor_mm_kwargs=hf_mm_kwargs,
+    )
+
+    video_phs = processed["mm_placeholders"].get("video", [])
+    assert len(video_phs) == 2, (
+        f"Expected exactly 2 video placeholders, got {len(video_phs)}"
+    )

tests/models/registry.py

@@ -421,15 +421,6 @@ _TEXT_GENERATION_EXAMPLE_MODELS = {
         },
         trust_remote_code=True,
     ),
-    "MiniMaxForCausalLM": _HfExamplesInfo("MiniMaxAI/MiniMax-Text-01-hf"),
-    "MiniMaxText01ForCausalLM": _HfExamplesInfo(
-        "MiniMaxAI/MiniMax-Text-01",
-        trust_remote_code=True,
-        revision="a59aa9cbc53b9fb8742ca4e9e1531b9802b6fdc3",
-    ),
-    "MiniMaxM1ForCausalLM": _HfExamplesInfo(
-        "MiniMaxAI/MiniMax-M1-40k", trust_remote_code=True
-    ),
     "MiniMaxM2ForCausalLM": _HfExamplesInfo(
         "MiniMaxAI/MiniMax-M2",
         trust_remote_code=True,
@@ -1113,10 +1104,6 @@ _MULTIMODAL_EXAMPLE_MODELS = {
         "openbmb/MiniCPM-V-4_6",
         min_transformers_version="5.7.0",
     ),
-    "MiniMaxVL01ForConditionalGeneration": _HfExamplesInfo(
-        "MiniMaxAI/MiniMax-VL-01",
-        trust_remote_code=True,
-    ),
     "MiniMaxM3SparseForConditionalGeneration": _HfExamplesInfo(
         "MiniMaxAI/MiniMax-M3",
         trust_remote_code=True,
@@ -1530,6 +1517,16 @@ _SPECULATIVE_DECODING_EXAMPLE_MODELS = {
         "Qwen/Qwen3-VL-8B-Instruct",
         speculative_model="taobao-mnn/Qwen3-VL-8B-Instruct-Eagle3",
     ),
+    "Eagle3Qwen3ForCausalLM": _HfExamplesInfo(
+        "Qwen/Qwen3-8B",
+        trust_remote_code=True,
+        speculative_model=(
+            "inference-optimization/"
+            "Qwen3-8B-from-Qwen3-8B_regen-speculators.eagle3-qwen3arch-ckpt1"
+        ),
+        tokenizer="Qwen/Qwen3-8B",
+        use_original_num_layers=True,
+    ),
     # [PEagle]
     "PEagleDraftModel": _HfExamplesInfo(
         "Qwen/Qwen3-8B",
@@ -1545,6 +1542,15 @@ _SPECULATIVE_DECODING_EXAMPLE_MODELS = {
         tokenizer="Qwen/Qwen3-8B",
         use_original_num_layers=True,
     ),
+    "PeagleQwen3ForCausalLM": _HfExamplesInfo(
+        "Qwen/Qwen3-8B",
+        trust_remote_code=True,
+        speculative_model=(
+            "inference-optimization/Qwen3-8B-speculators.peagle-qwen3arch-ckpt4"
+        ),
+        tokenizer="Qwen/Qwen3-8B",
+        use_original_num_layers=True,
+    ),
     # [MTP]
     "DeepSeekMTPModel": _HfExamplesInfo(
         "luccafong/deepseek_mtp_main_random",

tests/models/test_initialization.py

@@ -98,11 +98,6 @@ def can_initialize(
         vllm_config.validate_block_size()
         return scheduler_kv_cache_config
 
-    if model_arch == "MiniMaxVL01ForConditionalGeneration":
-        pytest.skip(
-            "pickle error when loading `transformers.models.auto.CONFIG_MAPPING`"
-        )
-
     if model_arch == "MoonshotKimiaForCausalLM":
         pytest.skip(
             "Kimi-Audio requires SpeechToTextConfig "

tests/models/utils.py

@@ -507,7 +507,13 @@ def dummy_hf_overrides(
     # Only set MoE related config when the model has MoE layers.
     # Otherwise all models detected as MoE by _get_transformers_backend_cls.
     if model_arch_config.num_experts > 0:
-        num_experts_per_tok = 1 if model_arch == "Llama4ForConditionalGeneration" else 2
+        num_experts_per_tok = 2
+        if model_arch in (
+            "Llama4ForConditionalGeneration",
+            "Llama4ForCausalLM",
+            "EagleLlama4ForCausalLM",
+        ):
+            num_experts_per_tok = 1
         update_dict.update(
             {
                 "num_experts": num_experts,

tests/multimodal/test_video.py

@@ -15,6 +15,7 @@ from vllm.multimodal.video import (
     DynamicVideoBackend,
     GLM46VVideoBackend,
     Molmo2VideoBackend,
+    Qwen2VLVideoBackend,
     Qwen3VLVideoBackend,
     VideoLoader,
     VideoSourceMetadata,
@@ -70,11 +71,12 @@ def test_video_loader_type_doesnt_exist():
 
 
 @pytest.mark.parametrize(
-    "model_repo, expected_loader_cls",
+    "model_repo, expected_loader_cls, hf_sample_kwargs",
     [
         pytest.param(
             "allenai/Molmo2-4B",
             Molmo2VideoBackend,
+            None,
             marks=pytest.mark.skip(
                 reason="Video processor not aligned, investigate later.",
             ),
@@ -83,23 +85,44 @@ def test_video_loader_type_doesnt_exist():
         pytest.param(
             "zai-org/GLM-4.1V-9B-Thinking",
             DynamicVideoBackend,
+            None,
             id="glm4v",
         ),
         pytest.param(
             "zai-org/GLM-4.6V-Flash",
             GLM46VVideoBackend,
+            None,
             id="glm46v",
         ),
         pytest.param(
             "Qwen/Qwen3-VL-4B-Instruct",
             Qwen3VLVideoBackend,
+            None,
             id="qwen3vl",
         ),
+        # Qwen2-VL/Qwen2.5-VL ship no ``video_processor_type`` in their
+        # preprocessor config, so resolution relies on the model_type ->
+        # video processor fallback in get_video_processor_cls_name_from_config.
+        # They also ship no default fps/num_frames, so the HF sampler needs an
+        # explicit target rate; pass fps=2 to match the loader default.
+        pytest.param(
+            "Qwen/Qwen2-VL-7B-Instruct",
+            Qwen2VLVideoBackend,
+            {"fps": 2},
+            id="qwen2vl",
+        ),
+        pytest.param(
+            "Qwen/Qwen2.5-VL-7B-Instruct",
+            Qwen2VLVideoBackend,
+            {"fps": 2},
+            id="qwen2_5_vl",
+        ),
     ],
 )
 def test_video_processor_from_model_repo(
     model_repo: str,
     expected_loader_cls: type,
+    hf_sample_kwargs: dict[str, int | float] | None,
 ):
     """Test that a model repo resolves to the correct video loader backend.
 
@@ -143,7 +166,7 @@ def test_video_processor_from_model_repo(
             fps=vllm_meta["fps"],
             duration=vllm_meta["duration"],
         )
-        hf_indices = processor.sample_frames(hf_metadata)
+        hf_indices = processor.sample_frames(hf_metadata, **(hf_sample_kwargs or {}))
         vllm_indices = np.array(vllm_meta["frames_indices"])
         np.testing.assert_array_equal(
             hf_indices,

tests/quantization/test_per_token_kv_cache.py

@@ -61,8 +61,8 @@ class QuantConfig:
     quant_max: float
     quant_min: float
     kv_quant_mode: KVQuantMode
-    # INT8 Triton stores truncate; FP8 hardware casts round.
-    uses_trunc: bool
+    # INT8 rounds explicitly; FP8 relies on dtype cast rounding.
+    rounds_before_store: bool
 
 
 INT8_CONFIG = QuantConfig(
@@ -71,7 +71,7 @@ INT8_CONFIG = QuantConfig(
     quant_max=127.0,
     quant_min=-128.0,
     kv_quant_mode=KVQuantMode.INT8_PER_TOKEN_HEAD,
-    uses_trunc=True,
+    rounds_before_store=True,
 )
 FP8_CONFIG = QuantConfig(
     cache_dtype=FP8_DTYPE,
@@ -79,7 +79,7 @@ FP8_CONFIG = QuantConfig(
     quant_max=FP8_MAX,
     quant_min=FP8_MIN,
     kv_quant_mode=KVQuantMode.FP8_PER_TOKEN_HEAD,
-    uses_trunc=False,
+    rounds_before_store=False,
 )
 
 QUANT_CONFIGS = [INT8_CONFIG, FP8_CONFIG]
@@ -104,7 +104,7 @@ def _quantize_per_token_head_ref(
     absmax = data.float().abs().amax(dim=2)  # [num_tokens, num_heads]
     scales = (absmax / cfg.quant_max).clamp(min=1e-6)
     scaled = data.float() * (1.0 / scales[:, :, None])
-    if cfg.uses_trunc:
+    if cfg.rounds_before_store:
         q = scaled.round().clamp(cfg.quant_min, cfg.quant_max).to(cfg.cache_dtype)
     else:
         q = scaled.clamp(cfg.quant_min, cfg.quant_max).to(cfg.cache_dtype)
@@ -255,7 +255,7 @@ def test_per_token_head_round_trip_accuracy(
 ):
     """Verify per-token-head round-trip: kernel dequant matches reference.
 
-    INT8: Triton truncates on float->int8 store.
+    INT8: round-to-nearest before int8 store.
     FP8: hardware cast (clamp then cast).
     """
     from vllm.v1.attention.ops.triton_reshape_and_cache_flash import (
@@ -315,6 +315,52 @@ def test_per_token_head_round_trip_accuracy(
                 )
 
 
+@torch.inference_mode()
+def test_int8_per_token_head_raw_cache_matches_round_reference():
+    """INT8 cache writes should match round-to-nearest quantization exactly."""
+    from vllm.v1.attention.ops.triton_reshape_and_cache_flash import (
+        triton_reshape_and_cache_flash_per_token_head_quant,
+    )
+
+    torch.set_default_device(DEVICE_TYPE)
+
+    head_size = 8
+    block_size = 4
+
+    key = torch.tensor(
+        [[[-127.0, -2.6, -2.4, -1.6, -1.4, -0.6, -0.4, 127.0]]],
+        dtype=torch.bfloat16,
+    )
+    value = -key
+
+    key_cache = torch.zeros(1, block_size, 1, head_size, dtype=torch.int8)
+    value_cache = torch.zeros_like(key_cache)
+    k_scale_cache = torch.ones(1, block_size, 1, dtype=torch.float32)
+    v_scale_cache = torch.ones_like(k_scale_cache)
+    slot_mapping = torch.tensor([2], dtype=torch.long)
+
+    triton_reshape_and_cache_flash_per_token_head_quant(
+        key,
+        value,
+        key_cache,
+        value_cache,
+        k_scale_cache,
+        v_scale_cache,
+        slot_mapping,
+    )
+
+    ref_k_quant, ref_k_scales = _quantize_per_token_head_ref(key, INT8_CONFIG)
+    ref_v_quant, ref_v_scales = _quantize_per_token_head_ref(value, INT8_CONFIG)
+
+    slot = slot_mapping.item()
+    blk = slot // block_size
+    off = slot % block_size
+    assert torch.equal(key_cache[blk, off], ref_k_quant[0])
+    assert torch.equal(value_cache[blk, off], ref_v_quant[0])
+    torch.testing.assert_close(k_scale_cache[blk, off], ref_k_scales[0])
+    torch.testing.assert_close(v_scale_cache[blk, off], ref_v_scales[0])
+
+
 # ===========================================================================
 # 4. Negative slot mapping (padding tokens should be skipped)
 # ===========================================================================
@@ -461,7 +507,7 @@ def test_triton_unified_attention_per_token_head_scale(
 
     scaled_k = key_cache_bf16.float() / k_scale_cache[:, :, :, None]
     scaled_v = value_cache_bf16.float() / v_scale_cache[:, :, :, None]
-    if qcfg.uses_trunc:
+    if qcfg.rounds_before_store:
         key_cache_q = (
             scaled_k.round().clamp(qcfg.quant_min, qcfg.quant_max).to(qcfg.cache_dtype)
         )

tests/v1/attention/test_attention_backends_selection.py

@@ -6,10 +6,12 @@ from types import SimpleNamespace
 
 import pytest
 
+from vllm.model_executor.layers.mamba.linear.minimax_linear_attn import (
+    MiniMaxText01LinearAttention,
+)
 from vllm.model_executor.layers.mamba.mamba_mixer import MambaMixer
 from vllm.model_executor.layers.mamba.mamba_mixer2 import MambaMixer2
 from vllm.model_executor.layers.mamba.short_conv import ShortConv
-from vllm.model_executor.models.minimax_text_01 import MiniMaxText01LinearAttention
 from vllm.v1.attention.backends.linear_attn import LinearAttentionBackend
 from vllm.v1.attention.backends.mamba1_attn import Mamba1AttentionBackend
 from vllm.v1.attention.backends.mamba2_attn import Mamba2AttentionBackend

tests/v1/attention/test_gdn_metadata_builder.py

@@ -16,6 +16,7 @@ from tests.v1.attention.utils import (
     create_vllm_config,
 )
 from vllm.config import SpeculativeConfig
+from vllm.config.compilation import CUDAGraphMode
 from vllm.v1.attention.backends.gdn_attn import (
     GDNAttentionMetadata,
     GDNAttentionMetadataBuilder,
@@ -123,9 +124,15 @@ GDN_BUILD_TEST_CASES = {
 
 def _create_gdn_builder(
     num_speculative_tokens: int = 0,
+    full_cuda_graph: bool = False,
 ) -> GDNAttentionMetadataBuilder:
     """Create a GDNAttentionMetadataBuilder with minimal config."""
-    vllm_config = create_vllm_config(block_size=BLOCK_SIZE)
+    vllm_config = create_vllm_config(
+        model_name="Qwen/Qwen3.5-0.8B",
+        block_size=BLOCK_SIZE,
+    )
+    if full_cuda_graph:
+        vllm_config.compilation_config.cudagraph_mode = CUDAGraphMode.FULL_AND_PIECEWISE
     if num_speculative_tokens > 0:
         vllm_config.speculative_config = SpeculativeConfig(
             method="ngram",
@@ -189,3 +196,28 @@ def test_has_initial_state_after_reclassification():
     assert meta.has_initial_state is not None
     # req0 has context_lens = 65 - 1 = 64 > 0, so has_initial_state[0] = True
     assert meta.has_initial_state[0].item() is True
+
+
+def test_full_cudagraph_spec_metadata_uses_request_count():
+    """FULL cudagraph token padding must not pad request-indexed metadata."""
+    num_speculative_tokens = 3
+    builder = _create_gdn_builder(
+        num_speculative_tokens=num_speculative_tokens,
+        full_cuda_graph=True,
+    )
+    batch = BatchSpec(seq_lens=[80, 96], query_lens=[4, 4])
+    meta = _build(builder, batch, num_decode_draft_tokens=[3, 3])
+
+    assert meta.num_spec_decodes == batch.batch_size
+    assert meta.num_spec_decode_tokens == batch.compute_num_tokens()
+    assert meta.spec_state_indices_tensor is not None
+    assert meta.spec_state_indices_tensor.shape == (
+        batch.batch_size,
+        num_speculative_tokens + 1,
+    )
+    assert meta.spec_sequence_masks is not None
+    assert meta.spec_sequence_masks.shape == (batch.batch_size,)
+    assert meta.spec_query_start_loc is not None
+    assert meta.spec_query_start_loc.shape == (batch.batch_size + 1,)
+    assert meta.num_accepted_tokens is not None
+    assert meta.num_accepted_tokens.shape == (batch.batch_size,)

tests/v1/core/prefix_cache/test_partial_prefix_cache_primitives.py

@@ -0,0 +1,460 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+from collections.abc import Callable
+
+import pytest
+
+import vllm.v1.core.kv_cache_utils as kv_cache_utils
+from vllm.distributed.kv_events import BlockRemoved, BlockStored
+from vllm.sampling_params import SamplingParams
+from vllm.utils.hashing import sha256
+from vllm.v1.core.block_pool import BlockPool
+from vllm.v1.core.kv_cache_utils import (
+    BlockHash,
+    BlockHashListWithBlockSize,
+    KVCacheBlock,
+    get_request_block_hasher,
+    hash_block_tokens,
+    init_none_hash,
+)
+from vllm.v1.request import Request
+
+pytestmark = pytest.mark.cpu_test
+
+
+@pytest.fixture(autouse=True)
+def _auto_init_hash_fn():
+    init_none_hash(sha256)
+
+
+def make_request(
+    request_id: str,
+    prompt_token_ids: list[int],
+    hash_block_size: int,
+    hash_fn: Callable,
+) -> Request:
+    sampling_params = SamplingParams(max_tokens=17)
+    sampling_params.update_from_generation_config({}, eos_token_id=100)
+    return Request(
+        request_id=request_id,
+        prompt_token_ids=prompt_token_ids,
+        sampling_params=sampling_params,
+        pooling_params=None,
+        block_hasher=get_request_block_hasher(hash_block_size, hash_fn),
+    )
+
+
+def boundary_hash(req: Request, hash_block_size: int, num_tokens: int) -> BlockHash:
+    # Every boundary at a hash_block_size multiple is just the fine-grained
+    # chain hash ending there.
+    return req.block_hashes[num_tokens // hash_block_size - 1]
+
+
+def cache_full_block_and_partial_tail(
+    token_ids: list[int],
+    *,
+    enable_kv_cache_events: bool = False,
+) -> tuple[BlockPool, Request, list[KVCacheBlock], BlockHash]:
+    hash_block_size = 2
+    block_size = 6
+    kv_cache_group_id = 0
+    req = make_request("0", token_ids, hash_block_size, sha256)
+    pool = BlockPool(
+        num_gpu_blocks=3,
+        enable_caching=True,
+        hash_block_size=hash_block_size,
+        enable_kv_cache_events=enable_kv_cache_events,
+    )
+    blocks = pool.get_new_blocks(2)
+
+    pool.cache_full_blocks(
+        request=req,
+        blocks=blocks,
+        num_cached_blocks=0,
+        num_full_blocks=1,
+        block_size=block_size,
+        kv_cache_group_id=kv_cache_group_id,
+    )
+    partial_hash = boundary_hash(req, hash_block_size, len(token_ids))
+    assert pool.cache_partial_block(
+        request=req,
+        block=blocks[1],
+        num_tokens=len(token_ids),
+        kv_cache_group_id=kv_cache_group_id,
+        block_size=block_size,
+    )
+    return pool, req, blocks, partial_hash
+
+
+def test_boundary_hashes_reuse_fine_grained_chain():
+    hash_block_size = 2
+    block_size = 6
+    token_ids = [0, 0, 1, 1, 2, 2, 3, 3, 4, 4]
+    req = make_request("0", token_ids, hash_block_size, sha256)
+
+    coarse = BlockHashListWithBlockSize(req.block_hashes, hash_block_size, block_size)
+    # The block_size=6 full-block hash is the fine hash at the 6-token boundary,
+    # not a concatenation of the three fine hashes inside the block.
+    assert coarse[0] == req.block_hashes[6 // hash_block_size - 1]
+    assert coarse[0] != BlockHash(
+        req.block_hashes[0] + req.block_hashes[1] + req.block_hashes[2]
+    )
+    # A partial tail at 10 tokens is the fine hash at the 10-token boundary,
+    # which chains over the entire prefix.
+    tail_hash = boundary_hash(req, hash_block_size, 10)
+    assert tail_hash == req.block_hashes[4]
+    assert tail_hash == hash_block_tokens(sha256, req.block_hashes[3], token_ids[8:10])
+
+
+def test_cache_partial_block_kv_cache_events():
+    hash_block_size = 4
+    block_size = 12
+    kv_cache_group_id = 2
+
+    pool = BlockPool(
+        num_gpu_blocks=2,
+        enable_caching=True,
+        hash_block_size=hash_block_size,
+        enable_kv_cache_events=True,
+    )
+    req = make_request(
+        "req_partial_events",
+        prompt_token_ids=list(range(hash_block_size * 2)),
+        hash_block_size=hash_block_size,
+        hash_fn=sha256,
+    )
+
+    block = pool.get_new_blocks(1)[0]
+    partial_entry_hash = pool.cache_partial_block(
+        request=req,
+        block=block,
+        num_tokens=hash_block_size * 2,
+        kv_cache_group_id=kv_cache_group_id,
+        block_size=block_size,
+    )
+
+    events = pool.take_events()
+    assert len(events) == 1
+    stored_event = events[0]
+    assert isinstance(stored_event, BlockStored)
+    assert partial_entry_hash is not None
+    assert stored_event.block_hashes == [
+        kv_cache_utils.maybe_convert_block_hash(req.block_hashes[1])
+    ]
+    assert stored_event.parent_block_hash == kv_cache_utils.maybe_convert_block_hash(
+        req.block_hashes[0]
+    )
+    assert stored_event.token_ids == req.all_token_ids[hash_block_size:]
+    assert stored_event.block_size == 4
+    assert stored_event.group_idx == kv_cache_group_id
+
+    duplicate_entry_hash = pool.cache_partial_block(
+        request=req,
+        block=block,
+        num_tokens=hash_block_size * 2,
+        kv_cache_group_id=kv_cache_group_id,
+        block_size=block_size,
+    )
+    assert duplicate_entry_hash == partial_entry_hash
+    assert pool.take_events() == []
+
+    pool.free_blocks([block])
+    pool.get_new_blocks(1)
+    events = pool.take_events()
+    assert len(events) == 1
+    removed_event = events[0]
+    assert isinstance(removed_event, BlockRemoved)
+    assert removed_event.block_hashes == stored_event.block_hashes
+    assert removed_event.group_idx == kv_cache_group_id
+
+
+def test_partial_block_replacement_emits_remove_then_store_events():
+    hash_block_size = 2
+    block_size = 6
+    kv_cache_group_id = 0
+    req = make_request("0", [0, 0, 1, 1, 2, 2, 3, 3], hash_block_size, sha256)
+    pool = BlockPool(
+        num_gpu_blocks=3,
+        enable_caching=True,
+        hash_block_size=hash_block_size,
+        enable_kv_cache_events=True,
+    )
+    blocks = pool.get_new_blocks(2)
+
+    pool.cache_full_blocks(
+        request=req,
+        blocks=blocks,
+        num_cached_blocks=0,
+        num_full_blocks=1,
+        block_size=block_size,
+        kv_cache_group_id=kv_cache_group_id,
+    )
+    partial_hash_8 = boundary_hash(req, hash_block_size, 8)
+    assert pool.cache_partial_block(
+        request=req,
+        block=blocks[1],
+        num_tokens=8,
+        kv_cache_group_id=kv_cache_group_id,
+        block_size=block_size,
+    )
+    assert pool.get_cached_block(partial_hash_8, [kv_cache_group_id]) == [blocks[1]]
+    pool.take_events()
+
+    req.append_output_token_ids([4, 4])
+    partial_hash_10 = boundary_hash(req, hash_block_size, 10)
+    assert pool.cache_partial_block(
+        request=req,
+        block=blocks[1],
+        num_tokens=10,
+        kv_cache_group_id=kv_cache_group_id,
+        block_size=block_size,
+    )
+    events = pool.take_events()
+
+    assert len(events) == 2
+    removed_event, stored_event = events
+    assert isinstance(removed_event, BlockRemoved)
+    assert removed_event.block_hashes == [
+        kv_cache_utils.maybe_convert_block_hash(partial_hash_8)
+    ]
+    assert removed_event.group_idx == kv_cache_group_id
+    assert isinstance(stored_event, BlockStored)
+    assert stored_event.block_hashes == [
+        kv_cache_utils.maybe_convert_block_hash(partial_hash_10)
+    ]
+    assert stored_event.parent_block_hash == kv_cache_utils.maybe_convert_block_hash(
+        boundary_hash(req, hash_block_size, 8)
+    )
+    assert stored_event.token_ids == req.all_token_ids[8:10]
+    assert stored_event.block_size == hash_block_size
+    assert stored_event.group_idx == kv_cache_group_id
+    assert pool.get_cached_block(partial_hash_8, [kv_cache_group_id]) is None
+    assert pool.get_cached_block(partial_hash_10, [kv_cache_group_id]) == [blocks[1]]
+
+
+def test_later_request_hits_cached_partial_tail():
+    hash_block_size = 2
+    block_size = 6
+    kv_cache_group_id = 0
+    cached_token_ids = [0, 0, 1, 1, 2, 2, 3, 3, 4, 4]
+    req = make_request("0", cached_token_ids, hash_block_size, sha256)
+    pool = BlockPool(
+        num_gpu_blocks=3,
+        enable_caching=True,
+        hash_block_size=hash_block_size,
+    )
+    blocks = pool.get_new_blocks(2)
+
+    pool.cache_full_blocks(
+        request=req,
+        blocks=blocks,
+        num_cached_blocks=0,
+        num_full_blocks=1,
+        block_size=block_size,
+        kv_cache_group_id=kv_cache_group_id,
+    )
+    partial_hash_10 = boundary_hash(req, hash_block_size, 10)
+    assert pool.cache_partial_block(
+        request=req,
+        block=blocks[1],
+        num_tokens=10,
+        kv_cache_group_id=kv_cache_group_id,
+        block_size=block_size,
+    )
+
+    replay = make_request("1", cached_token_ids, hash_block_size, sha256)
+    replay_hash_10 = boundary_hash(replay, hash_block_size, 10)
+    assert replay_hash_10 == partial_hash_10
+    assert pool.get_cached_block(replay_hash_10, [kv_cache_group_id]) == [blocks[1]]
+
+    extended = make_request("2", cached_token_ids + [10], hash_block_size, sha256)
+    extended_hash_10 = boundary_hash(extended, hash_block_size, 10)
+    assert extended_hash_10 == partial_hash_10
+    assert pool.get_cached_block(extended_hash_10, [kv_cache_group_id]) == [blocks[1]]
+
+
+def test_cache_partial_block_uses_fine_grained_boundary_hash():
+    hash_block_size = 2
+    block_size = 6
+    kv_cache_group_id = 0
+    token_ids = [0, 0, 1, 1, 2, 2, 3, 3, 4, 4]
+    req = make_request("0", token_ids, hash_block_size, sha256)
+    pool = BlockPool(
+        num_gpu_blocks=3,
+        enable_caching=True,
+        hash_block_size=hash_block_size,
+    )
+    blocks = pool.get_new_blocks(2)
+
+    pool.cache_full_blocks(
+        request=req,
+        blocks=blocks,
+        num_cached_blocks=0,
+        num_full_blocks=1,
+        block_size=block_size,
+        kv_cache_group_id=kv_cache_group_id,
+    )
+
+    partial_entry_hash = pool.cache_partial_block(
+        request=req,
+        block=blocks[1],
+        num_tokens=10,
+        kv_cache_group_id=kv_cache_group_id,
+        block_size=block_size,
+    )
+    # The partial entry is keyed by the fine-grained hash at the 10-token
+    # boundary, regardless of the owning group's block_size.
+    expected = boundary_hash(req, hash_block_size, 10)
+    assert partial_entry_hash == kv_cache_utils.make_block_hash_with_group_id(
+        expected, kv_cache_group_id
+    )
+    assert pool.get_cached_block(expected, [kv_cache_group_id]) == [blocks[1]]
+
+
+def test_cache_partial_block_requires_hash_boundary():
+    hash_block_size = 2
+    block_size = 4
+    req = make_request("0", [0, 0, 1, 1], hash_block_size, sha256)
+    pool = BlockPool(
+        num_gpu_blocks=2,
+        enable_caching=True,
+        hash_block_size=hash_block_size,
+    )
+    block = pool.get_new_blocks(1)[0]
+
+    with pytest.raises(AssertionError):
+        pool.cache_partial_block(
+            request=req,
+            block=block,
+            num_tokens=3,
+            kv_cache_group_id=0,
+            block_size=block_size,
+        )
+
+
+def test_cache_partial_block_duplicate_checks_all_blocks_for_hash():
+    hash_block_size = 2
+    block_size = 4
+    kv_cache_group_id = 0
+    req = make_request("0", [0, 0, 1, 1], hash_block_size, sha256)
+    pool = BlockPool(
+        num_gpu_blocks=4,
+        enable_caching=True,
+        hash_block_size=hash_block_size,
+    )
+    blocks = pool.get_new_blocks(2)
+
+    first_entry_hash = pool.cache_partial_block(
+        request=req,
+        block=blocks[0],
+        num_tokens=2,
+        kv_cache_group_id=kv_cache_group_id,
+        block_size=block_size,
+    )
+    second_entry_hash = pool.cache_partial_block(
+        request=req,
+        block=blocks[1],
+        num_tokens=2,
+        kv_cache_group_id=kv_cache_group_id,
+        block_size=block_size,
+    )
+    assert first_entry_hash == second_entry_hash
+
+    duplicate_entry_hash = pool.cache_partial_block(
+        request=req,
+        block=blocks[1],
+        num_tokens=2,
+        kv_cache_group_id=kv_cache_group_id,
+        block_size=block_size,
+    )
+    assert duplicate_entry_hash == second_entry_hash
+    assert pool.cached_block_hashes_by_block == {}
+
+
+def test_reset_prefix_cache_clears_partial_entry_metadata():
+    pool, req, blocks, partial_hash_10 = cache_full_block_and_partial_tail(
+        [0, 0, 1, 1, 2, 2, 3, 3, 4, 4]
+    )
+    full_hash = BlockHashListWithBlockSize(req.block_hashes, 2, 6)[0]
+
+    assert pool.get_cached_block(full_hash, [0]) == [blocks[0]]
+    assert pool.get_cached_block(partial_hash_10, [0]) == [blocks[1]]
+
+    pool.free_blocks(blocks)
+    assert pool.reset_prefix_cache()
+
+    assert pool.get_cached_block(full_hash, [0]) is None
+    assert pool.get_cached_block(partial_hash_10, [0]) is None
+    assert pool.cached_block_hashes_by_block == {}
+
+
+def test_evict_cached_block_removes_full_hash_and_partial_entry():
+    pool, req, blocks, partial_hash_10 = cache_full_block_and_partial_tail(
+        [0, 0, 1, 1, 2, 2, 3, 3, 4, 4]
+    )
+    full_hash = BlockHashListWithBlockSize(req.block_hashes, 2, 6)[0]
+
+    assert pool.get_cached_block(full_hash, [0]) == [blocks[0]]
+    assert pool.get_cached_block(partial_hash_10, [0]) == [blocks[1]]
+
+    pool.evict_blocks({blocks[0].block_id, blocks[1].block_id})
+
+    assert pool.get_cached_block(full_hash, [0]) is None
+    assert pool.get_cached_block(partial_hash_10, [0]) is None
+    assert pool.cached_block_hashes_by_block == {}
+
+
+def test_partial_block_promotes_to_direct_full_block_hash():
+    hash_block_size = 2
+    block_size = 6
+    kv_cache_group_id = 0
+    token_ids = [0, 0, 1, 1, 2, 2, 3, 3, 4, 4]
+    req = make_request("0", token_ids, hash_block_size, sha256)
+    pool = BlockPool(
+        num_gpu_blocks=3,
+        enable_caching=True,
+        hash_block_size=hash_block_size,
+    )
+    blocks = pool.get_new_blocks(2)
+
+    pool.cache_full_blocks(
+        request=req,
+        blocks=blocks,
+        num_cached_blocks=0,
+        num_full_blocks=1,
+        block_size=block_size,
+        kv_cache_group_id=kv_cache_group_id,
+    )
+    partial_hash_10 = boundary_hash(req, hash_block_size, 10)
+    assert pool.cache_partial_block(
+        request=req,
+        block=blocks[1],
+        num_tokens=10,
+        kv_cache_group_id=kv_cache_group_id,
+        block_size=block_size,
+    )
+    assert pool.get_cached_block(partial_hash_10, [kv_cache_group_id]) == [blocks[1]]
+
+    req.append_output_token_ids([5, 5])
+    full_hashes = BlockHashListWithBlockSize(
+        req.block_hashes, hash_block_size, block_size
+    )
+    promoted_full_hash = full_hashes[1]
+    # The promoted full-block hash is the fine hash at the 12-token boundary,
+    # not a concatenation of the fine hashes inside the block.
+    assert promoted_full_hash == req.block_hashes[12 // hash_block_size - 1]
+    assert promoted_full_hash != BlockHash(
+        req.block_hashes[3] + req.block_hashes[4] + req.block_hashes[5]
+    )
+
+    pool.cache_full_blocks(
+        request=req,
+        blocks=blocks,
+        num_cached_blocks=1,
+        num_full_blocks=2,
+        block_size=block_size,
+        kv_cache_group_id=kv_cache_group_id,
+    )
+    assert pool.get_cached_block(promoted_full_hash, [kv_cache_group_id]) == [blocks[1]]
+    assert pool.get_cached_block(partial_hash_10, [kv_cache_group_id]) is None

tests/v1/core/test_contiguous_kv_packing.py

@@ -1,16 +1,23 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-"""Tests for contiguous KV cache packing in _get_kv_cache_config_deepseek_v4."""
+"""Tests for contiguous KV cache packing."""
 
 from unittest.mock import MagicMock
 
 import pytest
 import torch
 
-from vllm.v1.core.kv_cache_utils import _get_kv_cache_config_deepseek_v4
+from vllm import envs
+from vllm.v1.core.kv_cache_utils import (
+    _get_kv_cache_config_deepseek_v4,
+    get_kv_cache_config_from_groups,
+)
 from vllm.v1.kv_cache_interface import (
+    FullAttentionSpec,
     KVCacheGroupSpec,
+    KVCacheTensor,
     MLAAttentionSpec,
+    SlidingWindowSpec,
     UniformTypeKVCacheSpecs,
 )
 
@@ -28,6 +35,25 @@ def _make_mla_spec(page_size: int, block_size: int = 256) -> MLAAttentionSpec:
     )
 
 
+def _make_full_spec() -> FullAttentionSpec:
+    return FullAttentionSpec(
+        block_size=16,
+        num_kv_heads=2,
+        head_size=64,
+        dtype=torch.float16,
+    )
+
+
+def _make_sw_spec() -> SlidingWindowSpec:
+    return SlidingWindowSpec(
+        block_size=16,
+        num_kv_heads=2,
+        head_size=64,
+        dtype=torch.float16,
+        sliding_window=128,
+    )
+
+
 def _make_groups(n_c4, n_c128, n_swa):
     PS_C4_MLA = 37440
     PS_C4_IDX = 8640
@@ -130,6 +156,73 @@ class TestInterleavedPacking:
         for i, v in enumerate(views):
             assert (v == i + 1).all(), f"View {i} was corrupted"
 
+    def test_hma_attention_groups_keep_default_backing(self, monkeypatch):
+        monkeypatch.setattr(envs, "VLLM_USE_PACKED_HMA_KV_CACHE", False, raising=False)
+        full = _make_full_spec()
+        sw = _make_sw_spec()
+        page_size = full.page_size_bytes
+        groups = [
+            KVCacheGroupSpec(["full.0", "full.1"], full),
+            KVCacheGroupSpec(["sw.0", "sw.2"], sw),
+            KVCacheGroupSpec(["sw.1", "sw.3"], sw),
+        ]
+
+        config = get_kv_cache_config_from_groups(
+            _mock_vllm_config(), groups, available_memory=page_size * 2 * 32
+        )
+
+        assert config.num_blocks == 32
+        assert sum(t.size for t in config.kv_cache_tensors) == page_size * 2 * 32
+        assert config.kv_cache_tensors == [
+            KVCacheTensor(size=page_size * 32, shared_by=["full.0", "sw.0", "sw.1"]),
+            KVCacheTensor(size=page_size * 32, shared_by=["full.1", "sw.2", "sw.3"]),
+        ]
+
+    def test_hma_attention_groups_use_packed_backing_with_flag(self, monkeypatch):
+        monkeypatch.setattr(envs, "VLLM_USE_PACKED_HMA_KV_CACHE", True, raising=False)
+        full = _make_full_spec()
+        sw = _make_sw_spec()
+        page_size = full.page_size_bytes
+        groups = [
+            KVCacheGroupSpec(["full.0", "full.1"], full),
+            KVCacheGroupSpec(["sw.0", "sw.2"], sw),
+            KVCacheGroupSpec(["sw.1", "sw.3"], sw),
+        ]
+
+        config = get_kv_cache_config_from_groups(
+            _mock_vllm_config(), groups, available_memory=page_size * 2 * 32
+        )
+
+        assert config.num_blocks == 32
+        assert {t.size for t in config.kv_cache_tensors} == {page_size * 2 * 32}
+        assert config.kv_cache_tensors == [
+            KVCacheTensor(
+                size=page_size * 2 * 32,
+                shared_by=["full.0", "sw.0", "sw.1"],
+                offset=0,
+                block_stride=page_size * 2,
+            ),
+            KVCacheTensor(
+                size=page_size * 2 * 32,
+                shared_by=["full.1", "sw.2", "sw.3"],
+                offset=page_size,
+                block_stride=page_size * 2,
+            ),
+        ]
+
+    def test_single_group_attention_keeps_unpacked_layout(self):
+        spec = _make_full_spec()
+        groups = [KVCacheGroupSpec(["full.0", "full.1"], spec)]
+
+        config = get_kv_cache_config_from_groups(
+            _mock_vllm_config(), groups, available_memory=spec.page_size_bytes * 2 * 32
+        )
+
+        assert sum(t.size for t in config.kv_cache_tensors) == (
+            spec.page_size_bytes * 2 * 32
+        )
+        assert [t.block_stride for t in config.kv_cache_tensors] == [0, 0]
+
 
 if __name__ == "__main__":
     pytest.main([__file__, "-v"])

tests/v1/core/test_kv_cache_utils.py

@@ -117,6 +117,7 @@ def new_kv_cache_spec(
     page_size_padded=None,
     sliding_window=None,
     attention_chunk_size=None,
+    indexes_kv_by_block_stride=False,
 ):
     return FullAttentionSpec(
         block_size=block_size,
@@ -126,6 +127,7 @@ def new_kv_cache_spec(
         page_size_padded=page_size_padded,
         sliding_window=sliding_window,
         attention_chunk_size=attention_chunk_size,
+        indexes_kv_by_block_stride=indexes_kv_by_block_stride,
     )
 
 
@@ -136,6 +138,7 @@ def new_sliding_window_spec(
     dtype=torch.float32,
     page_size_padded=None,
     sliding_window=1,
+    indexes_kv_by_block_stride=False,
 ):
     return SlidingWindowSpec(
         block_size=block_size,
@@ -144,6 +147,7 @@ def new_sliding_window_spec(
         dtype=dtype,
         page_size_padded=page_size_padded,
         sliding_window=sliding_window,
+        indexes_kv_by_block_stride=indexes_kv_by_block_stride,
     )
 
 
@@ -221,7 +225,7 @@ def test_kv_cache_block():
 
     # Test block hash setting and resetting
     block_hash = make_block_hash_with_group_id(BlockHash(b"abc"), 0)
-    block.block_hash = block_hash
+    block.set_block_hash(block_hash)
     assert block.block_hash == block_hash
 
     block.reset_hash()
@@ -1799,16 +1803,38 @@ def test_get_kv_cache_config_one_worker():
         ],
     )
 
-    # different hidden size that cannot be aligned by using different block size
+    # different hidden size that cannot be aligned by using different block size,
+    # but can be aligned by padding the smaller physical page.
+    swa_spec = new_sliding_window_spec(head_size=96, indexes_kv_by_block_stride=True)
     kv_cache_specs_hybrid = {
-        "layer_1": new_kv_cache_spec(head_size=64),
-        "layer_2": new_sliding_window_spec(head_size=96),
+        "layer_1": new_kv_cache_spec(head_size=64, indexes_kv_by_block_stride=True),
+        "layer_2": swa_spec,
     }
 
-    with pytest.raises(NotImplementedError):
-        get_kv_cache_configs(
-            vllm_config, [kv_cache_specs_hybrid], [mem_per_block_per_layer * 2 * 32]
-        )[0]
+    kv_cache_config_hybrid = get_kv_cache_configs(
+        vllm_config, [kv_cache_specs_hybrid], [mem_per_block_per_layer * 2 * 32]
+    )[0]
+    padded_page_size = swa_spec.page_size_bytes
+    assert kv_cache_config_hybrid == KVCacheConfig(
+        num_blocks=42,
+        kv_cache_tensors=[
+            KVCacheTensor(size=padded_page_size * 42, shared_by=["layer_1", "layer_2"]),
+        ],
+        kv_cache_groups=[
+            KVCacheGroupSpec(
+                ["layer_1"],
+                new_kv_cache_spec(
+                    head_size=64,
+                    page_size_padded=padded_page_size,
+                    indexes_kv_by_block_stride=True,
+                ),
+            ),
+            KVCacheGroupSpec(
+                ["layer_2"],
+                new_sliding_window_spec(head_size=96, indexes_kv_by_block_stride=True),
+            ),
+        ],
+    )
 
     # Test num_gpu_blocks_override
     vllm_config.cache_config.num_gpu_blocks_override = 16
@@ -2322,6 +2348,75 @@ def test_check_enough_kv_cache_memory_respects_num_gpu_blocks_override():
         get_kv_cache_configs(vllm_config, [kv_cache_specs], [large_available_memory])
 
 
+def test_unify_kv_cache_page_size_uses_padding_for_non_divisible_sizes():
+    """DFlash drafters can have a smaller head size than the target model.
+
+    For example, MiMo uses 192-dim target KV heads while its DFlash draft uses
+    128-dim KV heads. The resulting page sizes are 3:2 rather than an integer
+    block-size multiple, so the smaller page must be padded instead.
+    """
+    # Both layers' backends opt into the padded-page strided view (e.g.
+    # FlashAttention / its DiffKV subclass), so padding is allowed.
+    target_spec = new_kv_cache_spec(
+        block_size=16,
+        num_kv_heads=1,
+        head_size=192,
+        dtype=torch.bfloat16,
+        indexes_kv_by_block_stride=True,
+    )
+    draft_spec = new_sliding_window_spec(
+        block_size=16,
+        num_kv_heads=1,
+        head_size=128,
+        dtype=torch.bfloat16,
+        sliding_window=1024,
+        indexes_kv_by_block_stride=True,
+    )
+
+    unified_specs = kv_cache_utils.unify_kv_cache_spec_page_size(
+        {
+            "target_attn": target_spec,
+            "draft_attn": draft_spec,
+        }
+    )
+
+    assert unified_specs["target_attn"] == target_spec
+    unified_draft_spec = unified_specs["draft_attn"]
+    assert unified_draft_spec.block_size == draft_spec.block_size
+    assert unified_draft_spec.real_page_size_bytes == draft_spec.real_page_size_bytes
+    assert unified_draft_spec.page_size_padded == target_spec.page_size_bytes
+    assert unified_draft_spec.page_size_bytes == target_spec.page_size_bytes
+
+
+def test_unify_kv_cache_page_size_padding_requires_backend_support():
+    """Padding is gated on the backend declaring ``indexes_kv_by_block_stride``.
+
+    A backend that does not support the strided padded-page view must raise
+    rather than silently padding (and misreading KV at runtime).
+    """
+    target_spec = new_kv_cache_spec(
+        block_size=16,
+        num_kv_heads=1,
+        head_size=192,
+        dtype=torch.bfloat16,
+        indexes_kv_by_block_stride=True,
+    )
+    # The non-divisible draft layer needs padding but its backend does not
+    # support the strided padded-page view -> must raise, not silently pad.
+    draft_spec = new_sliding_window_spec(
+        block_size=16,
+        num_kv_heads=1,
+        head_size=128,
+        dtype=torch.bfloat16,
+        sliding_window=1024,
+        indexes_kv_by_block_stride=False,
+    )
+    specs = {"target_attn": target_spec, "draft_attn": draft_spec}
+
+    with pytest.raises(NotImplementedError):
+        kv_cache_utils.unify_kv_cache_spec_page_size(specs)
+
+
 def test_unify_hybrid_kv_cache_specs():
     # 1. has_full_attention and has_sliding_window
     before_spec_1 = new_kv_cache_spec()

tests/v1/core/test_prefix_caching.py

@@ -2003,7 +2003,7 @@ def test_maybe_evict_cached_block():
     assert len(pool.blocks) == len(block_hashes)
     # Manually add all blocks to cached_blocks
     for block, block_hash in zip(pool.blocks, block_hashes):
-        block.block_hash = block_hash
+        block.set_block_hash(block_hash)
         pool.cached_block_hash_to_block.insert(block_hash, block)
 
     block0, block1, block2, block3 = pool.blocks

tests/v1/core/test_scheduler.py

@@ -144,6 +144,43 @@ def test_async_scheduling_pp_allows_rescheduling_with_output_placeholders():
     assert req.request_id in output.num_scheduled_tokens
 
 
+def test_cached_request_data_resumed_all_token_ids_mrv1_only():
+    """all_token_ids carries a resumed request's token ids to the connector
+    for the V1 model runner, but is skipped entirely for the V2 model runner.
+    """
+    from vllm.v1.core.kv_cache_manager import KVCacheBlocks
+
+    scheduler = create_scheduler()
+    (req,) = create_requests(num_requests=1, num_tokens=8)
+    req.append_output_token_ids([101, 102, 103])
+
+    # A resumed request was not scheduled in the previous step.
+    assert req.request_id not in scheduler.prev_step_scheduled_req_ids
+
+    empty_blocks = KVCacheBlocks(blocks=((),))
+
+    def make_cached():
+        return scheduler._make_cached_request_data(
+            running_reqs=[],
+            resumed_reqs=[req],
+            num_scheduled_tokens={req.request_id: 1},
+            spec_decode_tokens={},
+            req_to_new_blocks={req.request_id: empty_blocks},
+        )
+
+    # V1 model runner: the full token id list is propagated.
+    assert not scheduler.use_v2_model_runner
+    cached = make_cached()
+    assert req.request_id in cached.resumed_req_ids
+    assert cached.all_token_ids[req.request_id] == list(req.all_token_ids)
+
+    # V2 model runner: all_token_ids is skipped entirely.
+    scheduler.use_v2_model_runner = True
+    cached = make_cached()
+    assert req.request_id in cached.resumed_req_ids
+    assert cached.all_token_ids == {}
+
+
 def test_schedule_partial_requests():
     """Test scheduling behavior with partial requests.
 

tests/v1/e2e/spec_decode/test_spec_decode.py

@@ -425,7 +425,7 @@ def _run_eagle_correctness(
             if "deepseek" in model_setup[1].lower():
                 m.setenv("VLLM_ROCM_USE_AITER", "1")
                 m.delenv("VLLM_MLA_DISABLE", raising=False)
-                attention_config = {"backend": "TRITON_MLA"}
+                attention_config = {"backend": "ROCM_AITER_MLA"}
             else:
                 m.setenv("VLLM_ROCM_USE_AITER", "1")
 

tests/v1/kv_connector/unit/offloading_connector/test_events.py

@@ -0,0 +1,355 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+from unittest.mock import MagicMock
+
+import pytest
+import torch
+
+from tests.v1.kv_connector.unit.utils import create_vllm_config
+from vllm.config import KVEventsConfig, KVTransferConfig
+from vllm.distributed.kv_events import BlockRemoved, BlockStored
+from vllm.distributed.kv_transfer.kv_connector.v1.offloading.events import (
+    OffloadingEventGroupSpec,
+    OffloadingEventsTracker,
+)
+from vllm.distributed.kv_transfer.kv_connector.v1.offloading.scheduler import (
+    GroupOffloadConfig,
+)
+from vllm.v1.core.kv_cache_utils import BlockHash, maybe_convert_block_hash
+from vllm.v1.kv_cache_interface import (
+    FullAttentionSpec,
+    KVCacheConfig,
+    KVCacheGroupSpec,
+    KVCacheSpecKind,
+)
+from vllm.v1.kv_offload.base import (
+    OffloadingEvent,
+    OffloadingKVEventsConfig,
+    OffloadKey,
+    make_offload_key,
+)
+from vllm.v1.kv_offload.cpu.common import CPULoadStoreSpec
+from vllm.v1.kv_offload.tiering.spec import TieringOffloadingSpec
+
+_CPU_MEDIUM = CPULoadStoreSpec.medium()
+_FULL_ATTENTION_EVENT_SPEC = OffloadingEventGroupSpec(
+    kv_cache_spec_kind=KVCacheSpecKind.FULL_ATTENTION.value,
+    kv_cache_spec_sliding_window=None,
+)
+
+
+def _tracker(
+    *,
+    enable_kv_cache_events: bool = True,
+    self_describing_kv_events: bool = True,
+) -> OffloadingEventsTracker:
+    return OffloadingEventsTracker(
+        OffloadingKVEventsConfig(
+            enable_kv_cache_events=enable_kv_cache_events,
+            self_describing_kv_events=self_describing_kv_events,
+        )
+    )
+
+
+def _hash(i: int) -> BlockHash:
+    return BlockHash(str(i).encode())
+
+
+def _wire_hash(block_hash: BlockHash):
+    return maybe_convert_block_hash(block_hash)
+
+
+def _request(*, block_hashes: list[BlockHash], token_count: int):
+    req = MagicMock()
+    req.block_hashes = block_hashes
+    req.all_token_ids = list(range(1, token_count + 1))
+    req.lora_request = None
+    return req
+
+
+def _group_config(
+    *,
+    group_idx: int = 0,
+    block_size: int = 4,
+    block_size_factor: int = 1,
+    sliding_window_size_in_blocks: int | None = None,
+) -> GroupOffloadConfig:
+    return GroupOffloadConfig(
+        group_idx=group_idx,
+        gpu_block_size=block_size,
+        offloaded_block_size=block_size * block_size_factor,
+        hash_block_size_factor=block_size_factor,
+        sliding_window_size_in_blocks=sliding_window_size_in_blocks,
+        kv_event_group_spec=_FULL_ATTENTION_EVENT_SPEC,
+    )
+
+
+def _record_chunks(
+    tracker: OffloadingEventsTracker,
+    req,
+    group_config: GroupOffloadConfig,
+    num_chunks: int,
+) -> list[OffloadKey]:
+    keys: list[OffloadKey] = []
+    hbf = group_config.hash_block_size_factor
+    for chunk_idx in range(num_chunks):
+        tail_hash = req.block_hashes[(chunk_idx + 1) * hbf - 1]
+        assert tail_hash is not None
+        key = make_offload_key(tail_hash, group_config.group_idx)
+        tracker.record_store(req, group_config, chunk_idx, key)
+        keys.append(key)
+    return keys
+
+
+def _stored_event(keys: list[OffloadKey]) -> OffloadingEvent:
+    return OffloadingEvent(keys=keys, medium=_CPU_MEDIUM, removed=False)
+
+
+def _removed_event(keys: list[OffloadKey]) -> OffloadingEvent:
+    return OffloadingEvent(keys=keys, medium=_CPU_MEDIUM, removed=True)
+
+
+def test_take_events_publishes_routable_block_stored():
+    block_size = 4
+    tracker = _tracker()
+    group_config = _group_config(block_size=block_size)
+    req = _request(
+        block_hashes=[_hash(i) for i in range(6)],
+        token_count=block_size * 6,
+    )
+    keys = _record_chunks(tracker, req, group_config, num_chunks=6)
+
+    batch1 = list(tracker.take_events([_stored_event(keys[:3])]))
+    assert len(batch1) == 3
+
+    for i, event in enumerate(batch1):
+        assert isinstance(event, BlockStored)
+        assert event.medium == _CPU_MEDIUM
+        assert event.block_hashes == [_wire_hash(_hash(i))]
+        assert event.block_size == block_size
+        assert event.token_ids == list(
+            range(i * block_size + 1, (i + 1) * block_size + 1)
+        )
+        if i == 0:
+            assert event.parent_block_hash is None
+        else:
+            assert event.parent_block_hash == _wire_hash(_hash(i - 1))
+        assert event.lora_id is None
+        assert event.lora_name is None
+        assert event.extra_keys is None
+        assert event.group_idx == 0
+        assert event.kv_cache_spec_kind == KVCacheSpecKind.FULL_ATTENTION.value
+        assert event.kv_cache_spec_sliding_window is None
+
+    batch2 = list(tracker.take_events([_stored_event(keys[3:])]))
+    assert len(batch2) == 3
+    assert batch2[0].parent_block_hash == batch1[-1].block_hashes[-1]
+
+    assert len(tracker._pending_event_metadata) == 6
+
+
+def test_take_events_factor_gt_1_chunk_store_and_remove():
+    block_size = 4
+    block_size_factor = 3
+    tracker = _tracker()
+    group_config = _group_config(
+        block_size=block_size, block_size_factor=block_size_factor
+    )
+    req = _request(
+        block_hashes=[_hash(i) for i in range(6)],
+        token_count=block_size * block_size_factor * 2,
+    )
+    keys = _record_chunks(tracker, req, group_config, num_chunks=2)
+
+    stored = list(tracker.take_events([_stored_event(keys)]))
+    assert len(stored) == 2
+
+    expected_hashes = []
+    for chunk_idx, event in enumerate(stored):
+        assert isinstance(event, BlockStored)
+        expected_chunk_hashes = [
+            _wire_hash(_hash(i))
+            for i in range(
+                chunk_idx * block_size_factor,
+                (chunk_idx + 1) * block_size_factor,
+            )
+        ]
+        assert event.block_hashes == expected_chunk_hashes
+        assert event.block_size == block_size
+        assert len(event.token_ids) == block_size * block_size_factor
+        if chunk_idx == 0:
+            assert event.parent_block_hash is None
+        else:
+            assert event.parent_block_hash == _wire_hash(_hash(block_size_factor - 1))
+        expected_hashes.extend(expected_chunk_hashes)
+
+    assert len(tracker._pending_event_metadata) == 2
+
+    removed = list(tracker.take_events([_removed_event(keys)]))
+    assert len(removed) == 1
+    assert isinstance(removed[0], BlockRemoved)
+    assert removed[0].block_hashes == expected_hashes
+    assert removed[0].medium == _CPU_MEDIUM
+    assert removed[0].group_idx == 0
+    assert not tracker._pending_event_metadata
+
+
+def test_take_events_factor_gt_1_store_is_order_independent():
+    block_size_factor = 3
+    tracker = _tracker()
+    group_config = _group_config(block_size_factor=block_size_factor)
+    req = _request(
+        block_hashes=[_hash(i) for i in range(6)],
+        token_count=4 * block_size_factor * 2,
+    )
+    keys = _record_chunks(tracker, req, group_config, num_chunks=2)
+    unknown_key = make_offload_key(_hash(12345), 0)
+
+    events = list(tracker.take_events([_stored_event([keys[1], unknown_key, keys[0]])]))
+
+    assert len(events) == 3
+    chunk1, placeholder, chunk0 = events
+    assert [len(event.block_hashes) for event in events] == [3, 1, 3]
+    assert placeholder.block_size == 0
+    assert placeholder.token_ids == []
+    assert chunk0.parent_block_hash is None
+    assert chunk1.parent_block_hash == chunk0.block_hashes[-1]
+
+
+def test_take_events_opt_out_keeps_placeholders():
+    tracker = _tracker(self_describing_kv_events=False)
+    group_config = _group_config()
+    req = _request(block_hashes=[_hash(i) for i in range(3)], token_count=12)
+    keys = _record_chunks(tracker, req, group_config, num_chunks=3)
+
+    assert not tracker.self_describing_enabled
+    assert not tracker._pending_event_metadata
+
+    events = list(
+        tracker.take_events(
+            [
+                _stored_event(keys),
+                _removed_event(keys),
+            ]
+        )
+    )
+    assert len(events) == 4
+    for event in events[:3]:
+        assert isinstance(event, BlockStored)
+        assert event.block_size == 0
+        assert event.token_ids == []
+        assert event.parent_block_hash is None
+    assert isinstance(events[3], BlockRemoved)
+    assert len(events[3].block_hashes) == 3
+
+
+def test_record_store_skips_sliding_window_group():
+    tracker = _tracker()
+    group_config = _group_config(sliding_window_size_in_blocks=2)
+    req = _request(block_hashes=[_hash(i) for i in range(3)], token_count=12)
+    keys = _record_chunks(tracker, req, group_config, num_chunks=3)
+
+    assert not tracker._pending_event_metadata
+
+    events = list(tracker.take_events([_stored_event(keys[:1])]))
+    assert len(events) == 1
+    assert isinstance(events[0], BlockStored)
+    assert events[0].block_size == 0
+
+
+def test_take_events_groups_removed_hashes_by_kv_group():
+    tracker = _tracker()
+    group0_config = _group_config(group_idx=0, block_size_factor=2)
+    group1_config = _group_config(group_idx=1, block_size_factor=2)
+    req0 = _request(block_hashes=[_hash(0), _hash(1)], token_count=8)
+    req1 = _request(block_hashes=[_hash(10), _hash(11)], token_count=8)
+    key0 = _record_chunks(tracker, req0, group0_config, num_chunks=1)[0]
+    key1 = _record_chunks(tracker, req1, group1_config, num_chunks=1)[0]
+
+    removed = list(tracker.take_events([_removed_event([key0, key1])]))
+
+    assert len(removed) == 2
+    by_group = {event.group_idx: event.block_hashes for event in removed}
+    assert by_group == {
+        0: [_wire_hash(_hash(0)), _wire_hash(_hash(1))],
+        1: [_wire_hash(_hash(10)), _wire_hash(_hash(11))],
+    }
+
+
+def test_take_events_supports_restore_after_eviction():
+    block_size = 4
+    tracker = _tracker()
+    group_config = _group_config(block_size=block_size)
+    req = _request(block_hashes=[_hash(0)], token_count=block_size)
+    key = _record_chunks(tracker, req, group_config, num_chunks=1)[0]
+
+    first_store = list(tracker.take_events([_stored_event([key])]))
+    assert len(first_store) == 1
+    assert isinstance(first_store[0], BlockStored)
+    assert first_store[0].token_ids == [1, 2, 3, 4]
+
+    removed = list(tracker.take_events([_removed_event([key])]))
+    assert len(removed) == 1
+    assert isinstance(removed[0], BlockRemoved)
+    assert not tracker._pending_event_metadata
+
+    req.all_token_ids = [5, 6, 7, 8]
+    tracker.record_store(req, group_config, offload_block_idx=0, offload_key=key)
+
+    second_store = list(tracker.take_events([_stored_event([key])]))
+    assert len(second_store) == 1
+    assert isinstance(second_store[0], BlockStored)
+    assert second_store[0].token_ids == [5, 6, 7, 8]
+
+
+def test_reset_cache_clears_side_table():
+    tracker = _tracker()
+    group_config = _group_config()
+    req = _request(block_hashes=[_hash(i) for i in range(3)], token_count=12)
+    _record_chunks(tracker, req, group_config, num_chunks=3)
+
+    assert tracker._pending_event_metadata
+
+    tracker.reset()
+
+    assert not tracker._pending_event_metadata
+
+
+def test_tiering_rejects_self_describing_kv_events():
+    vllm_config = create_vllm_config(
+        block_size=4,
+        max_num_batched_tokens=16,
+        disable_hybrid_kv_cache_manager=False,
+    )
+    vllm_config.kv_transfer_config = KVTransferConfig(
+        kv_connector="OffloadingConnector",
+        kv_role="kv_both",
+        kv_connector_extra_config={
+            "spec_name": "TieringOffloadingSpec",
+            "cpu_bytes_to_use": 1 << 20,
+            "self_describing_kv_events": True,
+            "secondary_tiers": [{"type": "example"}],
+        },
+    )
+    vllm_config.kv_events_config = KVEventsConfig(
+        enable_kv_cache_events=True,
+        publisher="null",
+    )
+    kv_cache_config = KVCacheConfig(
+        num_blocks=0,
+        kv_cache_tensors=[],
+        kv_cache_groups=[
+            KVCacheGroupSpec(
+                ["layer"],
+                FullAttentionSpec(
+                    block_size=4,
+                    num_kv_heads=1,
+                    head_size=1,
+                    dtype=torch.float32,
+                ),
+            )
+        ],
+    )
+
+    with pytest.raises(ValueError, match="TieringOffloadingSpec"):
+        TieringOffloadingSpec(vllm_config, kv_cache_config)

tests/v1/kv_connector/unit/offloading_connector/test_metrics.py

@@ -22,7 +22,7 @@ from vllm.v1.kv_offload.base import (
     OffloadingGaugeMetadata,
     OffloadingHistogramMetadata,
 )
-from vllm.v1.kv_offload.cpu.spec import CPUOffloadingSpec
+from vllm.v1.kv_offload.factory import OffloadingSpecFactory
 
 LOAD_BYTES = _TransferMetricName.LOAD_BYTES
 LOAD_TIME = _TransferMetricName.LOAD_TIME
@@ -33,6 +33,8 @@ STORE_SIZE = _TransferMetricName.STORE_SIZE
 STORES_SKIPPED = "vllm:kv_offload_stores_skipped"
 PENDING_STORES = "vllm:kv_offload_pending_stores"
 LOOKUP_LATENCY = "vllm:kv_offload_lookup_latency_seconds"
+MY_COUNTER = "my_counter"
+MY_LABEL = "my_label"
 
 
 class _FakeMetric:
@@ -67,6 +69,20 @@ class _FakeVllmConfig:
         )
 
 
+def _spec_cls_with_metric_definitions(
+    metric_definitions: dict[str, Any],
+) -> type:
+    """Build a fake offloading spec class reporting the given metric
+    definitions, so tests don't need to patch the real CPU spec."""
+
+    class _FakeOffloadingSpec:
+        @staticmethod
+        def build_metric_definitions(extra_config):
+            return metric_definitions
+
+    return _FakeOffloadingSpec
+
+
 def _metric_metadata():
     return {
         LOAD_BYTES: OffloadingCounterMetadata(
@@ -96,9 +112,17 @@ def _metric_metadata():
         LOOKUP_LATENCY: OffloadingHistogramMetadata(
             documentation="lookup latency",
         ),
+        MY_COUNTER: OffloadingCounterMetadata(
+            documentation="counter with a label",
+            labelnames=(MY_LABEL,),
+        ),
     }
 
 
+def _unlabeled(values: dict[str, Any], metric_name: str) -> Any:
+    return values[metric_name][()]
+
+
 def test_build_kv_connector_stats_with_none():
     """Test that build_kv_connector_stats returns empty stats when given None."""
     stats = OffloadingConnector.build_kv_connector_stats(data=None)
@@ -131,13 +155,13 @@ def test_build_kv_connector_stats_reconstructs_offload_stats():
             STORES_SKIPPED: _MetricType.COUNTER,
         },
         _StatsKey.DATA: {
-            LOAD_BYTES: 24,
-            LOAD_TIME: 1.5,
-            LOAD_SIZE: [16, 8],
-            STORE_BYTES: 3,
-            STORE_TIME: 0.3,
-            STORE_SIZE: [1, 2],
-            STORES_SKIPPED: 5,
+            LOAD_BYTES: {(): 24},
+            LOAD_TIME: {(): 1.5},
+            LOAD_SIZE: {(): [16, 8]},
+            STORE_BYTES: {(): 3},
+            STORE_TIME: {(): 0.3},
+            STORE_SIZE: {(): [1, 2]},
+            STORES_SKIPPED: {(): 5},
         },
     }
 
@@ -145,22 +169,28 @@ def test_build_kv_connector_stats_reconstructs_offload_stats():
 
     assert isinstance(stats, OffloadingConnectorStats)
     values = stats.data[_StatsKey.DATA]
-    assert values[LOAD_BYTES] == 24
-    assert values[LOAD_TIME] == 1.5
-    assert values[LOAD_SIZE] == [16, 8]
-    assert values[STORE_BYTES] == 3
-    assert values[STORE_TIME] == 0.3
-    assert values[STORE_SIZE] == [1, 2]
-    assert values[STORES_SKIPPED] == 5
+    assert _unlabeled(values, LOAD_BYTES) == 24
+    assert _unlabeled(values, LOAD_TIME) == 1.5
+    assert _unlabeled(values, LOAD_SIZE) == [16, 8]
+    assert _unlabeled(values, STORE_BYTES) == 3
+    assert _unlabeled(values, STORE_TIME) == 0.3
+    assert _unlabeled(values, STORE_SIZE) == [1, 2]
+    assert _unlabeled(values, STORES_SKIPPED) == 5
 
 
 def _make_stats_data(
     metric_data: dict[str, Any],
     metric_metadata: dict[str, Any],
 ) -> dict[str, Any]:
-    """Build a structured data dict from flat metric data and metadata."""
+    """Build a structured data dict from flat metric data and metadata.
+
+    Values for unlabeled metrics may be passed flat (wrapped here under the
+    empty label tuple); values for labeled metrics must already be passed as
+    a ``{labelvalues: value}`` map.
+    """
     metric_types = {}
-    for key in metric_data:
+    data = {}
+    for key, value in metric_data.items():
         md = metric_metadata[key]
         if isinstance(md, OffloadingCounterMetadata):
             metric_types[key] = _MetricType.COUNTER
@@ -168,9 +198,10 @@ def _make_stats_data(
             metric_types[key] = _MetricType.GAUGE
         elif isinstance(md, OffloadingHistogramMetadata):
             metric_types[key] = _MetricType.HISTOGRAM
+        data[key] = value if md.labelnames else {(): value}
     return {
         _StatsKey.TYPES: metric_types,
-        _StatsKey.DATA: metric_data,
+        _StatsKey.DATA: data,
     }
 
 
@@ -215,34 +246,106 @@ def test_aggregate_same_connector():
 
     assert result is stats1  # Should return self
     values = result.data[_StatsKey.DATA]
-    assert values[LOAD_BYTES] == 34
-    assert values[LOAD_TIME] == 2.6
-    assert values[LOAD_SIZE] == [16, 8, 3, 7]
-    assert values[STORE_BYTES] == 19
-    assert values[STORE_TIME] == 2.3
-    assert values[STORE_SIZE] == [1, 2, 16]
-    assert values[STORES_SKIPPED] == 4
-    assert values[PENDING_STORES] == 1
-    assert values[LOOKUP_LATENCY] == [0.1, 0.2, 0.3]
+    assert _unlabeled(values, LOAD_BYTES) == 34
+    assert _unlabeled(values, LOAD_TIME) == 2.6
+    assert _unlabeled(values, LOAD_SIZE) == [16, 8, 3, 7]
+    assert _unlabeled(values, STORE_BYTES) == 19
+    assert _unlabeled(values, STORE_TIME) == 2.3
+    assert _unlabeled(values, STORE_SIZE) == [1, 2, 16]
+    assert _unlabeled(values, STORES_SKIPPED) == 4
+    assert _unlabeled(values, PENDING_STORES) == 1
+    assert _unlabeled(values, LOOKUP_LATENCY) == [0.1, 0.2, 0.3]
+
+
+def test_aggregate_labeled_metrics():
+    metadata = _metric_metadata()
+    stats1 = OffloadingConnectorStats(
+        data=_make_stats_data(
+            {
+                MY_COUNTER: {
+                    ("a",): 10,
+                    ("b",): 3,
+                },
+            },
+            metadata,
+        ),
+    )
+    stats2 = OffloadingConnectorStats(
+        data=_make_stats_data(
+            {
+                MY_COUNTER: {
+                    ("a",): 7,
+                    ("c",): 5,
+                },
+            },
+            metadata,
+        ),
+    )
+
+    stats1.aggregate(stats2)
+
+    values = stats1.data[_StatsKey.DATA][MY_COUNTER]
+    assert values[("a",)] == 17
+    assert values[("b",)] == 3
+    assert values[("c",)] == 5
+
+
+def test_aggregate_labeled_metric_missing_from_self():
+    """Aggregating a labeled metric that self doesn't have at all yet."""
+    metadata = _metric_metadata()
+    stats1 = OffloadingConnectorStats()
+    stats2 = OffloadingConnectorStats(
+        data=_make_stats_data(
+            {
+                MY_COUNTER: {
+                    ("a",): 7,
+                    ("b",): 5,
+                },
+            },
+            metadata,
+        ),
+    )
+
+    stats1.aggregate(stats2)
+
+    values = stats1.data[_StatsKey.DATA][MY_COUNTER]
+    assert values[("a",)] == 7
+    assert values[("b",)] == 5
+    assert stats1.data[_StatsKey.TYPES][MY_COUNTER] == _MetricType.COUNTER
+
+
+def test_helper_methods_accept_labeled_metrics():
+    stats = OffloadingConnectorStats()
+
+    stats.increase_counter(MY_COUNTER, 3, ("a",))
+    stats.increase_counter(MY_COUNTER, 4, ("a",))
+    stats.set_gauge(PENDING_STORES, 2, ("b",))
+    stats.observe_histogram(LOOKUP_LATENCY, 0.1, ("b",))
+    stats.observe_histogram(LOOKUP_LATENCY, 0.2, ("b",))
+
+    values = stats.data[_StatsKey.DATA]
+    assert values[MY_COUNTER][("a",)] == 7
+    assert values[PENDING_STORES][("b",)] == 2
+    assert values[LOOKUP_LATENCY][("b",)] == [0.1, 0.2]
 
 
 def test_aggregate_merges_types():
     stats1 = OffloadingConnectorStats(
         data={
             _StatsKey.TYPES: {LOAD_BYTES: _MetricType.COUNTER},
-            _StatsKey.DATA: {LOAD_BYTES: 1},
+            _StatsKey.DATA: {LOAD_BYTES: {(): 1}},
         },
     )
     stats2 = OffloadingConnectorStats(
         data={
             _StatsKey.TYPES: {PENDING_STORES: _MetricType.GAUGE},
-            _StatsKey.DATA: {PENDING_STORES: 2},
+            _StatsKey.DATA: {PENDING_STORES: {(): 2}},
         },
     )
 
     result = stats1.aggregate(stats2)
 
-    assert result.data[_StatsKey.DATA][PENDING_STORES] == 2
+    assert _unlabeled(result.data[_StatsKey.DATA], PENDING_STORES) == 2
     assert result.data[_StatsKey.TYPES][PENDING_STORES] == _MetricType.GAUGE
 
 
@@ -283,6 +386,26 @@ def test_reduce():
     assert reduced[f"{LOOKUP_LATENCY}_sum"] == sum([0.1, 0.2, 0.3])
 
 
+def test_reduce_labeled_metrics():
+    metadata = _metric_metadata()
+    stats = OffloadingConnectorStats(
+        data=_make_stats_data(
+            {
+                MY_COUNTER: {
+                    ("a",): 17,
+                    ("b",): 3,
+                },
+            },
+            metadata,
+        ),
+    )
+
+    reduced = stats.reduce()
+
+    assert reduced[f"{MY_COUNTER}:{('a',)}"] == 17
+    assert reduced[f"{MY_COUNTER}:{('b',)}"] == 3
+
+
 def test_reset():
     """Test that reset() resets all connector stats."""
     metadata = _metric_metadata()
@@ -326,11 +449,11 @@ def test_prom_metrics_observes_manager_counter():
     prom_metrics.observe(
         {
             _StatsKey.TYPES: {STORES_SKIPPED: _MetricType.COUNTER},
-            _StatsKey.DATA: {STORES_SKIPPED: 7},
+            _StatsKey.DATA: {STORES_SKIPPED: {(): 7}},
         }
     )
 
-    counter = prom_metrics.offloading_metrics[(0, STORES_SKIPPED)]
+    counter = prom_metrics.offloading_metrics[(0, STORES_SKIPPED, ())]
     assert counter.increments == [7]
     counter_def = prom_metrics._offloading_metric_defs[STORES_SKIPPED]
     assert counter_def.kwargs["name"] == "vllm:kv_offload_stores_skipped"
@@ -360,22 +483,22 @@ def test_prom_metrics_observes_flat_transfer_metrics_and_legacy_metrics():
                 STORE_SIZE: _MetricType.HISTOGRAM,
             },
             _StatsKey.DATA: {
-                LOAD_BYTES: 24,
-                LOAD_TIME: 1.5,
-                LOAD_SIZE: [16, 8],
-                STORE_BYTES: 3,
-                STORE_TIME: 0.3,
-                STORE_SIZE: [1, 2],
+                LOAD_BYTES: {(): 24},
+                LOAD_TIME: {(): 1.5},
+                LOAD_SIZE: {(): [16, 8]},
+                STORE_BYTES: {(): 3},
+                STORE_TIME: {(): 0.3},
+                STORE_SIZE: {(): [1, 2]},
             },
         }
     )
 
-    assert prom_metrics.offloading_metrics[(0, LOAD_BYTES)].increments == [24]
-    assert prom_metrics.offloading_metrics[(0, LOAD_TIME)].increments == [1.5]
-    assert prom_metrics.offloading_metrics[(0, LOAD_SIZE)].observed == [16, 8]
-    assert prom_metrics.offloading_metrics[(0, STORE_BYTES)].increments == [3]
-    assert prom_metrics.offloading_metrics[(0, STORE_TIME)].increments == [0.3]
-    assert prom_metrics.offloading_metrics[(0, STORE_SIZE)].observed == [1, 2]
+    assert prom_metrics.offloading_metrics[(0, LOAD_BYTES, ())].increments == [24]
+    assert prom_metrics.offloading_metrics[(0, LOAD_TIME, ())].increments == [1.5]
+    assert prom_metrics.offloading_metrics[(0, LOAD_SIZE, ())].observed == [16, 8]
+    assert prom_metrics.offloading_metrics[(0, STORE_BYTES, ())].increments == [3]
+    assert prom_metrics.offloading_metrics[(0, STORE_TIME, ())].increments == [0.3]
+    assert prom_metrics.offloading_metrics[(0, STORE_SIZE, ())].observed == [1, 2]
 
     assert prom_metrics.counter_kv_bytes[(0, "CPU_to_GPU")].increments == [24]
     assert prom_metrics.counter_kv_transfer_time[(0, "CPU_to_GPU")].increments == [1.5]
@@ -396,7 +519,9 @@ def test_prom_metrics_observes_manager_gauge_and_histogram():
         ),
     }
     with patch.object(
-        CPUOffloadingSpec, "build_metric_definitions", return_value=metric_definitions
+        OffloadingSpecFactory,
+        "get_spec_cls",
+        return_value=_spec_cls_with_metric_definitions(metric_definitions),
     ):
         prom_metrics = OffloadPromMetrics(
             vllm_config=_FakeVllmConfig(store_threshold=0),  # type: ignore[arg-type]
@@ -416,20 +541,91 @@ def test_prom_metrics_observes_manager_gauge_and_histogram():
                 LOOKUP_LATENCY: _MetricType.HISTOGRAM,
             },
             _StatsKey.DATA: {
-                PENDING_STORES: 5,
-                LOOKUP_LATENCY: [0.2, 0.4],
+                PENDING_STORES: {(): 5},
+                LOOKUP_LATENCY: {(): [0.2, 0.4]},
             },
         }
     )
 
-    gauge = prom_metrics.offloading_metrics[(0, PENDING_STORES)]
-    histogram = prom_metrics.offloading_metrics[(0, LOOKUP_LATENCY)]
+    gauge = prom_metrics.offloading_metrics[(0, PENDING_STORES, ())]
+    histogram = prom_metrics.offloading_metrics[(0, LOOKUP_LATENCY, ())]
     assert gauge.set_values == [5]
     assert histogram.observed == [0.2, 0.4]
     histogram_def = prom_metrics._offloading_metric_defs[LOOKUP_LATENCY]
     assert histogram_def.kwargs["buckets"] == (0.1, 1.0)
 
 
+def test_prom_metrics_lazily_observes_labeled_metric():
+    metric_definitions = {
+        MY_COUNTER: OffloadingCounterMetadata(
+            documentation="counter with a label",
+            labelnames=(MY_LABEL,),
+        ),
+    }
+    with patch.object(
+        OffloadingSpecFactory,
+        "get_spec_cls",
+        return_value=_spec_cls_with_metric_definitions(metric_definitions),
+    ):
+        prom_metrics = OffloadPromMetrics(
+            vllm_config=_FakeVllmConfig(store_threshold=0),  # type: ignore[arg-type]
+            metric_types={
+                Gauge: _FakeMetric,
+                Counter: _FakeMetric,
+                Histogram: _FakeMetric,
+            },
+            labelnames=["model_name", "engine"],
+            per_engine_labelvalues={0: ["model", "0"]},
+        )
+
+    assert (0, MY_COUNTER, ("a",)) not in prom_metrics.offloading_metrics
+
+    prom_metrics.observe(
+        {
+            _StatsKey.TYPES: {MY_COUNTER: _MetricType.COUNTER},
+            _StatsKey.DATA: {MY_COUNTER: {("a",): 7}},
+        }
+    )
+
+    counter = prom_metrics.offloading_metrics[(0, MY_COUNTER, ("a",))]
+    assert counter.increments == [7]
+    assert counter.labelvalues == ("model", "0", "a")
+    counter_def = prom_metrics._offloading_metric_defs[MY_COUNTER]
+    assert counter_def.kwargs["labelnames"] == ["model_name", "engine", MY_LABEL]
+
+
+def test_prom_metrics_rejects_wrong_label_count():
+    metric_definitions = {
+        MY_COUNTER: OffloadingCounterMetadata(
+            documentation="counter with a label",
+            labelnames=(MY_LABEL,),
+        ),
+    }
+    with patch.object(
+        OffloadingSpecFactory,
+        "get_spec_cls",
+        return_value=_spec_cls_with_metric_definitions(metric_definitions),
+    ):
+        prom_metrics = OffloadPromMetrics(
+            vllm_config=_FakeVllmConfig(store_threshold=0),  # type: ignore[arg-type]
+            metric_types={
+                Gauge: _FakeMetric,
+                Counter: _FakeMetric,
+                Histogram: _FakeMetric,
+            },
+            labelnames=["model_name", "engine"],
+            per_engine_labelvalues={0: ["model", "0"]},
+        )
+
+    with pytest.raises(AssertionError, match="expects 1 labels"):
+        prom_metrics.observe(
+            {
+                _StatsKey.TYPES: {MY_COUNTER: _MetricType.COUNTER},
+                _StatsKey.DATA: {MY_COUNTER: {("a", "extra"): 7}},
+            }
+        )
+
+
 def test_prom_metrics_uses_configured_manager_metrics():
     prom_metrics = OffloadPromMetrics(
         vllm_config=_FakeVllmConfig(store_threshold=0),  # type: ignore[arg-type]
@@ -458,9 +654,9 @@ def test_aggregate_into_empty_stats():
                 PENDING_STORES: _MetricType.GAUGE,
             },
             _StatsKey.DATA: {
-                LOAD_BYTES: 42,
-                LOAD_SIZE: [10, 20],
-                PENDING_STORES: 3,
+                LOAD_BYTES: {(): 42},
+                LOAD_SIZE: {(): [10, 20]},
+                PENDING_STORES: {(): 3},
             },
         },
     )
@@ -469,9 +665,9 @@ def test_aggregate_into_empty_stats():
 
     assert result is empty
     values = result.data[_StatsKey.DATA]
-    assert values[LOAD_BYTES] == 42
-    assert values[LOAD_SIZE] == [10, 20]
-    assert values[PENDING_STORES] == 3
+    assert _unlabeled(values, LOAD_BYTES) == 42
+    assert _unlabeled(values, LOAD_SIZE) == [10, 20]
+    assert _unlabeled(values, PENDING_STORES) == 3
 
 
 def test_prom_metrics_multi_engine_routing():
@@ -490,14 +686,13 @@ def test_prom_metrics_multi_engine_routing():
     prom_metrics.observe(
         {
             _StatsKey.TYPES: {LOAD_BYTES: _MetricType.COUNTER},
-            _StatsKey.DATA: {LOAD_BYTES: 100},
+            _StatsKey.DATA: {LOAD_BYTES: {(): 100}},
         },
         engine_idx=1,
     )
 
-    engine0 = prom_metrics.offloading_metrics[(0, LOAD_BYTES)]
-    engine1 = prom_metrics.offloading_metrics[(1, LOAD_BYTES)]
-    assert engine0.increments == []
+    assert (0, LOAD_BYTES, ()) not in prom_metrics.offloading_metrics
+    engine1 = prom_metrics.offloading_metrics[(1, LOAD_BYTES, ())]
     assert engine1.increments == [100]
 
 
@@ -518,6 +713,6 @@ def test_prom_metrics_rejects_undeclared_metric():
         prom_metrics.observe(
             {
                 _StatsKey.TYPES: {"unknown:metric": _MetricType.COUNTER},
-                _StatsKey.DATA: {"unknown:metric": 1},
+                _StatsKey.DATA: {"unknown:metric": {(): 1}},
             }
         )

tests/v1/kv_connector/unit/offloading_connector/test_scheduler.py

@@ -1,6 +1,6 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-from collections.abc import Iterable
+from types import SimpleNamespace
 from unittest.mock import MagicMock
 
 import pytest
@@ -11,19 +11,16 @@ from tests.v1.kv_connector.unit.offloading_connector.utils import (
     to_keys,
 )
 from tests.v1.kv_connector.unit.utils import EOS_TOKEN_ID
-from vllm.distributed.kv_events import BlockRemoved, BlockStored
 from vllm.distributed.kv_transfer.kv_connector.v1.offloading.scheduler import (
     OffloadingConnectorScheduler,
     RequestOffloadState,
 )
-from vllm.v1.core.kv_cache_utils import BlockHash
 from vllm.v1.kv_cache_interface import (
     FullAttentionSpec,
     KVCacheGroupSpec,
     SlidingWindowSpec,
 )
 from vllm.v1.kv_offload.base import (
-    OffloadingEvent,
     OffloadingManager,
     OffloadPolicy,
     ReqContext,
@@ -145,31 +142,6 @@ def test_offloading_connector(request_runner, async_scheduling: bool):
     runner.connector_scheduler._maximal_prefix_lookup = lambda key, req_context: 1
     runner.run(decoded_tokens=[EOS_TOKEN_ID], expected_loaded=(3, 4, 5))
 
-    # test take_events
-    def to_hashes(int_hashes: list[int]) -> list[BlockHash]:
-        return [BlockHash(str(i).encode()) for i in int_hashes]
-
-    def take_events() -> Iterable[OffloadingEvent]:
-        yield OffloadingEvent(keys=to_keys([1, 2, 3]), medium="A", removed=False)
-        yield OffloadingEvent(keys=to_keys([4, 5, 6]), medium="B", removed=True)
-
-    runner.manager.take_events.side_effect = take_events
-    events = list(runner.scheduler_connector.take_events())
-    assert len(events) == 2
-    event = events[0]
-    assert isinstance(event, BlockStored)
-    assert event.block_hashes == to_hashes([1, 2, 3])
-    assert event.block_size == 0
-    assert event.medium == "A"
-    assert event.token_ids == []
-    assert event.parent_block_hash is None
-    assert event.lora_id is None
-    assert event.lora_name is None
-    event = events[1]
-    assert isinstance(event, BlockRemoved)
-    assert event.block_hashes == to_hashes([4, 5, 6])
-    assert event.medium == "B"
-
 
 @pytest.mark.parametrize("async_scheduling", [True, False])
 def test_request_preemption(request_runner, async_scheduling: bool):
@@ -1278,11 +1250,11 @@ def test_reset_cache_finalizes_finished_request_with_pending_store(
     )
 
     finalized: list[str] = []
-    runner.manager.on_request_finished.side_effect = (
-        lambda req_context: finalized.append(req_context.req_id)
+    runner.manager.on_request_finished.side_effect = lambda req_context: (
+        finalized.append(req_context.req_id)
     )
-    runner.manager.prepare_store.side_effect = (
-        lambda keys, req_context: generate_store_output(keys)
+    runner.manager.prepare_store.side_effect = lambda keys, req_context: (
+        generate_store_output(keys)
     )
 
     # Decode a couple of blocks and keep every transfer in flight, so the
@@ -1314,6 +1286,100 @@ def test_reset_cache_finalizes_finished_request_with_pending_store(
     assert req_id not in cs._req_status
 
 
+def test_pending_transfer_defers_prefix_lookup():
+    """A request with an in-flight store must not issue a load on re-admission.
+
+    With async scheduling, a preempted request's store can be flushed by the
+    worker before the scheduler consumes its completion. If the request is
+    re-admitted in that window, the connector should defer it instead of
+    looking up offloaded blocks and later asserting when a load is queued while
+    the store job is still tracked.
+    """
+    scheduler = object.__new__(OffloadingConnectorScheduler)
+    scheduler.manager = MagicMock(spec=OffloadingManager)
+
+    request = SimpleNamespace(request_id="req-0")
+    group_state = SimpleNamespace(block_ids=[1, 2, 3])
+    req_status = SimpleNamespace(
+        group_states=[group_state],
+        transfer_jobs={123},
+    )
+    scheduler._req_status = {request.request_id: req_status}
+
+    matched_tokens, is_async = scheduler.get_num_new_matched_tokens(
+        request,
+        num_computed_tokens=0,
+    )
+
+    assert matched_tokens is None
+    assert is_async is False
+    assert group_state.block_ids == []
+    scheduler.manager.lookup.assert_not_called()
+
+
+def test_async_preempt_readmit_before_transfer_output_is_deferred(request_runner):
+    """A preempted request can be scheduled again before flush output is read.
+
+    EngineCore.step_with_batch_queue() may schedule a new batch while a prior
+    preemption batch is still queued. The store completion from jobs_to_flush is
+    only cleared when that queued output reaches update_from_output(), so the
+    re-admission path must defer while the scheduler still tracks the store.
+    """
+    block_size = 4
+    block_size_factor = 3
+    offloaded_block_size = block_size * block_size_factor
+
+    runner = request_runner(
+        block_size=block_size,
+        num_gpu_blocks=100,
+        async_scheduling=True,
+        block_size_factor=block_size_factor,
+    )
+    free_block_queue = runner.scheduler.kv_cache_manager.block_pool.free_block_queue
+    num_free_blocks_empty = free_block_queue.num_free_blocks
+
+    req_id = "0"
+    runner.new_request(token_ids=[0] * offloaded_block_size * 2)
+    runner.manager.prepare_store.side_effect = lambda keys, req_context: (
+        generate_store_output(keys)
+    )
+
+    runner.run(decoded_tokens=[0], complete_transfers=False)
+    runner.run(
+        decoded_tokens=[0] * (2 * offloaded_block_size - block_size),
+        complete_transfers=False,
+    )
+
+    req_status = runner.connector_scheduler._req_status[req_id]
+    pending_store_jobs = set(req_status.transfer_jobs)
+    assert pending_store_jobs
+    assert all(
+        runner.connector_scheduler._jobs[jid].is_store for jid in pending_store_jobs
+    )
+
+    free_block_queue.num_free_blocks = 0
+    preempt_output = runner.scheduler.schedule()
+    assert preempt_output.preempted_req_ids == {req_id}
+    assert preempt_output.kv_connector_metadata is not None
+    assert pending_store_jobs <= preempt_output.kv_connector_metadata.jobs_to_flush
+    assert req_status.transfer_jobs == pending_store_jobs
+
+    # Simulate the async batch-queue window: schedule again before the
+    # preemption batch's ModelRunnerOutput is consumed by update_from_output().
+    free_block_queue.num_free_blocks = num_free_blocks_empty
+    assert runner.scheduler.reset_prefix_cache()
+    runner.connector_scheduler._maximal_prefix_lookup = lambda key, req_context: len(
+        key
+    )
+
+    readmit_output = runner.scheduler.schedule()
+
+    assert readmit_output.num_scheduled_tokens == {}
+    assert readmit_output.kv_connector_metadata is not None
+    assert readmit_output.kv_connector_metadata.load_jobs == {}
+    assert req_status.transfer_jobs == pending_store_jobs
+
+
 @pytest.mark.parametrize("async_scheduling", [True, False])
 def test_swa_alignment_skip(request_runner, async_scheduling: bool):
     """SWA blocks unreachable by the load path are skipped during store.

tests/v1/kv_connector/unit/offloading_connector/utils.py

@@ -14,7 +14,12 @@ from tests.v1.kv_connector.unit.utils import (
     create_vllm_config,
 )
 from vllm import SamplingParams
-from vllm.config import KVTransferConfig, VllmConfig, set_current_vllm_config
+from vllm.config import (
+    KVEventsConfig,
+    KVTransferConfig,
+    VllmConfig,
+    set_current_vllm_config,
+)
 from vllm.distributed.kv_transfer.kv_connector.v1 import KVConnectorRole
 from vllm.distributed.kv_transfer.kv_connector.v1.offloading.common import (
     OffloadingConnectorMetadata,
@@ -198,6 +203,9 @@ class RequestRunner:
             "spec_module_path": "tests.v1.kv_connector.unit.offloading_connector.utils",  # noqa: E501
             # Preserve legacy behavior for tests; new opt-in tests override.
             "offload_prompt_only": False,
+            # Exercise the self-describing KV events path by default;
+            # opt-out tests override this to cover the legacy placeholders.
+            "self_describing_kv_events": True,
         }
         if block_size_factor > 1:
             extra_config["block_size"] = block_size * block_size_factor
@@ -209,6 +217,13 @@ class RequestRunner:
             kv_role="kv_both",
             kv_connector_extra_config=extra_config,
         )
+        vllm_config.kv_events_config = KVEventsConfig(
+            # Enable so the offloading events tracker is active, but use the
+            # null publisher: these tests drain take_events directly and a
+            # real ZMQ publisher would bind a port per test.
+            enable_kv_cache_events=True,
+            publisher="null",
+        )
 
         if kv_cache_groups is None:
             kv_cache_groups = [

tests/v1/kv_connector/unit/test_mooncake_store_coordinator.py

@@ -7,7 +7,10 @@ from vllm.distributed.kv_transfer.kv_connector.v1.mooncake.store.coordinator imp
     ExternalCachedBlockPool,
     MooncakeStoreCoordinator,
 )
-from vllm.v1.core.kv_cache_utils import BlockHash, BlockHashListWithBlockSize
+from vllm.distributed.kv_transfer.kv_connector.v1.mooncake.store.data import (
+    chunk_hashes_for_block_size,
+)
+from vllm.v1.core.kv_cache_utils import BlockHash
 from vllm.v1.kv_cache_interface import (
     FullAttentionSpec,
     KVCacheGroupSpec,
@@ -182,7 +185,7 @@ def test_coordinator_group_block_size_double_hash():
     ]
     coord = _make_coord(groups, hash_block_size=16)
     hs = _hashes(4)
-    big_hashes = list(BlockHashListWithBlockSize(hs, 16, 32))
+    big_hashes = list(chunk_hashes_for_block_size(hs, 16, 32))
     exists = {(0, bytes(h)) for h in hs}
     exists |= {(1, bytes(bh)) for bh in big_hashes}
     cmap = ExternalCachedBlockPool(exists)

tests/v1/kv_connector/unit/test_mooncake_store_hma_e2e.py

@@ -323,8 +323,8 @@ def test_recv_skips_swa_blocks_before_window():
 
 def test_chunked_token_database_hash_block_size_smaller_than_block_size():
     """DSv4-style: hash_block_size=4, group block_size=16 — process_tokens
-    must merge every 4 fine hashes into one chunk hash via
-    BlockHashListWithBlockSize."""
+    keys each 16-token chunk by its last fine hash, keeping the Mooncake key
+    at one digest instead of concatenating all 4 fine hashes."""
     md = KeyMetadata("m", 0, 0, 0, 0, group_id=3)
     db = ChunkedTokenDatabase(md, block_size=16, hash_block_size=4)
     db.set_kv_caches_base_addr([0])
@@ -335,8 +335,7 @@ def test_chunked_token_database_hash_block_size_smaller_than_block_size():
     assert len(out) == 2
     assert out[0][0] == 0 and out[0][1] == 16
     assert out[1][0] == 16 and out[1][1] == 32
-    # Each chunk's hash is the concatenation of 4 fine hashes.
-    expected0 = b"".join(fine_hashes[0:4]).hex()
-    expected1 = b"".join(fine_hashes[4:8]).hex()
-    assert out[0][2].chunk_hash == expected0
-    assert out[1][2].chunk_hash == expected1
+    # Each chunk's hash is its last (4th) fine hash, which already chains the
+    # prior three.
+    assert out[0][2].chunk_hash == fine_hashes[3].hex()
+    assert out[1][2].chunk_hash == fine_hashes[7].hex()

tests/v1/kv_connector/unit/test_mooncake_store_worker.py

@@ -23,6 +23,7 @@ from vllm.distributed.kv_transfer.kv_connector.v1.mooncake.store import (
     worker as mooncake_store_worker,
 )
 from vllm.distributed.kv_transfer.kv_connector.v1.mooncake.store.data import (
+    BlobBlockHashes,
     ChunkedTokenDatabase,
     KeyMetadata,
     LoadSpec,
@@ -32,6 +33,7 @@ from vllm.distributed.kv_transfer.kv_connector.v1.mooncake.store.data import (
 from vllm.distributed.kv_transfer.kv_connector.v1.mooncake.store.metrics import (
     MooncakeStoreConnectorStats,
 )
+from vllm.v1.core.kv_cache_utils import BlockHash
 
 
 def _default_send_coord() -> mooncake_store_worker.MooncakeStoreCoordinator:
@@ -1179,9 +1181,9 @@ def test_store_sending_thread_kv_events_use_group_chunk_metadata():
     assert full_event.group_idx == 0
     assert full_event.block_size == 32
     assert full_event.token_ids == list(range(32))
-    assert full_event.block_hashes == [
-        maybe_convert_block_hash(BlockHash(b"".join(hs)))
-    ]
+    # block_size=32 over hash_block_size=8 (scale 4): the chunk is keyed by its
+    # last sub-hash, not the concatenation of all four.
+    assert full_event.block_hashes == [maybe_convert_block_hash(BlockHash(hs[3]))]
 
     assert swa_event.group_idx == 1
     assert swa_event.block_size == 8
@@ -1749,3 +1751,33 @@ def test_store_worker_close_swallows_store_errors():
     worker.close()
 
     assert worker.store is None
+
+
+def test_blob_block_hashes_wire_roundtrip():
+    """The lookup wire format sends a ``hash_len`` frame plus the raw hashes
+    concatenated back-to-back; the server rebuilds them through a zero-copy
+    ``BlobBlockHashes`` view over the frame buffer."""
+    hashes = [BlockHash(bytes([i]) * 16) for i in range(5)]
+    hash_len = len(hashes[0])
+
+    # Client side (LookupKeyClient._lookup): flat payload frame.
+    blob = b"".join(hashes)
+
+    # Server side (LookupKeyServer): view over the frame buffer (a memoryview),
+    # never materializing the full hash list upfront.
+    view = BlobBlockHashes(memoryview(blob), hash_len)
+
+    assert len(view) == 5
+    assert list(view) == hashes  # default Sequence iter terminates via IndexError
+    assert [bytes(h) for h in view] == hashes
+    assert bytes(view[-1]) == hashes[-1]
+    assert [bytes(h) for h in view[1:3]] == hashes[1:3]
+    with pytest.raises(IndexError):
+        _ = view[5]
+
+
+def test_blob_block_hashes_empty():
+    """Empty lookups send hash_len=0 and an empty payload."""
+    view = BlobBlockHashes(memoryview(b""), 0)
+    assert len(view) == 0
+    assert list(view) == []

tests/v1/kv_connector/unit/test_moriio_connector.py

@@ -36,13 +36,33 @@ from vllm.utils.network_utils import (
     get_ip,
     make_zmq_path,
 )
-from vllm.v1.kv_cache_interface import KVCacheConfig
+from vllm.v1.kv_cache_interface import (
+    FullAttentionSpec,
+    KVCacheConfig,
+    KVCacheGroupSpec,
+    KVCacheTensor,
+)
 
 from .utils import create_request, create_scheduler
 
 
 def _make_test_kv_cache_config() -> KVCacheConfig:
-    return KVCacheConfig(num_blocks=0, kv_cache_tensors=[], kv_cache_groups=[])
+    layer_names = ["layer0", "layer1", "layer2"]
+    return KVCacheConfig(
+        num_blocks=2,
+        kv_cache_tensors=[KVCacheTensor(size=0, shared_by=layer_names)],
+        kv_cache_groups=[
+            KVCacheGroupSpec(
+                layer_names=layer_names,
+                kv_cache_spec=FullAttentionSpec(
+                    block_size=16,
+                    num_kv_heads=4,
+                    head_size=64,
+                    dtype=torch.float16,
+                ),
+            )
+        ],
+    )
 
 
 aiter_available = importlib.util.find_spec("aiter") is not None
@@ -175,9 +195,18 @@ class FakeMoRIIOConnectorWorker(MoRIIOConnectorWorker):
     REMOTE_ENGINE_ID = "remote_engine"
 
     def __init__(
-        self, *args, hand_shake_latency: float = 1.8, kv_cache_layout="HND", **kwargs
+        self,
+        vllm_config,
+        engine_id,
+        *args,
+        hand_shake_latency: float = 1.8,
+        kv_cache_layout="HND",
+        kv_cache_config=None,
+        **kwargs,
     ):
-        super().__init__(*args, **kwargs)
+        super().__init__(
+            vllm_config, engine_id, kv_cache_config or _make_test_kv_cache_config()
+        )
 
 
 def create_vllm_config(

tests/v1/kv_connector/unit/test_moriio_kv_layout.py

@@ -0,0 +1,228 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+import importlib.util
+from types import SimpleNamespace
+
+import pytest
+import torch
+
+from vllm.platforms import current_platform
+from vllm.v1.kv_cache_interface import FullAttentionSpec, MLAAttentionSpec
+
+aiter_available = importlib.util.find_spec("aiter") is not None
+mori_available = importlib.util.find_spec("mori") is not None
+
+if not (current_platform.is_rocm() and mori_available):
+    pytest.skip(
+        "MoRIIOs are only available on ROCm with mori package installed",
+        allow_module_level=True,
+    )
+
+moriio_layout = importlib.import_module(
+    "vllm.distributed.kv_transfer.kv_connector.v1.moriio.moriio_layout"
+)
+
+
+def _full_spec(block_size: int = 4) -> FullAttentionSpec:
+    return FullAttentionSpec(
+        block_size=block_size,
+        num_kv_heads=2,
+        head_size=3,
+        dtype=torch.bfloat16,
+    )
+
+
+def _mla_spec(block_size: int = 4) -> MLAAttentionSpec:
+    return MLAAttentionSpec(
+        block_size=block_size,
+        num_kv_heads=1,
+        head_size=3,
+        dtype=torch.bfloat16,
+    )
+
+
+def _worker(
+    kv_caches: dict[str, torch.Tensor],
+    layer_to_spec: dict[str, object],
+    num_blocks: int = 8,
+) -> SimpleNamespace:
+    return SimpleNamespace(
+        kv_caches=kv_caches,
+        layer_to_spec=layer_to_spec,
+        num_blocks=num_blocks,
+        block_size=4,
+    )
+
+
+def _remote_meta(num_blocks: int = 16) -> SimpleNamespace:
+    return SimpleNamespace(num_blocks=num_blocks)
+
+
+def test_separated_kv_layout_uses_kv_axis_zero_and_block_axis_one():
+    cache = torch.empty((2, 8, 4, 2, 3), dtype=torch.bfloat16)
+    worker = _worker({"layer": cache}, {"layer": _full_spec()})
+
+    geometry = moriio_layout.get_layer_transfer_geometry(
+        "layer", cache, worker.layer_to_spec, remote_num_blocks=16
+    )
+    assert geometry.block_stride == 24
+    assert geometry.local_kv_stride == 192
+    assert geometry.remote_kv_stride == 384
+    assert geometry.split_kv_regions
+
+    assert moriio_layout.compute_block_transfer_offsets(
+        "layer", cache, worker.layer_to_spec, [1, 3], [4, 5], _remote_meta().num_blocks
+    ) == ([48, 144, 432, 528], [192, 240, 960, 1008], [48, 48, 48, 48])
+
+
+def test_interleaved_kv_layout_uses_block_axis_zero_and_kv_axis_one():
+    cache = torch.empty((8, 2, 4, 2, 3), dtype=torch.bfloat16)
+    worker = _worker({"layer": cache}, {"layer": _full_spec()})
+
+    geometry = moriio_layout.get_layer_transfer_geometry(
+        "layer", cache, worker.layer_to_spec, remote_num_blocks=16
+    )
+    assert geometry.block_stride == 48
+    assert geometry.local_kv_stride == 24
+    assert geometry.remote_kv_stride == 24
+    assert not geometry.split_kv_regions
+
+    assert moriio_layout.compute_block_transfer_offsets(
+        "layer", cache, worker.layer_to_spec, [1, 3], [4, 5], _remote_meta().num_blocks
+    ) == ([96, 288], [384, 480], [96, 96])
+
+
+def test_mla_key_only_layout_transfers_one_slab_per_block():
+    cache = torch.empty((8, 4, 3), dtype=torch.bfloat16)
+    worker = _worker({"layer": cache}, {"layer": _mla_spec()})
+
+    geometry = moriio_layout.get_layer_transfer_geometry(
+        "layer", cache, worker.layer_to_spec, remote_num_blocks=16
+    )
+    assert geometry.block_stride == 12
+    assert geometry.local_kv_stride is None
+    assert geometry.remote_kv_stride is None
+    assert geometry.transfers_per_block == 1
+
+    assert moriio_layout.compute_block_transfer_offsets(
+        "layer", cache, worker.layer_to_spec, [1, 3], [4, 5], _remote_meta().num_blocks
+    ) == ([24, 72], [96, 120], [24, 24])
+
+
+def test_mixed_layers_compute_distinct_offsets_per_layer():
+    kv_caches = {
+        "separated": torch.empty((2, 8, 4, 2, 3), dtype=torch.bfloat16),
+        "interleaved": torch.empty((8, 2, 4, 2, 3), dtype=torch.bfloat16),
+        "indexer": torch.empty((8, 4, 3), dtype=torch.bfloat16),
+    }
+    worker = _worker(
+        kv_caches,
+        {
+            "separated": _full_spec(),
+            "interleaved": _full_spec(),
+            "indexer": _mla_spec(),
+        },
+    )
+
+    separated = moriio_layout.compute_block_transfer_offsets(
+        "separated",
+        kv_caches["separated"],
+        worker.layer_to_spec,
+        [1, 3],
+        [4, 5],
+        _remote_meta().num_blocks,
+    )
+    interleaved = moriio_layout.compute_block_transfer_offsets(
+        "interleaved",
+        kv_caches["interleaved"],
+        worker.layer_to_spec,
+        [1, 3],
+        [4, 5],
+        _remote_meta().num_blocks,
+    )
+    indexer = moriio_layout.compute_block_transfer_offsets(
+        "indexer",
+        kv_caches["indexer"],
+        worker.layer_to_spec,
+        [1, 3],
+        [4, 5],
+        _remote_meta().num_blocks,
+    )
+
+    assert separated != interleaved
+    assert separated != indexer
+    assert interleaved != indexer
+
+
+def test_block_id_length_mismatch_raises_value_error():
+    cache = torch.empty((8, 2, 4, 2, 3), dtype=torch.bfloat16)
+    worker = _worker({"layer": cache}, {"layer": _full_spec()})
+
+    with pytest.raises(ValueError, match="must have the same length"):
+        moriio_layout.compute_block_transfer_offsets(
+            "layer", cache, worker.layer_to_spec, [1, 3], [4], _remote_meta().num_blocks
+        )
+
+
+def test_registration_regions_do_not_split_interleaved_or_mla_cache():
+    separated = torch.empty((2, 8, 4, 2, 3), dtype=torch.bfloat16)
+    interleaved = torch.empty((8, 2, 4, 2, 3), dtype=torch.bfloat16)
+    indexer = torch.empty((8, 4, 3), dtype=torch.bfloat16)
+    worker = _worker(
+        {
+            "separated": separated,
+            "interleaved": interleaved,
+            "indexer": indexer,
+        },
+        {
+            "separated": _full_spec(),
+            "interleaved": _full_spec(),
+            "indexer": _mla_spec(),
+        },
+    )
+
+    separated_regions = moriio_layout.iter_layer_registration_regions(
+        "separated", separated, worker.layer_to_spec
+    )
+    interleaved_regions = moriio_layout.iter_layer_registration_regions(
+        "interleaved", interleaved, worker.layer_to_spec
+    )
+    indexer_regions = moriio_layout.iter_layer_registration_regions(
+        "indexer", indexer, worker.layer_to_spec
+    )
+
+    assert [region[0].data_ptr() for region in separated_regions] == [
+        separated[0].data_ptr(),
+        separated[1].data_ptr(),
+    ]
+    assert separated_regions[0][1] == 8 * 48
+    assert separated_regions[1][1] == 8 * 48
+
+    assert len(interleaved_regions) == 1
+    assert interleaved_regions[0][0].data_ptr() == interleaved.data_ptr()
+    assert interleaved_regions[0][1] == 8 * 2 * 48
+
+    assert len(indexer_regions) == 1
+    assert indexer_regions[0][0].data_ptr() == indexer.data_ptr()
+    assert indexer_regions[0][1] == 8 * 24
+
+
+def test_registration_regions_use_layer_num_blocks():
+    cache = torch.empty((4, 2, 4, 2, 3), dtype=torch.bfloat16)
+    worker = _worker({"layer": cache}, {"layer": _full_spec()}, num_blocks=8)
+
+    regions = moriio_layout.iter_layer_registration_regions(
+        "layer", cache, worker.layer_to_spec
+    )
+
+    assert len(regions) == 1
+    assert regions[0][1] == 4 * 2 * 48
+
+
+def test_unsupported_shape_raises_value_error():
+    cache = torch.empty((8, 4, 2, 3), dtype=torch.bfloat16)
+    worker = _worker({"layer": cache}, {"layer": _full_spec()})
+
+    with pytest.raises(ValueError, match="Unsupported MoRIIO K/V cache shape"):
+        moriio_layout.get_layer_transfer_geometry("layer", cache, worker.layer_to_spec)

tests/v1/kv_connector/unit/test_multi_connector.py

@@ -219,6 +219,7 @@ def test_multi_example_connector_consistency():
         enforce_eager=True,
         gpu_memory_utilization=0.5,
         kv_transfer_config=kv_transfer_config,
+        async_scheduling=False,
     )
     # Run generation - this should trigger saving KV cache
     # Use a single prompt to avoid race conditions depending on the order of scheduling

tests/v1/kv_connector/unit/test_offloading_connector.py

@@ -138,6 +138,10 @@ def _wait_for_prefix_cache_reset(llm: LLM) -> None:
 
 
 def _latency_test(llm: LLM, subscriber: MockSubscriber | None):
+    # TODO: Reintroduce latency test on ROCm once MRV2 supports cross
+    # layer KV Cache. See https://github.com/vllm-project/vllm/pull/45947
+    if current_platform.is_rocm():
+        return
     sampling_params = SamplingParams(max_tokens=1)
 
     num_times_cpu_better_than_cold = 0

tests/v1/kv_offload/cpu/test_manager.py

@@ -14,12 +14,13 @@ from vllm.v1.kv_offload.base import (
     ReqContext,
     make_offload_key,
 )
-from vllm.v1.kv_offload.cpu.common import CPULoadStoreSpec
+from vllm.v1.kv_offload.cpu.common import (
+    CPULoadStoreSpec,
+    CPUOffloadingMetrics,
+)
 from vllm.v1.kv_offload.cpu.manager import CPUOffloadingManager
 from vllm.v1.kv_offload.cpu.policies.arc import ARCCachePolicy
 
-STORES_SKIPPED = "vllm:kv_offload_stores_skipped"
-
 
 def make_req_context(
     req_id: str = "", kv_transfer_params: dict | None = None
@@ -181,10 +182,45 @@ def test_filter_reused_manager_reports_stores_skipped_counter():
     )
     stats = manager.get_stats()
     assert stats is not None
-    assert stats.reduce()[STORES_SKIPPED] == 3
+    assert stats.reduce()[CPUOffloadingMetrics.STORES_SKIPPED] == 3
     stats = manager.get_stats()
     assert stats is not None
-    assert stats.reduce()[STORES_SKIPPED] == 0
+    assert stats.reduce()[CPUOffloadingMetrics.STORES_SKIPPED] == 0
+
+
+def test_cpu_manager_reports_cache_usage_gauge():
+    def check_usage_stats(manager: CPUOffloadingManager, value: float):
+        stats = manager.get_stats()
+        assert stats is not None
+        assert stats.reduce()[
+            CPUOffloadingMetrics.CPU_CACHE_USAGE_PERC
+        ] == pytest.approx(value)
+
+    # Zero-capacity manager always reports 0.0
+    manager = make_cpu_manager(num_blocks=0)
+    check_usage_stats(manager, 0.0)
+
+    # Empty manager (4 blocks, none allocated): usage = 0.0
+    manager = make_cpu_manager(num_blocks=4)
+    check_usage_stats(manager, 0.0)
+
+    # After allocating 2 of 4 blocks: usage = 0.5
+    manager.prepare_store(to_keys([1, 2]), _EMPTY_REQ_CTX)
+    check_usage_stats(manager, 0.5)
+
+    # After filling all 4 blocks: usage = 1.0
+    manager.prepare_store(to_keys([3, 4]), _EMPTY_REQ_CTX)
+    check_usage_stats(manager, 1.0)
+
+    # After completing store, the blocks becomes evictable as it is not actively used
+    # and usage drops.
+    manager.complete_store(to_keys([1, 2]), _EMPTY_REQ_CTX)
+    check_usage_stats(manager, 0.5)
+
+    # After completing store, the blocks becomes evictable as it is not actively used
+    # and usage drops.
+    manager.complete_store(to_keys([3, 4]), _EMPTY_REQ_CTX)
+    check_usage_stats(manager, 0.0)
 
 
 def test_cpu_manager():
@@ -258,25 +294,25 @@ def test_cpu_manager():
     # prepare store with no space ([2, 3] is being loaded)
     assert cpu_manager.prepare_store(to_keys([6, 7, 8]), _EMPTY_REQ_CTX) is None
 
-    # complete load [2, 3]
+    # complete load [2, 3]. Load changes the eviction list, making 2, 3 recent.
     cpu_manager.complete_load(to_keys([2, 3]), _EMPTY_REQ_CTX)
 
-    # prepare store [6, 7, 8] -> evicts [2, 3, 4] (oldest)
+    # prepare store [6, 7, 8] -> evicts [4, 5, 2] (oldest)
     prepare_store_output = cpu_manager.prepare_store(to_keys([6, 7, 8]), _EMPTY_REQ_CTX)
     verify_store_output(
         prepare_store_output,
         ExpectedPrepareStoreOutput(
             keys_to_store=[6, 7, 8],
-            store_block_ids=[3, 2, 1],
-            evicted_keys=[2, 3, 4],
+            store_block_ids=[1, 0, 3],
+            evicted_keys=[4, 5, 2],
         ),
     )
 
     # complete store [6, 7, 8]
     cpu_manager.complete_store(to_keys([6, 7, 8]), _EMPTY_REQ_CTX)
 
-    # touch [5, 6, 7] (move to end of LRU order)
-    cpu_manager.touch(to_keys([5, 6, 7]), _EMPTY_REQ_CTX)
+    # touch [3, 6, 7] (move to end of LRU order)
+    cpu_manager.touch(to_keys([3, 6, 7]), _EMPTY_REQ_CTX)
 
     # prepare store [7, 9] -> evicts [8] (oldest following previous touch)
     prepare_store_output = cpu_manager.prepare_store(to_keys([9]), _EMPTY_REQ_CTX)
@@ -284,7 +320,7 @@ def test_cpu_manager():
         prepare_store_output,
         ExpectedPrepareStoreOutput(
             keys_to_store=[9],
-            store_block_ids=[1],
+            store_block_ids=[3],
             evicted_keys=[8],
         ),
     )
@@ -299,7 +335,7 @@ def test_cpu_manager():
     verify_events(
         cpu_manager.take_events(),
         expected_stores=({3, 4, 5}, {6, 7, 8}),
-        expected_evictions=({2, 3, 4}, {8}),
+        expected_evictions=({4, 5, 2}, {8}),
     )
 
 

tests/v1/kv_offload/tiering/test_tiering_offloading.py

@@ -295,6 +295,8 @@ class TestTieringOffloadingManager:
         self.manager.prepare_store(blocks, _CTX)
         self.manager.complete_store(blocks, _CTX, success=True)
         self._simulate_on_schedule_end()
+        # for secondary tiers to drain jobs, so primary tier's blocks are evictable.
+        self._simulate_on_schedule_end()
 
         self.secondary_tier1.touch = MagicMock(wraps=self.secondary_tier1.touch)
         self.secondary_tier2.touch = MagicMock(wraps=self.secondary_tier2.touch)
@@ -303,7 +305,7 @@ class TestTieringOffloadingManager:
         self.manager.touch(blocks, _CTX)
 
         # Verify touch was called on primary tier (check LRU order)
-        primary_keys = list(self.primary_tier._policy.blocks.keys())
+        primary_keys = list(self.primary_tier._policy.evictable_blocks.keys())
         assert primary_keys[-3:] == list(reversed(blocks))
 
         # Verify touch was propagated to all secondary tiers