Merge remote-tracking branch 'upstream/main' into upstream_merge_2025_06_04

Author: gshtras

.buildkite/release-pipeline.yaml

@@ -1,5 +1,6 @@
 steps:
   - label: "Build wheel - CUDA 12.8"
+    id: build-wheel-cuda-12-8
     agents:
       queue: cpu_queue_postmerge
     commands:
@@ -11,6 +12,7 @@ steps:
       DOCKER_BUILDKIT: "1"
 
   - label: "Build wheel - CUDA 12.6"
+    id: build-wheel-cuda-12-6
     agents:
       queue: cpu_queue_postmerge
     commands:
@@ -28,6 +30,7 @@ steps:
 
   - label: "Build wheel - CUDA 11.8"
     # depends_on: block-build-cu118-wheel
+    id: build-wheel-cuda-11-8
     agents:
       queue: cpu_queue_postmerge
     commands:
@@ -44,13 +47,26 @@ steps:
 
   - label: "Build release image"
     depends_on: block-release-image-build
+    id: build-release-image
     agents:
       queue: cpu_queue_postmerge
     commands:
       - "aws ecr-public get-login-password --region us-east-1 | docker login --username AWS --password-stdin public.ecr.aws/q9t5s3a7"
       - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --build-arg CUDA_VERSION=12.8.1 --tag public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT --target vllm-openai --progress plain -f docker/Dockerfile ."
       - "docker push public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT"
 
+  - label: "Annotate release workflow"
+    depends_on:
+      - build-release-image
+      - build-wheel-cuda-12-8
+      - build-wheel-cuda-12-6
+      - build-wheel-cuda-11-8
+    id: annotate-release-workflow
+    agents:
+      queue: cpu_queue_postmerge
+    commands:
+      - "bash .buildkite/scripts/annotate-release.sh"
+
   - label: "Build and publish TPU release image"
     depends_on: ~
     if: build.env("NIGHTLY") == "1"
@@ -70,9 +86,10 @@ steps:
       DOCKER_BUILDKIT: "1"
 
   - input: "Provide Release version here"
+    id: input-release-version
     fields:
       - text: "What is the release version?"
-        key: "release-version"
+        key: release-version
 
   - block: "Build CPU release image"
     key: block-cpu-release-image-build

.buildkite/scripts/annotate-release.sh

@@ -0,0 +1,31 @@
+#!/bin/bash
+
+set -ex
+
+# Get release version and strip leading 'v' if present
+RELEASE_VERSION=$(buildkite-agent meta-data get release-version | sed 's/^v//')
+
+if [ -z "$RELEASE_VERSION" ]; then
+  echo "Error: RELEASE_VERSION is empty. 'release-version' metadata might not be set or is invalid."
+  exit 1
+fi
+
+buildkite-agent annotate --style 'info' --context 'release-workflow' << EOF
+To download the wheel:
+\`\`\`
+aws s3 cp s3://vllm-wheels/${RELEASE_VERSION}/vllm-${RELEASE_VERSION}-cp38-abi3-manylinux1_x86_64.whl .
+aws s3 cp s3://vllm-wheels/${RELEASE_VERSION}+cu126/vllm-${RELEASE_VERSION}+cu126-cp38-abi3-manylinux1_x86_64.whl .
+aws s3 cp s3://vllm-wheels/${RELEASE_VERSION}+cu118/vllm-${RELEASE_VERSION}+cu118-cp38-abi3-manylinux1_x86_64.whl . 
+\`\`\`
+
+To download and upload the image:
+
+\`\`\`
+docker pull public.ecr.aws/q9t5s3a7/vllm-release-repo:${BUILDKITE_COMMIT}
+docker tag public.ecr.aws/q9t5s3a7/vllm-release-repo:${BUILDKITE_COMMIT} vllm/vllm-openai
+docker tag vllm/vllm-openai vllm/vllm-openai:latest
+docker tag vllm/vllm-openai vllm/vllm-openai:v${RELEASE_VERSION}
+docker push vllm/vllm-openai:latest
+docker push vllm/vllm-openai:v${RELEASE_VERSION}
+\`\`\`
+EOF 

.buildkite/scripts/hardware_ci/run-tpu-v1-test.sh

@@ -150,7 +150,7 @@ run_and_track_test 9 "test_multimodal.py" \
 run_and_track_test 10 "test_pallas.py" \
     "python3 -m pytest -s -v /workspace/vllm/tests/v1/tpu/test_pallas.py"
 run_and_track_test 11 "test_struct_output_generate.py" \
-    "python3 -m pytest -s -v /workspace/vllm/tests/v1/entrypoints/llm/test_struct_output_generate.py -k 'not test_structured_output_with_reasoning_matrices'"
+    "python3 -m pytest -s -v /workspace/vllm/tests/v1/entrypoints/llm/test_struct_output_generate.py -k \"not test_structured_output_with_reasoning_matrices\""
 run_and_track_test 12 "test_moe_pallas.py" \
     "python3 -m pytest -s -v /workspace/vllm/tests/tpu/test_moe_pallas.py"
 run_and_track_test 13 "test_lora.py" \

.pre-commit-config.yaml

@@ -11,6 +11,8 @@ repos:
   hooks:
   - id: yapf
     args: [--in-place, --verbose]
+    # Keep the same list from yapfignore here to avoid yapf failing without any inputs
+    exclude: '(.buildkite|benchmarks|build|examples)/.*'
 - repo: https://github.com/astral-sh/ruff-pre-commit
   rev: v0.11.7
   hooks:

benchmarks/benchmark_utils.py

@@ -66,4 +66,9 @@ def iterencode(self, o: Any, *args, **kwargs) -> Any:
 
 def write_to_json(filename: str, records: list) -> None:
     with open(filename, "w") as f:
-        json.dump(records, f, cls=InfEncoder)
+        json.dump(
+            records,
+            f,
+            cls=InfEncoder,
+            default=lambda o: f"<{type(o).__name__} object is not JSON serializable>",
+        )

csrc/quantization/cutlass_w8a8/c3x/scaled_mm_sm100_fp8_dispatch.cuh

@@ -15,6 +15,7 @@ using c3x::cutlass_gemm_caller;
 template <typename InType, typename OutType,
           template <typename, typename, typename> typename Epilogue>
 struct sm100_fp8_config_default {
+  // M in (128, inf)
   static_assert(std::is_same<InType, cutlass::float_e4m3_t>());
   using KernelSchedule = cutlass::gemm::collective::KernelScheduleAuto;
   using EpilogueSchedule = cutlass::epilogue::collective::EpilogueScheduleAuto;
@@ -25,6 +26,34 @@ struct sm100_fp8_config_default {
                             KernelSchedule, EpilogueSchedule>;
 };
 
+template <typename InType, typename OutType,
+          template <typename, typename, typename> typename Epilogue>
+struct sm100_fp8_config_M128 {
+  // M in (64, 128]
+  static_assert(std::is_same<InType, cutlass::float_e4m3_t>());
+  using KernelSchedule = cutlass::gemm::collective::KernelScheduleAuto;
+  using EpilogueSchedule = cutlass::epilogue::collective::EpilogueScheduleAuto;
+  using TileShape = Shape<_128, _128, _64>;
+  using ClusterShape = Shape<_2, _2, _1>;
+  using Cutlass3xGemm =
+      cutlass_3x_gemm_sm100<InType, OutType, Epilogue, TileShape, ClusterShape,
+                            KernelSchedule, EpilogueSchedule>;
+};
+
+template <typename InType, typename OutType,
+          template <typename, typename, typename> typename Epilogue>
+struct sm100_fp8_config_M64 {
+  // M in [1, 64]
+  static_assert(std::is_same<InType, cutlass::float_e4m3_t>());
+  using KernelSchedule = cutlass::gemm::collective::KernelScheduleAuto;
+  using EpilogueSchedule = cutlass::epilogue::collective::EpilogueScheduleAuto;
+  using TileShape = Shape<_64, _64, _256>;
+  using ClusterShape = Shape<_1, _8, _1>;
+  using Cutlass3xGemm =
+      cutlass_3x_gemm_sm100<InType, OutType, Epilogue, TileShape, ClusterShape,
+                            KernelSchedule, EpilogueSchedule>;
+};
+
 template <typename InType, typename OutType,
           template <typename, typename, typename> typename Epilogue,
           typename... EpilogueArgs>
@@ -39,8 +68,28 @@ inline void cutlass_gemm_sm100_fp8_dispatch(torch::Tensor& out,
   using Cutlass3xGemmDefault =
       typename sm100_fp8_config_default<InType, OutType,
                                         Epilogue>::Cutlass3xGemm;
-  return cutlass_gemm_caller<Cutlass3xGemmDefault>(
-      out, a, b, std::forward<EpilogueArgs>(args)...);
+  using Cutlass3xGemmM64 =
+      typename sm100_fp8_config_M64<InType, OutType, Epilogue>::Cutlass3xGemm;
+  using Cutlass3xGemmM128 =
+      typename sm100_fp8_config_M128<InType, OutType, Epilogue>::Cutlass3xGemm;
+
+  uint32_t const m = a.size(0);
+  uint32_t const mp2 =
+      std::max(static_cast<uint32_t>(64), next_pow_2(m));  // next power of 2
+
+  if (mp2 <= 64) {
+    // m in [1, 64]
+    return cutlass_gemm_caller<Cutlass3xGemmM64>(
+        out, a, b, std::forward<EpilogueArgs>(args)...);
+  } else if (mp2 <= 128) {
+    // m in (64, 128]
+    return cutlass_gemm_caller<Cutlass3xGemmM128>(
+        out, a, b, std::forward<EpilogueArgs>(args)...);
+  } else {
+    // m in (128, inf)
+    return cutlass_gemm_caller<Cutlass3xGemmDefault>(
+        out, a, b, std::forward<EpilogueArgs>(args)...);
+  }
 }
 
 template <template <typename, typename, typename> typename Epilogue,

docker/Dockerfile.nightly_torch

@@ -312,4 +312,7 @@ RUN --mount=type=cache,target=/root/.cache/uv \
 # Logging to confirm the torch versions
 RUN pip freeze | grep -E 'torch|xformers|vllm|flashinfer'
 
+# Logging to confirm all the packages are installed
+RUN pip freeze
+
 #################### UNITTEST IMAGE #############################

docs/models/extensions/tensorizer.md

@@ -10,7 +10,7 @@ shorter Pod startup times and CPU memory usage. Tensor encryption is also suppor
 
 For more information on CoreWeave's Tensorizer, please refer to
 [CoreWeave's Tensorizer documentation](https://github.com/coreweave/tensorizer). For more information on serializing a vLLM model, as well a general usage guide to using Tensorizer with vLLM, see
-the [vLLM example script](https://docs.vllm.ai/en/latest/examples/tensorize_vllm_model.html).
+the [vLLM example script](https://docs.vllm.ai/en/latest/examples/others/tensorize_vllm_model.html).
 
 !!! note
     Note that to use this feature you will need to install `tensorizer` by running `pip install vllm[tensorizer]`.

examples/online_serving/multi_instance_data_parallel.py

@@ -0,0 +1,58 @@
+# SPDX-License-Identifier: Apache-2.0
+import asyncio
+from typing import Optional
+
+from vllm.engine.arg_utils import AsyncEngineArgs
+from vllm.engine.async_llm_engine import AsyncLLMEngine
+from vllm.outputs import RequestOutput
+from vllm.sampling_params import SamplingParams
+
+"""
+To run this example, run the following commands simultaneously with
+different CUDA_VISIBLE_DEVICES:
+    python examples/online_serving/multi_instance_data_parallel.py
+
+    vllm serve ibm-research/PowerMoE-3b -dp 2 -dpr 1 \
+        --data-parallel-address 127.0.0.1 --data-parallel-rpc-port 62300 \
+        --data-parallel-size-local 1 --enforce-eager --headless
+
+Once both instances have completed the handshake, this example will
+send a request to the instance with DP rank 1.
+"""
+
+
+async def main():
+    engine_args = AsyncEngineArgs(
+        model="ibm-research/PowerMoE-3b",
+        data_parallel_size=2,
+        dtype="auto",
+        max_model_len=2048,
+        data_parallel_address="127.0.0.1",
+        data_parallel_rpc_port=62300,
+        data_parallel_size_local=1,
+        enforce_eager=True,
+    )
+
+    engine_client = AsyncLLMEngine.from_engine_args(engine_args)
+
+    sampling_params = SamplingParams(
+        temperature=0.7,
+        top_p=0.9,
+        max_tokens=100,
+    )
+
+    prompt = "Who won the 2004 World Series?"
+    final_output: Optional[RequestOutput] = None
+    async for output in engine_client.generate(
+        prompt=prompt,
+        sampling_params=sampling_params,
+        request_id="abcdef",
+        data_parallel_rank=1,
+    ):
+        final_output = output
+    if final_output:
+        print(final_output.outputs[0].text)
+
+
+if __name__ == "__main__":
+    asyncio.run(main())

requirements/nightly_torch_test.txt

@@ -9,7 +9,9 @@ pytest-shard
 pytest-timeout
 
 librosa # required by audio tests in entrypoints/openai
-sentence-transformers
+sentence-transformers # required for embedding tests
+transformers==4.51.3
+transformers_stream_generator # required for qwen-vl test
 numba == 0.61.2; python_version > '3.9'
 # testing utils
 boto3
@@ -38,4 +40,7 @@ matplotlib # required for qwen-vl test
 # required for  Multi-Modal Models Test (Standard)
 num2words # required for smolvlm test
 pqdm
-timm # required for internvl test
+timm # required for internvl test
+
+schemathesis>=3.39.15  # Required for openai schema test.
+mteb>=1.38.11, <2 # required for mteb test