Merge branch 'main' into dcp-bugfix

Author: gjc0824

.buildkite/scripts/hardware_ci/run-cpu-test-arm.sh

@@ -7,58 +7,56 @@ set -ex
 # allow to bind to different cores
 CORE_RANGE=${CORE_RANGE:-0-16}
 OMP_CORE_RANGE=${OMP_CORE_RANGE:-0-16}
-NUMA_NODE=${NUMA_NODE:-0}
 
-export CMAKE_BUILD_PARALLEL_LEVEL=32
+export CMAKE_BUILD_PARALLEL_LEVEL=16
 
 # Setup cleanup
 remove_docker_container() {
     set -e;
-    docker rm -f cpu-test-"$NUMA_NODE" || true;
+    docker rm -f cpu-test || true;
 }
 trap remove_docker_container EXIT
 remove_docker_container
 
 # Try building the docker image
-numactl -C "$CORE_RANGE" -N "$NUMA_NODE" docker build --tag cpu-test-"$NUMA_NODE" --target vllm-test -f docker/Dockerfile.cpu .
+docker build --tag cpu-test --target vllm-test -f docker/Dockerfile.cpu .
 
-# Run the image, setting --shm-size=4g for tensor parallel.
-docker run -itd --cpuset-cpus="$CORE_RANGE" --cpuset-mems="$NUMA_NODE" --entrypoint /bin/bash -v ~/.cache/huggingface:/root/.cache/huggingface --privileged=true -e HF_TOKEN --env VLLM_CPU_KVCACHE_SPACE=16 --env VLLM_CPU_CI_ENV=1 -e E2E_OMP_THREADS="$OMP_CORE_RANGE" --shm-size=4g --name cpu-test-"$NUMA_NODE" cpu-test-"$NUMA_NODE"
+# Run the image
+docker run -itd --cpuset-cpus="$CORE_RANGE" --entrypoint /bin/bash -v ~/.cache/huggingface:/root/.cache/huggingface -e HF_TOKEN --env VLLM_CPU_KVCACHE_SPACE=16 --env VLLM_CPU_CI_ENV=1 -e E2E_OMP_THREADS="$OMP_CORE_RANGE" --shm-size=4g --name cpu-test cpu-test
 
 function cpu_tests() {
   set -e
-  export NUMA_NODE=$2
 
-  docker exec cpu-test-"$NUMA_NODE" bash -c "
+  docker exec cpu-test bash -c "
     set -e
     pip list"
 
   # offline inference
-  docker exec cpu-test-"$NUMA_NODE" bash -c "
+  docker exec cpu-test bash -c "
     set -e
     python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m"
 
   # Run kernel tests
-  docker exec cpu-test-"$NUMA_NODE" bash -c "
+  docker exec cpu-test bash -c "
     set -e
     pytest -x -v -s tests/kernels/test_onednn.py
     pytest -x -v -s tests/kernels/attention/test_cpu_attn.py"
 
   # basic online serving
-  docker exec cpu-test-"$NUMA_NODE" bash -c '
+  docker exec cpu-test bash -c '
     set -e
-    VLLM_CPU_OMP_THREADS_BIND=$E2E_OMP_THREADS vllm serve meta-llama/Llama-3.2-3B-Instruct --max-model-len 2048 &
+    VLLM_CPU_OMP_THREADS_BIND=$E2E_OMP_THREADS vllm serve Qwen/Qwen3-0.6B --max-model-len 2048 &
     server_pid=$!
     timeout 600 bash -c "until curl localhost:8000/v1/models; do sleep 1; done" || exit 1
     vllm bench serve \
       --backend vllm \
       --dataset-name random \
-      --model meta-llama/Llama-3.2-3B-Instruct \
+      --model Qwen/Qwen3-0.6B \
       --num-prompts 20 \
       --endpoint /v1/completions
     kill -s SIGTERM $server_pid &'
 }
 
 # All of CPU tests are expected to be finished less than 40 mins.
 export -f cpu_tests
-timeout 2h bash -c "cpu_tests $CORE_RANGE $NUMA_NODE"
+timeout 2h bash -c cpu_tests

.buildkite/test-pipeline.yaml

@@ -632,6 +632,7 @@ steps:
   # we can only upgrade after this is resolved
   # TODO(jerryzh168): resolve the above comment
   - uv pip install --system torchao==0.13.0 --index-url https://download.pytorch.org/whl/cu129
+  - uv pip install --system conch-triton-kernels
   - VLLM_TEST_FORCE_LOAD_FORMAT=auto pytest -v -s quantization/ --ignore quantization/test_blackwell_moe.py
 
 - label: LM Eval Small Models # 53min
@@ -972,7 +973,6 @@ steps:
   - vllm/model_executor/layers/layernorm.py
   - vllm/model_executor/layers/activation.py
   - vllm/model_executor/layers/quantization/input_quant_fp8.py
-  - vllm/model_executor/layers/fused_moe/layer.py
   - tests/compile/test_fusion_attn.py
   - tests/compile/test_silu_mul_quant_fusion.py
   - tests/compile/distributed/test_fusion_all_reduce.py

.github/workflows/issue_autolabel.yml

@@ -105,6 +105,31 @@ jobs:
                   }
                 ],
               },
+              cpu: {
+                // Keyword search - matches whole words only (with word boundaries)
+                keywords: [
+                  {
+                    term: "CPU Backend",
+                    searchIn: "title"
+                  },
+                  {
+                    term: "x86",
+                    searchIn: "title"
+                  },
+                  {
+                    term: "ARM",
+                    searchIn: "title"
+                  },
+                  {
+                    term: "Apple Silicon",
+                    searchIn: "title"
+                  },
+                  {
+                    term: "IBM Z",
+                    searchIn: "title"
+                  },
+                ],
+              },
               // Add more label configurations here as needed
               // example: {
               //   keywords: [...],

README.md

@@ -21,6 +21,7 @@ Join us at the [PyTorch Conference, October 22-23](https://events.linuxfoundatio
 
 *Latest News* 🔥
 
+- [2025/11] We hosted [vLLM Bangkok Meetup](https://luma.com/v0f647nv). We explored vLLM and LMCache inference and low-resource language adaptation with speakers from Embedded LLM, AMD, and Red Hat. Please find the meetup slides [here](https://drive.google.com/drive/folders/1H0DS57F8HQ5q3kSOSoRmucPJWL3E0A_X?usp=sharing).
 - [2025/11] We hosted [the first vLLM Europe Meetup in Zurich](https://luma.com/0gls27kb) focused on quantization, distributed inference, and reinforcement learning at scale with speakers from Mistral, IBM, and Red Hat. Please find the meetup slides [here](https://docs.google.com/presentation/d/1UC9PTLCHYXQpOmJDSFg6Sljra3iVXzc09DeEI7dnxMc/edit?usp=sharing) and recording [here](https://www.youtube.com/watch?v=6m6ZE6yVEDI)
 - [2025/11] We hosted [vLLM Beijing Meetup](https://mp.weixin.qq.com/s/xSrYXjNgr1HbCP4ExYNG1w) focusing on distributed inference and diverse accelerator support with vLLM! Please find the meetup slides [here](https://drive.google.com/drive/folders/1nQJ8ZkLSjKxvu36sSHaceVXtttbLvvu-?usp=drive_link).
 - [2025/10] We hosted [vLLM Shanghai Meetup](https://mp.weixin.qq.com/s/__xb4OyOsImz-9eAVrdlcg) focused on hands-on vLLM inference optimization! Please find the meetup slides [here](https://drive.google.com/drive/folders/1KqwjsFJLfEsC8wlDugnrR61zsWHt94Q6).

csrc/moe/marlin_moe_wna16/marlin_template.h

@@ -489,14 +489,16 @@ __global__ void Marlin(
   #pragma unroll
         for (int i = 0; i < 4; i++) {
           int idx = tid4 * 4 + i;
-          idx = idx < block_num_valid_tokens ? idx : 0;
-          if constexpr (w_type == vllm::kFE2M1f && s_type == vllm::kFE4M3fn) {
-            sh_block_topk_weights[idx] = __hmul2(
-                global_scale, Dtype::num2num2(Dtype::float2num(
-                                  topk_weights_ptr[sh_block_sorted_ids[idx]])));
-          } else {
-            sh_block_topk_weights[idx] = Dtype::num2num2(
-                Dtype::float2num(topk_weights_ptr[sh_block_sorted_ids[idx]]));
+          if (idx < block_num_valid_tokens) {
+            if constexpr (w_type == vllm::kFE2M1f && s_type == vllm::kFE4M3fn) {
+              sh_block_topk_weights[idx] =
+                  __hmul2(global_scale,
+                          Dtype::num2num2(Dtype::float2num(
+                              topk_weights_ptr[sh_block_sorted_ids[idx]])));
+            } else {
+              sh_block_topk_weights[idx] = Dtype::num2num2(
+                  Dtype::float2num(topk_weights_ptr[sh_block_sorted_ids[idx]]));
+            }
           }
         }
       }

docker/Dockerfile

@@ -244,9 +244,15 @@ RUN mkdir -p /tmp/deepgemm/dist && touch /tmp/deepgemm/dist/.deepgemm_skipped
 
 COPY tools/ep_kernels/install_python_libraries.sh /tmp/install_python_libraries.sh
 # Install EP kernels(pplx-kernels and DeepEP)
+ARG PPLX_COMMIT_HASH
+ARG DEEPEP_COMMIT_HASH
 RUN --mount=type=cache,target=/root/.cache/uv \
     export TORCH_CUDA_ARCH_LIST='9.0a 10.0a' && \
-    /tmp/install_python_libraries.sh /tmp/ep_kernels_workspace wheel && \
+    /tmp/install_python_libraries.sh \
+        --workspace /tmp/ep_kernels_workspace \
+        --mode wheel \
+        ${PPLX_COMMIT_HASH:+--pplx-ref "$PPLX_COMMIT_HASH"} \
+        ${DEEPEP_COMMIT_HASH:+--deepep-ref "$DEEPEP_COMMIT_HASH"} && \
     find /tmp/ep_kernels_workspace/nvshmem -name '*.a' -delete
 
 # Check the size of the wheel if RUN_WHEEL_CHECK is true
@@ -392,8 +398,8 @@ RUN --mount=type=bind,from=build,src=/workspace/dist,target=/vllm-workspace/dist
 # Install FlashInfer pre-compiled kernel cache and binaries
 # https://docs.flashinfer.ai/installation.html
 RUN --mount=type=cache,target=/root/.cache/uv \
-    uv pip install --system flashinfer-cubin==0.5.2 \
-    && uv pip install --system flashinfer-jit-cache==0.5.2 \
+    uv pip install --system flashinfer-cubin==0.5.3 \
+    && uv pip install --system flashinfer-jit-cache==0.5.3 \
         --extra-index-url https://flashinfer.ai/whl/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.') \
     && flashinfer show-config
 

docker/Dockerfile.cpu

@@ -119,7 +119,6 @@ FROM base AS vllm-test-deps
 
 WORKDIR /workspace/vllm
 
-# TODO: Update to 2.9.0 when there is a new build for intel_extension_for_pytorch for that version
 RUN --mount=type=bind,src=requirements/test.in,target=requirements/test.in \
     cp requirements/test.in requirements/cpu-test.in && \
     sed -i '/mamba_ssm/d' requirements/cpu-test.in && \
@@ -132,9 +131,6 @@ RUN --mount=type=bind,src=requirements/test.in,target=requirements/test.in \
       esac; \
     }; \
     remove_packages_not_supported_on_aarch64 && \
-    sed -i 's/^torch==.*/torch==2.8.0/g' requirements/cpu-test.in && \
-    sed -i 's/torchaudio.*/torchaudio/g' requirements/cpu-test.in && \
-    sed -i 's/torchvision.*/torchvision/g' requirements/cpu-test.in && \
     uv pip compile requirements/cpu-test.in -o requirements/cpu-test.txt --index-strategy unsafe-best-match --torch-backend cpu
 
 RUN --mount=type=cache,target=/root/.cache/uv \

docs/community/meetups.md

@@ -10,6 +10,7 @@ Stay tuned for upcoming meetups! Follow us on [Twitter/X](https://x.com/vllm_pro
 
 Below you'll find slides and recordings from our previous meetups:
 
+- [vLLM Bangkok Meetup](https://luma.com/v0f647nv), November 21st 2025. [[Slides]](https://drive.google.com/drive/folders/1H0DS57F8HQ5q3kSOSoRmucPJWL3E0A_X?usp=sharing)
 - [vLLM Zurich Meetup](https://luma.com/0gls27kb), November 6th 2025. [[Slides]](https://docs.google.com/presentation/d/1UC9PTLCHYXQpOmJDSFg6Sljra3iVXzc09DeEI7dnxMc/edit?usp=sharing) [[Recording]](https://www.youtube.com/watch?v=6m6ZE6yVEDI)
 - [vLLM Beijing Meetup](https://mp.weixin.qq.com/s/xSrYXjNgr1HbCP4ExYNG1w), November 1st 2025. [[Slides]](https://drive.google.com/drive/folders/1nQJ8ZkLSjKxvu36sSHaceVXtttbLvvu-?usp=drive_link)
 - [vLLM Shanghai Meetup](https://mp.weixin.qq.com/s/__xb4OyOsImz-9eAVrdlcg), October 25th 2025. [[Slides]](https://drive.google.com/drive/folders/1KqwjsFJLfEsC8wlDugnrR61zsWHt94Q6)

docs/contributing/model/basic.md

@@ -29,7 +29,7 @@ The initialization code should look like this:
     ```python
     from torch import nn
     from vllm.config import VllmConfig
-    from vllm.attention import Attention
+    from vllm.attention.layer import Attention
 
     class MyAttention(nn.Module):
         def __init__(self, vllm_config: VllmConfig, prefix: str):

docs/design/cuda_graphs.md

@@ -84,12 +84,14 @@ See the following figures for a quick comparison between the previous and curren
 ```python
 class BatchDescriptor(NamedTuple):
     num_tokens: int
-    uniform_decode: bool = False
+    num_reqs: int
+    uniform: bool = False
+    has_lora: bool = False
 ```
 
-where `num_tokens` can be the padded token length, and `uniform_decode` is determined by if `max_query_len` of a batch is equal to the desired `max_query_len` of a uniform_decode, and the num_scheduled_tokens is divisible by that desired `max_query_len`.
+where `num_tokens` can be the padded token length, and `uniform` indicates if all the requests have the same query lengths. Many attention backends only support full cudagraphs when the batches are uniform; pure decode batches are uniform but may not be query length 1 (i.e. `num_tokens == num_reqs`), this occurs in the validation pass of spec-decode where "decode" batches will have a query length of  `1+num_spec_tokens`.
 
-The goal of this structure is to uniquely identify a (padded) batch with minimal possible items corresponding to a CUDA Graphs item. We are safe to exclude items like `uniform_query_len` because it is a constant at runtime for a certain setup currently. For example, it should be either `1` for a commonly pure decode or `1+num_spec_tokens` for a validation phase of speculative decode.
+The goal of this structure is to uniquely identify a (padded) batch with minimal possible items corresponding to a CUDA Graphs item.
 
 !!! note
     The prototype of `BatchDescriptor` may be extended for more general situations in the future, e.g., include more items, like `uniform_query_len` to support multiple different uniform decode lengths settings (<https://github.com/vllm-project/vllm/pull/23679>), or other modifications needed to support CUDA Graphs for models whose inputs are not necessarily token length aware (for example, some multi-modal inputs).