Merge pull request #478 from ROCm/unified_fp8

upstream_merge_25_03_12

Author: gshtras

.buildkite/release-pipeline.yaml

@@ -57,6 +57,8 @@ steps:
     agents:
       queue: tpu_queue_postmerge
     commands:
+      - "rm /var/log/syslog"
+      - "rm /var/log/kern.log"
       - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --tag vllm/vllm-tpu:nightly --tag vllm/vllm-tpu:$BUILDKITE_COMMIT --progress plain -f Dockerfile.tpu ."
       - "docker push vllm/vllm-tpu:nightly"
       - "docker push vllm/vllm-tpu:$BUILDKITE_COMMIT"

.buildkite/run-cpu-test.sh

@@ -19,13 +19,14 @@ remove_docker_container
 
 # Run the image, setting --shm-size=4g for tensor parallel.
 docker run -itd --entrypoint /bin/bash -v ~/.cache/huggingface:/root/.cache/huggingface --cpuset-cpus="$CORE_RANGE"  \
- --cpuset-mems="$NUMA_NODE" --privileged=true --network host -e HF_TOKEN --env VLLM_CPU_KVCACHE_SPACE=4 --shm-size=4g --name cpu-test-"$BUILDKITE_BUILD_NUMBER"-"$NUMA_NODE" cpu-test-"$BUILDKITE_BUILD_NUMBER"
+ --cpuset-mems="$NUMA_NODE" --privileged=true -e HF_TOKEN --env VLLM_CPU_KVCACHE_SPACE=4 --shm-size=4g --name cpu-test-"$BUILDKITE_BUILD_NUMBER"-"$NUMA_NODE" cpu-test-"$BUILDKITE_BUILD_NUMBER"
 docker run -itd --entrypoint /bin/bash -v ~/.cache/huggingface:/root/.cache/huggingface --cpuset-cpus="$CORE_RANGE" \
- --cpuset-mems="$NUMA_NODE" --privileged=true --network host -e HF_TOKEN --env VLLM_CPU_KVCACHE_SPACE=4 --shm-size=4g --name cpu-test-"$BUILDKITE_BUILD_NUMBER"-avx2-"$NUMA_NODE" cpu-test-"$BUILDKITE_BUILD_NUMBER"-avx2
+ --cpuset-mems="$NUMA_NODE" --privileged=true -e HF_TOKEN --env VLLM_CPU_KVCACHE_SPACE=4 --shm-size=4g --name cpu-test-"$BUILDKITE_BUILD_NUMBER"-avx2-"$NUMA_NODE" cpu-test-"$BUILDKITE_BUILD_NUMBER"-avx2
 
 function cpu_tests() {
   set -e
   export NUMA_NODE=$2
+  export BUILDKITE_BUILD_NUMBER=$3
 
   # offline inference
   docker exec cpu-test-"$BUILDKITE_BUILD_NUMBER"-avx2-"$NUMA_NODE" bash -c "
@@ -36,6 +37,7 @@ function cpu_tests() {
   docker exec cpu-test-"$BUILDKITE_BUILD_NUMBER"-"$NUMA_NODE" bash -c "
     set -e
     pip install -r vllm/requirements/test.txt
+    pip install -r vllm/requirements/cpu.txt
     pytest -v -s tests/models/decoder_only/language -m cpu_model
     pytest -v -s tests/models/embedding/language -m cpu_model
     pytest -v -s tests/models/encoder_decoder/language -m cpu_model
@@ -85,4 +87,4 @@ function cpu_tests() {
 
 # All of CPU tests are expected to be finished less than 40 mins.
 export -f cpu_tests
-timeout 40m bash -c "cpu_tests $CORE_RANGE $NUMA_NODE"
+timeout 40m bash -c "cpu_tests $CORE_RANGE $NUMA_NODE $BUILDKITE_BUILD_NUMBER"

.buildkite/run-tpu-test.sh

@@ -19,7 +19,6 @@ docker run --privileged --net host --shm-size=16G -it \
     vllm-tpu /bin/bash -c "python3 -m pip install git+https://github.com/thuml/depyf.git \
     && python3 -m pip install pytest \
     && python3 -m pip install lm_eval[api]==0.4.4 \
-    && pytest -v -s /workspace/vllm/tests/entrypoints/openai/test_accuracy.py \
     && pytest -v -s /workspace/vllm/tests/tpu/test_custom_dispatcher.py \
     && python3 /workspace/vllm/tests/tpu/test_compilation.py \
     && python3 /workspace/vllm/tests/tpu/test_quantization_accuracy.py \

.github/workflows/scripts/create_release.js

@@ -1,4 +1,4 @@
-// Uses Github's API to create the release and wait for result.
+// Uses GitHub's API to create the release and wait for result.
 // We use a JS script since github CLI doesn't provide a way to wait for the release's creation and returns immediately.
 
 module.exports = async (github, context, core) => {

CMakeLists.txt

@@ -156,20 +156,6 @@ endif()
 #
 get_torch_gpu_compiler_flags(VLLM_GPU_FLAGS ${VLLM_GPU_LANG})
 
-#
-# Get supported FP8 format based on GPU arches
-#
-get_supported_fp8_format(FP8_FORMAT ${VLLM_GPU_LANG} "${VLLM_GPU_ARCHES}")
-if(${FP8_FORMAT} STREQUAL "E4M3FN")
-  message(STATUS "FP8 format: E4M3FN")
-  list(APPEND VLLM_GPU_FLAGS "-DUSE_CUDA_FP8_FORMAT")
-elseif(${FP8_FORMAT} STREQUAL "E4M3FNUZ")
-  message(STATUS "FP8 format: E4M3FNUZ")
-  list(APPEND VLLM_GPU_FLAGS "-DUSE_HIP_FP8_FORMAT")
-elseif(${FP8_FORMAT} STREQUAL "CONFLICT")
-  message(FATAL_ERROR "Target architectures support different types of FP8 formats!")
-endif()
-
 #
 # Set nvcc parallelism.
 #
@@ -437,7 +423,8 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
   # 2:4 Sparse Kernels
 
   # The 2:4 sparse kernels cutlass_scaled_sparse_mm and cutlass_compressor
-  # require CUDA 12.2 or later (and only work on Hopper and Blackwell).
+  # require CUDA 12.2 or later (and only work on Hopper).
+  cuda_archs_loose_intersection(SCALED_MM_ARCHS "9.0a;" "${CUDA_ARCHS}")
   if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.2 AND SCALED_MM_ARCHS)
     set(SRCS "csrc/sparse/cutlass/sparse_scaled_mm_c3x.cu")
     set_gencode_flags_for_srcs(
@@ -585,6 +572,14 @@ set_gencode_flags_for_srcs(
   CUDA_ARCHS "${CUDA_ARCHS}")
 
 if(VLLM_GPU_LANG STREQUAL "CUDA")
+  set(VLLM_MOE_WNA16_SRC
+    "csrc/moe/moe_wna16.cu")
+
+  set_gencode_flags_for_srcs(
+    SRCS "${VLLM_MOE_WNA16_SRC}"
+    CUDA_ARCHS "${CUDA_ARCHS}")
+
+  list(APPEND VLLM_MOE_EXT_SRC "${VLLM_MOE_WNA16_SRC}")
   cuda_archs_loose_intersection(MARLIN_MOE_ARCHS "8.0;8.6;8.7;8.9;9.0;10.0;10.1;12.0" "${CUDA_ARCHS}")
   if (MARLIN_MOE_ARCHS)
     set(MARLIN_MOE_SRC

Dockerfile.cpu

@@ -22,7 +22,7 @@ ENV LD_PRELOAD="/usr/lib/x86_64-linux-gnu/libtcmalloc_minimal.so.4:/usr/local/li
 
 RUN echo 'ulimit -c 0' >> ~/.bashrc
 
-RUN pip install intel_extension_for_pytorch==2.5.0
+RUN pip install intel_extension_for_pytorch==2.6.0
 
 WORKDIR /workspace
 

Dockerfile.tpu

@@ -15,6 +15,9 @@ ARG GIT_REPO_CHECK=0
 RUN --mount=type=bind,source=.git,target=.git \
     if [ "$GIT_REPO_CHECK" != 0 ]; then bash tools/check_repo.sh; fi
 
+# Remove existing versions of dependencies
+RUN pip uninstall -y torch torch_xla torchvision
+
 ENV VLLM_TARGET_DEVICE="tpu"
 RUN --mount=type=cache,target=/root/.cache/pip \
     --mount=type=bind,source=.git,target=.git \

Dockerfile.xpu

@@ -1,4 +1,4 @@
-FROM intel/oneapi-basekit:2024.2.1-0-devel-ubuntu22.04 AS vllm-base
+FROM intel/deep-learning-essentials:2025.0.1-0-devel-ubuntu22.04 AS vllm-base
 
 RUN wget -O- https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB | gpg --dearmor | tee /usr/share/keyrings/intel-oneapi-archive-keyring.gpg > /dev/null && \
     echo "deb [signed-by=/usr/share/keyrings/intel-oneapi-archive-keyring.gpg] https://apt.repos.intel.com/oneapi all main " | tee /etc/apt/sources.list.d/oneAPI.list && \
@@ -21,7 +21,8 @@ RUN apt-get update -y && \
     python3 \
     python3-dev \
     python3-pip \
-    # vim \
+    libze-intel-gpu-dev \
+    libze-intel-gpu1 \
     wget
 
 WORKDIR /workspace/vllm
@@ -32,19 +33,10 @@ RUN --mount=type=cache,target=/root/.cache/pip \
     pip install --no-cache-dir \
     -r requirements/xpu.txt
 
-RUN git clone https://github.com/intel/pti-gpu && \
-    cd pti-gpu/sdk && \
-    git checkout 6c491f07a777ed872c2654ca9942f1d0dde0a082 && \
-    mkdir build && \
-    cd build && \
-    cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_TOOLCHAIN_FILE=../cmake/toolchains/icpx_toolchain.cmake -DBUILD_TESTING=OFF .. && \
-    make -j && \
-    cmake --install . --config Release --prefix "/usr/local"
-
 ENV LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/usr/local/lib/"
 
 COPY . .
-ARG GIT_REPO_CHECK
+ARG GIT_REPO_CHECK=0
 RUN --mount=type=bind,source=.git,target=.git \
     if [ "$GIT_REPO_CHECK" != 0 ]; then bash tools/check_repo.sh; fi
 
@@ -54,6 +46,12 @@ RUN --mount=type=cache,target=/root/.cache/pip \
     --mount=type=bind,source=.git,target=.git \
     python3 setup.py install
 
+# Please refer xpu doc, we need manually install intel-extension-for-pytorch 2.6.10+xpu due to there are some conflict dependencies with torch 2.6.0+xpu
+# FIXME: This will be fix in ipex 2.7. just leave this here for awareness.
+RUN --mount=type=cache,target=/root/.cache/pip \
+    pip install intel-extension-for-pytorch==2.6.10+xpu \
+    --extra-index-url=https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
+
 CMD ["/bin/bash"]
 
 FROM vllm-base AS vllm-openai

README.md

@@ -91,7 +91,7 @@ pip install vllm
 ```
 
 Visit our [documentation](https://docs.vllm.ai/en/latest/) to learn more.
-- [Installation](https://docs.vllm.ai/en/latest/getting_started/installation/index.html)
+- [Installation](https://docs.vllm.ai/en/latest/getting_started/installation.html)
 - [Quickstart](https://docs.vllm.ai/en/latest/getting_started/quickstart.html)
 - [List of Supported Models](https://docs.vllm.ai/en/latest/models/supported_models.html)
 
@@ -151,9 +151,9 @@ If you use vLLM for your research, please cite our [paper](https://arxiv.org/abs
 
 ## Contact Us
 
-- For technical questions and feature requests, please use Github issues or discussions.
+- For technical questions and feature requests, please use GitHub issues or discussions.
 - For discussing with fellow users and coordinating contributions and development, please use Slack.
-- For security disclosures, please use Github's security advisory feature.
+- For security disclosures, please use GitHub's security advisory feature.
 - For collaborations and partnerships, please contact us at vllm-questions AT lists.berkeley.edu.
 
 ## Media Kit

benchmarks/README.md

@@ -1,29 +1,181 @@
 # Benchmarking vLLM
 
-## Downloading the ShareGPT dataset
+This README guides you through running benchmark tests with the extensive
+datasets supported on vLLM. It’s a living document, updated as new features and datasets
+become available.
 
-You can download the dataset by running:
+## Dataset Overview
+
+<table style="width:100%; border-collapse: collapse;">
+  <thead>
+    <tr>
+      <th style="width:15%; text-align: left;">Dataset</th>
+      <th style="width:10%; text-align: center;">Online</th>
+      <th style="width:10%; text-align: center;">Offline</th>
+      <th style="width:65%; text-align: left;">Data Path</th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr>
+      <td><strong>ShareGPT</strong></td>
+      <td style="text-align: center;">✅</td>
+      <td style="text-align: center;">✅</td>
+      <td><code>wget https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered/resolve/main/ShareGPT_V3_unfiltered_cleaned_split.json</code></td>
+    </tr>
+    <tr>
+      <td><strong>BurstGPT</strong></td>
+      <td style="text-align: center;">✅</td>
+      <td style="text-align: center;">✅</td>
+      <td><code>wget https://github.com/HPMLL/BurstGPT/releases/download/v1.1/BurstGPT_without_fails_2.csv</code></td>
+    </tr>
+    <tr>
+      <td><strong>Sonnet</strong></td>
+      <td style="text-align: center;">✅</td>
+      <td style="text-align: center;">✅</td>
+      <td>Local file: <code>benchmarks/sonnet.txt</code></td>
+    </tr>
+    <tr>
+      <td><strong>Random</strong></td>
+      <td style="text-align: center;">✅</td>
+      <td style="text-align: center;">✅</td>
+      <td><code>synthetic</code></td>
+    </tr>
+    <tr>
+      <td><strong>HuggingFace</strong></td>
+      <td style="text-align: center;">✅</td>
+      <td style="text-align: center;">🚧</td>
+      <td>Specify your dataset path on HuggingFace</td>
+    </tr>
+    <tr>
+      <td><strong>VisionArena</strong></td>
+      <td style="text-align: center;">✅</td>
+      <td style="text-align: center;">🚧</td>
+      <td><code>lmarena-ai/vision-arena-bench-v0.1</code> (a HuggingFace dataset)</td>
+    </tr>
+  </tbody>
+</table>
+✅: supported  
+🚧: to be supported
+
+**Note**: VisionArena’s `dataset-name` should be set to `hf`
+
+---
+## Example - Online Benchmark
+
+First start serving your model
 
 ```bash
-wget https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered/resolve/main/ShareGPT_V3_unfiltered_cleaned_split.json
+MODEL_NAME="NousResearch/Hermes-3-Llama-3.1-8B"
+vllm serve ${MODEL_NAME} --disable-log-requests
 ```
 
-## Downloading the ShareGPT4V dataset
+Then run the benchmarking script
+
+```bash
+# download dataset
+# wget https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered/resolve/main/ShareGPT_V3_unfiltered_cleaned_split.json
+MODEL_NAME="NousResearch/Hermes-3-Llama-3.1-8B"
+NUM_PROMPTS=10
+BACKEND="openai-chat"
+DATASET_NAME="sharegpt"
+DATASET_PATH="<your data path>/ShareGPT_V3_unfiltered_cleaned_split.json"
+python3 benchmarks/benchmark_serving.py --backend ${BACKEND} --model ${MODEL_NAME} --endpoint /v1/chat/completions --dataset-name ${DATASET_NAME} --dataset-path ${DATASET_PATH} --num-prompts ${NUM_PROMPTS}
+```
+
+If successful, you will see the following output
+
+```
+============ Serving Benchmark Result ============
+Successful requests:                     10        
+Benchmark duration (s):                  5.78      
+Total input tokens:                      1369      
+Total generated tokens:                  2212      
+Request throughput (req/s):              1.73      
+Output token throughput (tok/s):         382.89    
+Total Token throughput (tok/s):          619.85    
+---------------Time to First Token----------------
+Mean TTFT (ms):                          71.54     
+Median TTFT (ms):                        73.88     
+P99 TTFT (ms):                           79.49     
+-----Time per Output Token (excl. 1st token)------
+Mean TPOT (ms):                          7.91      
+Median TPOT (ms):                        7.96      
+P99 TPOT (ms):                           8.03      
+---------------Inter-token Latency----------------
+Mean ITL (ms):                           7.74      
+Median ITL (ms):                         7.70      
+P99 ITL (ms):                            8.39      
+==================================================
+```
 
-The json file refers to several image datasets (coco, llava, etc.). The benchmark scripts
-will ignore a datapoint if the referred image is missing.
+### VisionArena Benchmark for Vision Language Models
 
 ```bash
-wget https://huggingface.co/datasets/Lin-Chen/ShareGPT4V/resolve/main/sharegpt4v_instruct_gpt4-vision_cap100k.json
-mkdir coco -p
-wget http://images.cocodataset.org/zips/train2017.zip -O coco/train2017.zip
-unzip coco/train2017.zip -d coco/
+# need a model with vision capability here
+vllm serve Qwen/Qwen2-VL-7B-Instruct --disable-log-requests
 ```
 
-# Downloading the BurstGPT dataset
+```bash
+MODEL_NAME="Qwen/Qwen2-VL-7B-Instruct"
+NUM_PROMPTS=10
+BACKEND="openai-chat"
+DATASET_NAME="hf"
+DATASET_PATH="lmarena-ai/vision-arena-bench-v0.1"
+DATASET_SPLIT='train'
+
+python3 benchmarks/benchmark_serving.py \
+  --backend "${BACKEND}" \
+  --model "${MODEL_NAME}" \
+  --endpoint "/v1/chat/completions" \
+  --dataset-name "${DATASET_NAME}" \
+  --dataset-path "${DATASET_PATH}" \
+  --hf-split "${DATASET_SPLIT}" \
+  --num-prompts "${NUM_PROMPTS}"
+```
 
-You can download the BurstGPT v1.1 dataset by running:
+---
+## Example - Offline Throughput Benchmark
 
 ```bash
-wget https://github.com/HPMLL/BurstGPT/releases/download/v1.1/BurstGPT_without_fails_2.csv
+MODEL_NAME="NousResearch/Hermes-3-Llama-3.1-8B"
+NUM_PROMPTS=10
+DATASET_NAME="sonnet"
+DATASET_PATH="benchmarks/sonnet.txt"
+
+python3 benchmarks/benchmark_throughput.py \
+  --model "${MODEL_NAME}" \
+  --dataset-name "${DATASET_NAME}" \
+  --dataset-path "${DATASET_PATH}" \
+  --num-prompts "${NUM_PROMPTS}"
+  ```
+
+If successful, you will see the following output
+
+```
+Throughput: 7.35 requests/s, 4789.20 total tokens/s, 1102.83 output tokens/s
 ```
+
+### Benchmark with LoRA Adapters
+
+``` bash
+MODEL_NAME="meta-llama/Llama-2-7b-hf"
+BACKEND="vllm"
+DATASET_NAME="sharegpt"
+DATASET_PATH="/home/jovyan/data/vllm_benchmark_datasets/ShareGPT_V3_unfiltered_cleaned_split.json"
+NUM_PROMPTS=10
+MAX_LORAS=2
+MAX_LORA_RANK=8
+ENABLE_LORA="--enable-lora"
+LORA_PATH="yard1/llama-2-7b-sql-lora-test"
+
+python3 benchmarks/benchmark_throughput.py \
+  --model "${MODEL_NAME}" \
+  --backend "${BACKEND}" \
+  --dataset_path "${DATASET_PATH}" \
+  --dataset_name "${DATASET_NAME}" \
+  --num-prompts "${NUM_PROMPTS}" \
+  --max-loras "${MAX_LORAS}" \
+  --max-lora-rank "${MAX_LORA_RANK}" \
+  ${ENABLE_LORA} \
+  --lora-path "${LORA_PATH}"
+  ```