Merge remote-tracking branch 'mirror/main'

.buildkite/lm-eval-harness/configs/Meta-Llama-3.2-1B-Instruct-INT8-compressed-tensors.yaml

+# bash .buildkite/lm-eval-harness/run-lm-eval-gsm-vllm-baseline.sh -m neuralmagic/Llama-3.2-1B-Instruct-quantized.w8a8 -b "auto" -l 1000 -f 5 -t 1
+model_name: "neuralmagic/Llama-3.2-1B-Instruct-quantized.w8a8"
+tasks:
+- name: "gsm8k"
+  metrics:
+  - name: "exact_match,strict-match"
+    value: 0.356
+  - name: "exact_match,flexible-extract"
+    value: 0.358
+limit: 1000
+num_fewshot: 5

.buildkite/lm-eval-harness/configs/models-small.txt

 Meta-Llama-3-8B-Instruct.yaml
 Meta-Llama-3-8B-Instruct-FP8-compressed-tensors.yaml
-Meta-Llama-3-8B-Instruct-INT8-compressed-tensors.yaml
+Meta-Llama-3.2-1B-Instruct-INT8-compressed-tensors.yaml
 Meta-Llama-3-8B-Instruct-INT8-compressed-tensors-asym.yaml
 Meta-Llama-3-8B-Instruct-nonuniform-compressed-tensors.yaml
 Meta-Llama-3-8B-Instruct-Channelwise-compressed-tensors.yaml

.buildkite/test-pipeline.yaml

@@ -230,14 +230,12 @@ steps:
   commands:
   - pytest -v -s compile/test_basic_correctness.py
 
-# TODO: re-write in comparison tests, and fix symbolic shape
-# for quantization ops.
-# - label: "PyTorch Fullgraph Test" # 18min
-#   source_file_dependencies:
-#   - vllm/
-#   - tests/compile
-#   commands:
-#   - pytest -v -s compile/test_full_graph.py
+- label: "PyTorch Fullgraph Test" # 18min
+  source_file_dependencies:
+  - vllm/
+  - tests/compile
+  commands:
+  - pytest -v -s compile/test_full_graph.py
 
 - label: Kernels Test %N # 1h each
   mirror_hardwares: [amd]
@@ -312,13 +310,22 @@ steps:
     - pytest -v -s models/test_oot_registration.py # it needs a clean process
     - pytest -v -s models/*.py --ignore=models/test_oot_registration.py
 
-- label: Decoder-only Language Models Test # 1h36min
+- label: Decoder-only Language Models Test (Standard) # 35min
   #mirror_hardwares: [amd]
   source_file_dependencies:
   - vllm/
   - tests/models/decoder_only/language
   commands:
-    - pytest -v -s models/decoder_only/language
+    - pytest -v -s models/decoder_only/language/test_models.py
+    - pytest -v -s models/decoder_only/language/test_big_models.py
+
+- label: Decoder-only Language Models Test (Extended) # 1h20min
+  nightly: true
+  source_file_dependencies:
+  - vllm/
+  - tests/models/decoder_only/language
+  commands:
+    - pytest -v -s models/decoder_only/language --ignore=models/decoder_only/language/test_models.py --ignore=models/decoder_only/language/test_big_models.py
 
 - label: Decoder-only Multi-Modal Models Test # 1h31min
   #mirror_hardwares: [amd]

.github/workflows/actionlint.yml

@@ -34,4 +34,5 @@ jobs:
 
       - name: "Run actionlint"
         run: |
+          echo "::add-matcher::.github/workflows/matchers/actionlint.json"
           tools/actionlint.sh -color

.github/workflows/add_label_automerge.yml

@@ -8,7 +8,7 @@ jobs:
         runs-on: ubuntu-latest
         steps:
             -   name: Add label
-                uses: actions/github-script@v7
+                uses: actions/github-script@60a0d83039c74a4aee543508d2ffcb1c3799cdea # v7.0.1
                 with:
                     script: |
                         github.rest.issues.addLabels({

.github/workflows/clang-format.yml

@@ -17,9 +17,9 @@ jobs:
       matrix:
         python-version: ["3.11"]
     steps:
-    - uses: actions/checkout@v4
+    - uses: actions/checkout@eef61447b9ff4aafe5dcd4e0bbf5d482be7e7871 # v4.2.1
     - name: Set up Python ${{ matrix.python-version }}
-      uses: actions/setup-python@v5
+      uses: actions/setup-python@f677139bbe7f9c59b41e40162b753c062f5d49a3 # v5.2.0
       with:
         python-version: ${{ matrix.python-version }}
     - name: Install dependencies
@@ -38,4 +38,4 @@ jobs:
         )
         find csrc/ \( -name '*.h' -o -name '*.cpp' -o -name '*.cu' -o -name '*.cuh' \) -print \
             | grep -vFf <(printf "%s\n" "${EXCLUDES[@]}") \
-            | xargs clang-format --dry-run --Werror
\ No newline at end of file
+            | xargs clang-format --dry-run --Werror

.github/workflows/matchers/mypy.json

+{
+  "problemMatcher": [
+    {
+      "owner": "mypy",
+      "pattern": [
+        {
+          "regexp": "^(.+):(\\d+):\\s(error|warning):\\s(.+)$",
+          "file": 1,
+          "line": 2,
+          "severity": 3,
+          "message": 4
+        }
+      ]
+    }
+  ]
+}

.github/workflows/matchers/ruff.json

+{
+    "problemMatcher": [
+      {
+        "owner": "ruff",
+        "pattern": [
+          {
+            "regexp": "^(.+?):(\\d+):(\\d+): (\\w+): (.+)$",
+            "file": 1,
+            "line": 2,
+            "column": 3,
+            "code": 4,
+            "message": 5
+          }
+        ]
+      }
+    ]
+  }

.github/workflows/mypy.yaml

@@ -17,9 +17,9 @@ jobs:
       matrix:
         python-version: ["3.8", "3.9", "3.10", "3.11", "3.12"]
     steps:
-    - uses: actions/checkout@v4
+    - uses: actions/checkout@eef61447b9ff4aafe5dcd4e0bbf5d482be7e7871 # v4.2.1
     - name: Set up Python ${{ matrix.python-version }}
-      uses: actions/setup-python@v5
+      uses: actions/setup-python@f677139bbe7f9c59b41e40162b753c062f5d49a3 # v5.2.0
       with:
         python-version: ${{ matrix.python-version }}
     - name: Install dependencies
@@ -32,4 +32,5 @@ jobs:
         pip install types-setuptools
     - name: Mypy
       run: |
-        tools/mypy.sh
+        echo "::add-matcher::.github/workflows/matchers/mypy.json"
+        tools/mypy.sh 1

.github/workflows/publish.yml

@@ -21,7 +21,7 @@ jobs:
       upload_url: ${{ steps.create_release.outputs.upload_url }}
     steps:
       - name: Checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@eef61447b9ff4aafe5dcd4e0bbf5d482be7e7871 # v4.2.1
 
       - name: Extract branch info
         shell: bash
@@ -30,7 +30,7 @@ jobs:
 
       - name: Create Release
         id: create_release
-        uses: "actions/github-script@v7"
+        uses: actions/github-script@60a0d83039c74a4aee543508d2ffcb1c3799cdea # v7.0.1
         env:
           RELEASE_TAG: ${{ env.release_tag }}
         with:
@@ -54,10 +54,10 @@ jobs:
 
     steps:
       - name: Checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@eef61447b9ff4aafe5dcd4e0bbf5d482be7e7871 # v4.2.1
 
       - name: Setup ccache
-        uses: hendrikmuhs/ccache-action@v1.2
+        uses: hendrikmuhs/ccache-action@ed74d11c0b343532753ecead8a951bb09bb34bc9 # v1.2.14
         with:
           create-symlink: true
           key: ${{ github.job }}-${{ matrix.python-version }}-${{ matrix.cuda-version }}
@@ -68,7 +68,7 @@ jobs:
           bash -x .github/workflows/scripts/env.sh
 
       - name: Set up Python
-        uses: actions/setup-python@v5
+        uses: actions/setup-python@f677139bbe7f9c59b41e40162b753c062f5d49a3 # v5.2.0
         with:
             python-version: ${{ matrix.python-version }}
 
@@ -92,7 +92,7 @@ jobs:
           echo "asset_name=${asset_name}" >> "$GITHUB_ENV"
 
       - name: Upload Release Asset
-        uses: actions/upload-release-asset@v1
+        uses: actions/upload-release-asset@e8f9f06c4b078e705bd2ea027f0926603fc9b4d5 # v1.0.2
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
         with:

.github/workflows/reminder_comment.yml

@@ -8,7 +8,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Remind to run full CI on PR
-        uses: actions/github-script@v7
+        uses: actions/github-script@60a0d83039c74a4aee543508d2ffcb1c3799cdea # v7.0.1
         with:
           script: |
             github.rest.issues.createComment({

.github/workflows/ruff.yml

@@ -17,9 +17,9 @@ jobs:
       matrix:
         python-version: ["3.8", "3.9", "3.10", "3.11", "3.12"]
     steps:
-    - uses: actions/checkout@v4
+    - uses: actions/checkout@eef61447b9ff4aafe5dcd4e0bbf5d482be7e7871 # v4.2.1
     - name: Set up Python ${{ matrix.python-version }}
-      uses: actions/setup-python@v5
+      uses: actions/setup-python@f677139bbe7f9c59b41e40162b753c062f5d49a3 # v5.2.0
       with:
         python-version: ${{ matrix.python-version }}
     - name: Install dependencies
@@ -28,7 +28,8 @@ jobs:
         pip install -r requirements-lint.txt
     - name: Analysing the code with ruff
       run: |
-        ruff check .
+        echo "::add-matcher::.github/workflows/matchers/ruff.json"
+        ruff check --output-format github .
     - name: Spelling check with codespell
       run: |
         codespell --toml pyproject.toml

.github/workflows/yapf.yml

@@ -16,9 +16,9 @@ jobs:
       matrix:
         python-version: ["3.8", "3.9", "3.10", "3.11", "3.12"]
     steps:
-    - uses: actions/checkout@v4
+    - uses: actions/checkout@eef61447b9ff4aafe5dcd4e0bbf5d482be7e7871 # v4.2.1
     - name: Set up Python ${{ matrix.python-version }}
-      uses: actions/setup-python@v5
+      uses: actions/setup-python@f677139bbe7f9c59b41e40162b753c062f5d49a3 # v5.2.0
       with:
         python-version: ${{ matrix.python-version }}
     - name: Install dependencies

CMakeLists.txt

@@ -87,24 +87,6 @@ endif()
 #
 find_package(Torch REQUIRED)
 
-#
-message(STATUS "Enabling core extension.")
-
-# Define _core_C extension
-#  built for (almost) every target platform, (excludes TPU and Neuron)
-
-set(VLLM_EXT_SRC
-  "csrc/core/torch_bindings.cpp")
-
-define_gpu_extension_target(
-  _core_C
-  DESTINATION vllm
-  LANGUAGE CXX
-  SOURCES ${VLLM_EXT_SRC}
-  COMPILE_FLAGS ${CXX_COMPILE_FLAGS}
-  USE_SABI 3
-  WITH_SOABI)
-
 #
 # Forward the non-CUDA device extensions to external CMake scripts.
 #
@@ -191,12 +173,12 @@ endif()
 
 #
 # Use FetchContent for C++ dependencies that are compiled as part of vLLM's build process.
-# Configure it to place files in vllm/.deps, in order to play nicely with sccache.
+# setup.py will override FETCHCONTENT_BASE_DIR to play nicely with sccache.
+# Each dependency that produces build artifacts should override its BINARY_DIR to avoid
+# conflicts between build types. It should instead be set to ${CMAKE_BINARY_DIR}/<dependency>.
 #
 include(FetchContent)
-get_filename_component(PROJECT_ROOT_DIR "${CMAKE_CURRENT_SOURCE_DIR}" ABSOLUTE)
-file(MAKE_DIRECTORY "${FETCHCONTENT_BASE_DIR}")
-set(FETCHCONTENT_BASE_DIR "${PROJECT_ROOT_DIR}/.deps")
+file(MAKE_DIRECTORY ${FETCHCONTENT_BASE_DIR}) # Ensure the directory exists
 message(STATUS "FetchContent base directory: ${FETCHCONTENT_BASE_DIR}")
 
 #
@@ -280,7 +262,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
     message(STATUS "Building Marlin kernels for archs: ${MARLIN_ARCHS}")
   else()
     message(STATUS "Not building Marlin kernels as no compatible archs found"
-                   "in CUDA target architectures")
+                   " in CUDA target architectures")
   endif()
 
   #
@@ -460,7 +442,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
     message(STATUS "Building Marlin MOE kernels for archs: ${MARLIN_MOE_ARCHS}")
   else()
     message(STATUS "Not building Marlin MOE kernels as no compatible archs found"
-                   "in CUDA target architectures")
+                   " in CUDA target architectures")
   endif()
 endif()
 
@@ -540,6 +522,8 @@ else()
           GIT_REPOSITORY https://github.com/vllm-project/flash-attention.git
           GIT_TAG 013f0c4fc47e6574060879d9734c1df8c5c273bd
           GIT_PROGRESS TRUE
+          # Don't share the vllm-flash-attn build between build types
+          BINARY_DIR ${CMAKE_BINARY_DIR}/vllm-flash-attn
   )
 ]]
 endif()

Dockerfile.openvino

@@ -15,11 +15,11 @@ RUN --mount=type=bind,source=.git,target=.git \
     if [ "$GIT_REPO_CHECK" != 0 ]; then bash tools/check_repo.sh ; fi
 
 # install build requirements
-RUN PIP_EXTRA_INDEX_URL="https://download.pytorch.org/whl/cpu" python3 -m pip install -r /workspace/vllm/requirements-build.txt
+RUN PIP_EXTRA_INDEX_URL="https://download.pytorch.org/whl/cpu" python3 -m pip install -r /workspace/requirements-build.txt
 # build vLLM with OpenVINO backend
-RUN PIP_EXTRA_INDEX_URL="https://download.pytorch.org/whl/cpu" VLLM_TARGET_DEVICE="openvino" python3 -m pip install /workspace/vllm/
+RUN PIP_EXTRA_INDEX_URL="https://download.pytorch.org/whl/cpu" VLLM_TARGET_DEVICE="openvino" python3 -m pip install /workspace
 
-COPY examples/ /workspace/vllm/examples
-COPY benchmarks/ /workspace/vllm/benchmarks
+COPY examples/ /workspace/examples
+COPY benchmarks/ /workspace/benchmarks
 
 CMD ["/bin/bash"]

Dockerfile.ppc64le

@@ -33,4 +33,4 @@ WORKDIR /workspace/
 
 RUN ln -s /workspace/vllm/tests && ln -s /workspace/vllm/examples && ln -s /workspace/vllm/benchmarks
 
-ENTRYPOINT ["python3", "-m", "vllm.entrypoints.openai.api_server"]
+ENTRYPOINT ["/opt/conda/bin/python3", "-m", "vllm.entrypoints.openai.api_server"]

benchmarks/benchmark_latency.py

 """Benchmark the latency of processing a single batch of requests."""
 import argparse
+import dataclasses
 import json
 import time
 from pathlib import Path
@@ -10,43 +11,19 @@ import torch
 from tqdm import tqdm
 
 from vllm import LLM, SamplingParams
-from vllm.engine.arg_utils import DEVICE_OPTIONS, EngineArgs
+from vllm.engine.arg_utils import EngineArgs
 from vllm.inputs import PromptType
-from vllm.model_executor.layers.quantization import QUANTIZATION_METHODS
 from vllm.utils import FlexibleArgumentParser
 
 
 def main(args: argparse.Namespace):
     print(args)
 
+    engine_args = EngineArgs.from_cli_args(args)
+
     # NOTE(woosuk): If the request cannot be processed in a single batch,
     # the engine will automatically process the request in multiple batches.
-    llm = LLM(
-        model=args.model,
-        speculative_model=args.speculative_model,
-        num_speculative_tokens=args.num_speculative_tokens,
-        speculative_draft_tensor_parallel_size=\
-            args.speculative_draft_tensor_parallel_size,
-        tokenizer=args.tokenizer,
-        quantization=args.quantization,
-        tensor_parallel_size=args.tensor_parallel_size,
-        trust_remote_code=args.trust_remote_code,
-        dtype=args.dtype,
-        max_model_len=args.max_model_len,
-        enforce_eager=args.enforce_eager,
-        kv_cache_dtype=args.kv_cache_dtype,
-        quantization_param_path=args.quantization_param_path,
-        device=args.device,
-        ray_workers_use_nsight=args.ray_workers_use_nsight,
-        enable_chunked_prefill=args.enable_chunked_prefill,
-        download_dir=args.download_dir,
-        block_size=args.block_size,
-        gpu_memory_utilization=args.gpu_memory_utilization,
-        load_format=args.load_format,
-        distributed_executor_backend=args.distributed_executor_backend,
-        otlp_traces_endpoint=args.otlp_traces_endpoint,
-        enable_prefix_caching=args.enable_prefix_caching,
-    )
+    llm = LLM(**dataclasses.asdict(engine_args))
 
     sampling_params = SamplingParams(
         n=args.n,
@@ -125,19 +102,6 @@ if __name__ == '__main__':
     parser = FlexibleArgumentParser(
         description='Benchmark the latency of processing a single batch of '
         'requests till completion.')
-    parser.add_argument('--model', type=str, default='facebook/opt-125m')
-    parser.add_argument('--speculative-model', type=str, default=None)
-    parser.add_argument('--num-speculative-tokens', type=int, default=None)
-    parser.add_argument('--speculative-draft-tensor-parallel-size',
-                        '-spec-draft-tp',
-                        type=int,
-                        default=None)
-    parser.add_argument('--tokenizer', type=str, default=None)
-    parser.add_argument('--quantization',
-                        '-q',
-                        choices=[*QUANTIZATION_METHODS, None],
-                        default=None)
-    parser.add_argument('--tensor-parallel-size', '-tp', type=int, default=1)
     parser.add_argument('--input-len', type=int, default=32)
     parser.add_argument('--output-len', type=int, default=128)
     parser.add_argument('--batch-size', type=int, default=8)
@@ -154,45 +118,6 @@ if __name__ == '__main__':
                         type=int,
                         default=30,
                         help='Number of iterations to run.')
-    parser.add_argument('--trust-remote-code',
-                        action='store_true',
-                        help='trust remote code from huggingface')
-    parser.add_argument(
-        '--max-model-len',
-        type=int,
-        default=None,
-        help='Maximum length of a sequence (including prompt and output). '
-        'If None, will be derived from the model.')
-    parser.add_argument(
-        '--dtype',
-        type=str,
-        default='auto',
-        choices=['auto', 'half', 'float16', 'bfloat16', 'float', 'float32'],
-        help='data type for model weights and activations. '
-        'The "auto" option will use FP16 precision '
-        'for FP32 and FP16 models, and BF16 precision '
-        'for BF16 models.')
-    parser.add_argument('--enforce-eager',
-                        action='store_true',
-                        help='enforce eager mode and disable CUDA graph')
-    parser.add_argument(
-        '--kv-cache-dtype',
-        type=str,
-        choices=['auto', 'fp8', 'fp8_e5m2', 'fp8_e4m3'],
-        default="auto",
-        help='Data type for kv cache storage. If "auto", will use model '
-        'data type. CUDA 11.8+ supports fp8 (=fp8_e4m3) and fp8_e5m2. '
-        'ROCm (AMD GPU) supports fp8 (=fp8_e4m3)')
-    parser.add_argument(
-        '--quantization-param-path',
-        type=str,
-        default=None,
-        help='Path to the JSON file containing the KV cache scaling factors. '
-        'This should generally be supplied, when KV cache dtype is FP8. '
-        'Otherwise, KV cache scaling factors default to 1.0, which may cause '
-        'accuracy issues. FP8_E5M2 (without scaling) is only supported on '
-        'cuda version greater than 11.8. On ROCm (AMD GPU), FP8_E4M3 is '
-        'instead supported for common inference criteria.')
     parser.add_argument(
         '--profile',
         action='store_true',
@@ -203,78 +128,12 @@ if __name__ == '__main__':
         default=None,
         help=('path to save the pytorch profiler output. Can be visualized '
               'with ui.perfetto.dev or Tensorboard.'))
-    parser.add_argument("--device",
-                        type=str,
-                        default="auto",
-                        choices=DEVICE_OPTIONS,
-                        help='device type for vLLM execution')
-    parser.add_argument('--block-size',
-                        type=int,
-                        default=16,
-                        help='block size of key/value cache')
-    parser.add_argument(
-        '--enable-chunked-prefill',
-        action='store_true',
-        help='If True, the prefill requests can be chunked based on the '
-        'max_num_batched_tokens')
-    parser.add_argument("--enable-prefix-caching",
-                        action='store_true',
-                        help="Enable automatic prefix caching")
-    parser.add_argument(
-        "--ray-workers-use-nsight",
-        action='store_true',
-        help="If specified, use nsight to profile ray workers",
-    )
-    parser.add_argument('--download-dir',
-                        type=str,
-                        default=None,
-                        help='directory to download and load the weights, '
-                        'default to the default cache dir of huggingface')
     parser.add_argument(
         '--output-json',
         type=str,
         default=None,
         help='Path to save the latency results in JSON format.')
-    parser.add_argument('--gpu-memory-utilization',
-                        type=float,
-                        default=0.9,
-                        help='the fraction of GPU memory to be used for '
-                        'the model executor, which can range from 0 to 1.'
-                        'If unspecified, will use the default value of 0.9.')
-    parser.add_argument(
-        '--load-format',
-        type=str,
-        default=EngineArgs.load_format,
-        choices=[
-            'auto', 'pt', 'safetensors', 'npcache', 'dummy', 'tensorizer',
-            'bitsandbytes'
-        ],
-        help='The format of the model weights to load.\n\n'
-        '* "auto" will try to load the weights in the safetensors format '
-        'and fall back to the pytorch bin format if safetensors format '
-        'is not available.\n'
-        '* "pt" will load the weights in the pytorch bin format.\n'
-        '* "safetensors" will load the weights in the safetensors format.\n'
-        '* "npcache" will load the weights in pytorch format and store '
-        'a numpy cache to speed up the loading.\n'
-        '* "dummy" will initialize the weights with random values, '
-        'which is mainly for profiling.\n'
-        '* "tensorizer" will load the weights using tensorizer from '
-        'CoreWeave. See the Tensorize vLLM Model script in the Examples'
-        'section for more information.\n'
-        '* "bitsandbytes" will load the weights using bitsandbytes '
-        'quantization.\n')
-    parser.add_argument(
-        '--distributed-executor-backend',
-        choices=['ray', 'mp'],
-        default=None,
-        help='Backend to use for distributed serving. When more than 1 GPU '
-        'is used, will be automatically set to "ray" if installed '
-        'or "mp" (multiprocessing) otherwise.')
-    parser.add_argument(
-        '--otlp-traces-endpoint',
-        type=str,
-        default=None,
-        help='Target URL to which OpenTelemetry traces will be sent.')
+
+    parser = EngineArgs.add_cli_args(parser)
     args = parser.parse_args()
     main(args)

benchmarks/benchmark_prefix_caching.py

@@ -25,6 +25,7 @@ ShareGPT example usage:
         --input-length-range 128:256
 """
 
+import dataclasses
 import json
 import random
 import time
@@ -33,6 +34,7 @@ from typing import List, Optional, Tuple
 from transformers import PreTrainedTokenizerBase
 
 from vllm import LLM, SamplingParams
+from vllm.engine.arg_utils import EngineArgs
 from vllm.utils import FlexibleArgumentParser
 
 try:
@@ -129,12 +131,9 @@ def main(args):
         filtered_datasets = [(PROMPT, prompt_len, args.output_len)
                              ] * args.num_prompts
 
-    llm = LLM(model=args.model,
-              tokenizer_mode='auto',
-              trust_remote_code=True,
-              enforce_eager=True,
-              tensor_parallel_size=args.tensor_parallel_size,
-              enable_prefix_caching=args.enable_prefix_caching)
+    engine_args = EngineArgs.from_cli_args(args)
+
+    llm = LLM(**dataclasses.asdict(engine_args))
 
     sampling_params = SamplingParams(temperature=0, max_tokens=args.output_len)
 
@@ -162,18 +161,11 @@ if __name__ == "__main__":
     parser = FlexibleArgumentParser(
         description=
         'Benchmark the performance with or without automatic prefix caching.')
-    parser.add_argument('--model',
-                        type=str,
-                        default='baichuan-inc/Baichuan2-13B-Chat')
     parser.add_argument("--dataset-path",
                         type=str,
                         default=None,
                         help="Path to the dataset.")
-    parser.add_argument('--tensor-parallel-size', '-tp', type=int, default=1)
     parser.add_argument('--output-len', type=int, default=10)
-    parser.add_argument('--enable-prefix-caching',
-                        action='store_true',
-                        help='enable prefix caching')
     parser.add_argument('--num-prompts',
                         type=int,
                         default=1,
@@ -190,9 +182,7 @@ if __name__ == "__main__":
                         default='128:256',
                         help='Range of input lengths for sampling prompts,'
                         'specified as "min:max" (e.g., "128:256").')
-    parser.add_argument("--seed",
-                        type=int,
-                        default=0,
-                        help='Random seed for reproducibility')
+
+    parser = EngineArgs.add_cli_args(parser)
     args = parser.parse_args()
     main(args)

benchmarks/benchmark_prioritization.py

 """Benchmark offline prioritization."""
 import argparse
+import dataclasses
 import json
 import random
 import time
@@ -7,7 +8,8 @@ from typing import List, Optional, Tuple
 
 from transformers import AutoTokenizer, PreTrainedTokenizerBase
 
-from vllm.model_executor.layers.quantization import QUANTIZATION_METHODS
+from vllm.engine.arg_utils import EngineArgs
+from vllm.utils import FlexibleArgumentParser
 
 
 def sample_requests(
@@ -62,46 +64,11 @@ def sample_requests(
 
 def run_vllm(
     requests: List[Tuple[str, int, int]],
-    model: str,
-    tokenizer: str,
-    quantization: Optional[str],
-    tensor_parallel_size: int,
-    seed: int,
     n: int,
-    trust_remote_code: bool,
-    dtype: str,
-    max_model_len: Optional[int],
-    enforce_eager: bool,
-    kv_cache_dtype: str,
-    quantization_param_path: Optional[str],
-    device: str,
-    enable_prefix_caching: bool,
-    enable_chunked_prefill: bool,
-    max_num_batched_tokens: int,
-    gpu_memory_utilization: float = 0.9,
-    download_dir: Optional[str] = None,
+    engine_args: EngineArgs,
 ) -> float:
     from vllm import LLM, SamplingParams
-    llm = LLM(
-        model=model,
-        tokenizer=tokenizer,
-        quantization=quantization,
-        tensor_parallel_size=tensor_parallel_size,
-        seed=seed,
-        trust_remote_code=trust_remote_code,
-        dtype=dtype,
-        max_model_len=max_model_len,
-        gpu_memory_utilization=gpu_memory_utilization,
-        enforce_eager=enforce_eager,
-        kv_cache_dtype=kv_cache_dtype,
-        quantization_param_path=quantization_param_path,
-        device=device,
-        enable_prefix_caching=enable_prefix_caching,
-        download_dir=download_dir,
-        enable_chunked_prefill=enable_chunked_prefill,
-        max_num_batched_tokens=max_num_batched_tokens,
-        disable_log_stats=False,
-    )
+    llm = LLM(**dataclasses.asdict(engine_args))
 
     # Add the requests to the engine.
     prompts = []
@@ -142,16 +109,8 @@ def main(args: argparse.Namespace):
                                    args.output_len)
 
     if args.backend == "vllm":
-        elapsed_time = run_vllm(requests, args.model, args.tokenizer,
-                                args.quantization, args.tensor_parallel_size,
-                                args.seed, args.n, args.trust_remote_code,
-                                args.dtype, args.max_model_len,
-                                args.enforce_eager, args.kv_cache_dtype,
-                                args.quantization_param_path, args.device,
-                                args.enable_prefix_caching,
-                                args.enable_chunked_prefill,
-                                args.max_num_batched_tokens,
-                                args.gpu_memory_utilization, args.download_dir)
+        elapsed_time = run_vllm(requests, args.n,
+                                EngineArgs.from_cli_args(args))
     else:
         raise ValueError(f"Unknown backend: {args.backend}")
     total_num_tokens = sum(prompt_len + output_len
@@ -173,7 +132,7 @@ def main(args: argparse.Namespace):
 
 
 if __name__ == "__main__":
-    parser = argparse.ArgumentParser(description="Benchmark the throughput.")
+    parser = FlexibleArgumentParser(description="Benchmark the throughput.")
     parser.add_argument("--backend",
                         type=str,
                         choices=["vllm", "hf", "mii"],
@@ -191,13 +150,6 @@ if __name__ == "__main__":
                         default=None,
                         help="Output length for each request. Overrides the "
                         "output length from the dataset.")
-    parser.add_argument("--model", type=str, default="facebook/opt-125m")
-    parser.add_argument("--tokenizer", type=str, default=None)
-    parser.add_argument('--quantization',
-                        '-q',
-                        choices=[*QUANTIZATION_METHODS, None],
-                        default=None)
-    parser.add_argument("--tensor-parallel-size", "-tp", type=int, default=1)
     parser.add_argument("--n",
                         type=int,
                         default=1,
@@ -206,81 +158,13 @@ if __name__ == "__main__":
                         type=int,
                         default=200,
                         help="Number of prompts to process.")
-    parser.add_argument("--seed", type=int, default=0)
-    parser.add_argument('--trust-remote-code',
-                        action='store_true',
-                        help='trust remote code from huggingface')
-    parser.add_argument(
-        '--max-model-len',
-        type=int,
-        default=None,
-        help='Maximum length of a sequence (including prompt and output). '
-        'If None, will be derived from the model.')
-    parser.add_argument(
-        '--dtype',
-        type=str,
-        default='auto',
-        choices=['auto', 'half', 'float16', 'bfloat16', 'float', 'float32'],
-        help='data type for model weights and activations. '
-        'The "auto" option will use FP16 precision '
-        'for FP32 and FP16 models, and BF16 precision '
-        'for BF16 models.')
-    parser.add_argument('--gpu-memory-utilization',
-                        type=float,
-                        default=0.9,
-                        help='the fraction of GPU memory to be used for '
-                        'the model executor, which can range from 0 to 1.'
-                        'If unspecified, will use the default value of 0.9.')
-    parser.add_argument("--enforce-eager",
-                        action="store_true",
-                        help="enforce eager execution")
-    parser.add_argument(
-        '--kv-cache-dtype',
-        type=str,
-        choices=['auto', 'fp8', 'fp8_e5m2', 'fp8_e4m3'],
-        default="auto",
-        help='Data type for kv cache storage. If "auto", will use model '
-        'data type. CUDA 11.8+ supports fp8 (=fp8_e4m3) and fp8_e5m2. '
-        'ROCm (AMD GPU) supports fp8 (=fp8_e4m3)')
-    parser.add_argument(
-        '--quantization-param-path',
-        type=str,
-        default=None,
-        help='Path to the JSON file containing the KV cache scaling factors. '
-        'This should generally be supplied, when KV cache dtype is FP8. '
-        'Otherwise, KV cache scaling factors default to 1.0, which may cause '
-        'accuracy issues. FP8_E5M2 (without scaling) is only supported on '
-        'cuda version greater than 11.8. On ROCm (AMD GPU), FP8_E4M3 is '
-        'instead supported for common inference criteria.')
-    parser.add_argument(
-        "--device",
-        type=str,
-        default="cuda",
-        choices=["cuda", "cpu"],
-        help='device type for vLLM execution, supporting CUDA and CPU.')
-    parser.add_argument(
-        "--enable-prefix-caching",
-        action='store_true',
-        help="enable automatic prefix caching for vLLM backend.")
-    parser.add_argument("--enable-chunked-prefill",
-                        action='store_true',
-                        help="enable chunked prefill for vLLM backend.")
-    parser.add_argument('--max-num-batched-tokens',
-                        type=int,
-                        default=None,
-                        help='maximum number of batched tokens per '
-                        'iteration')
-    parser.add_argument('--download-dir',
-                        type=str,
-                        default=None,
-                        help='directory to download and load the weights, '
-                        'default to the default cache dir of huggingface')
     parser.add_argument(
         '--output-json',
         type=str,
         default=None,
         help='Path to save the throughput results in JSON format.')
 
+    parser = EngineArgs.add_cli_args(parser)
     args = parser.parse_args()
     if args.tokenizer is None:
         args.tokenizer = args.model

benchmarks/benchmark_serving.py

@@ -53,6 +53,8 @@ try:
 except ImportError:
     from argparse import ArgumentParser as FlexibleArgumentParser
 
+MILLISECONDS_TO_SECONDS_CONVERSION = 1000
+
 
 @dataclass
 class BenchmarkMetrics:
@@ -60,6 +62,7 @@ class BenchmarkMetrics:
     total_input: int
     total_output: int
     request_throughput: float
+    request_goodput: float
     output_throughput: float
     total_token_throughput: float
     mean_ttft_ms: float
@@ -202,6 +205,7 @@ def sample_hf_requests(
     dataset_split: str,
     num_requests: int,
     tokenizer: PreTrainedTokenizerBase,
+    random_seed: int,
     fixed_output_len: Optional[int] = None,
 ) -> List[Tuple[str, str, int, Optional[Dict[str, Collection[str]]]]]:
     dataset = load_dataset(dataset_path,
@@ -210,8 +214,8 @@ def sample_hf_requests(
                            streaming=True)
     assert "conversations" in dataset.features, (
         "HF Dataset must have 'conversations' column.")
-    filtered_dataset = dataset.shuffle().filter(
-        lambda x: len(x["conversations"]) >= 2)
+    filter_func = lambda x: len(x["conversations"]) >= 2
+    filtered_dataset = dataset.shuffle(seed=random_seed).filter(filter_func)
     sampled_requests: List[Tuple[str, int, int, Dict[str,
                                                      Collection[str]]]] = []
     for data in filtered_dataset:
@@ -315,12 +319,15 @@ def calculate_metrics(
     tokenizer: PreTrainedTokenizerBase,
     selected_percentile_metrics: List[str],
     selected_percentiles: List[float],
+    gootput_config_dict: Dict[str, float],
 ) -> Tuple[BenchmarkMetrics, List[int]]:
     actual_output_lens: List[int] = []
     total_input = 0
     completed = 0
+    good_completed = 0
     itls: List[float] = []
     tpots: List[float] = []
+    all_tpots: List[float] = []
     ttfts: List[float] = []
     e2els: List[float] = []
     for i in range(len(outputs)):
@@ -334,9 +341,13 @@ def calculate_metrics(
                           add_special_tokens=False).input_ids)
             actual_output_lens.append(output_len)
             total_input += input_requests[i][1]
+            tpot = 0
             if output_len > 1:
-                tpots.append(
-                    (outputs[i].latency - outputs[i].ttft) / (output_len - 1))
+                tpot = (outputs[i].latency - outputs[i].ttft) / (output_len -
+                                                                 1)
+                tpots.append(tpot)
+            # Note: if output_len <= 1, we regard tpot as 0 for goodput
+            all_tpots.append(tpot)
             itls += outputs[i].itl
             ttfts.append(outputs[i].ttft)
             e2els.append(outputs[i].latency)
@@ -344,6 +355,28 @@ def calculate_metrics(
         else:
             actual_output_lens.append(0)
 
+    if gootput_config_dict:
+        valid_metrics = []
+        slo_values = []
+
+        if "ttft" in gootput_config_dict:
+            valid_metrics.append(ttfts)
+            slo_values.append(gootput_config_dict["ttft"] /
+                              MILLISECONDS_TO_SECONDS_CONVERSION)
+        if "tpot" in gootput_config_dict:
+            valid_metrics.append(all_tpots)
+            slo_values.append(gootput_config_dict["tpot"] /
+                              MILLISECONDS_TO_SECONDS_CONVERSION)
+        if "e2el" in gootput_config_dict:
+            valid_metrics.append(e2els)
+            slo_values.append(gootput_config_dict["e2el"] /
+                              MILLISECONDS_TO_SECONDS_CONVERSION)
+
+        for req_metric in zip(*valid_metrics):
+            is_good_req = all([s >= r for s, r in zip(slo_values, req_metric)])
+            if is_good_req:
+                good_completed += 1
+
     if completed == 0:
         warnings.warn(
             "All requests failed. This is likely due to a misconfiguration "
@@ -354,6 +387,7 @@ def calculate_metrics(
         total_input=total_input,
         total_output=sum(actual_output_lens),
         request_throughput=completed / dur_s,
+        request_goodput=good_completed / dur_s,
         output_throughput=sum(actual_output_lens) / dur_s,
         total_token_throughput=(total_input + sum(actual_output_lens)) / dur_s,
         mean_ttft_ms=np.mean(ttfts or 0) *
@@ -397,6 +431,8 @@ async def benchmark(
     selected_percentile_metrics: List[str],
     selected_percentiles: List[str],
     ignore_eos: bool,
+    gootput_config_dict: Dict[str, float],
+    max_concurrency: Optional[int],
 ):
     if backend in ASYNC_REQUEST_FUNCS:
         request_func = ASYNC_REQUEST_FUNCS[backend]
@@ -445,9 +481,25 @@ async def benchmark(
             print("Profiler started")
 
     print(f"Traffic request rate: {request_rate}")
+    print(f"Maximum request concurrency: {max_concurrency}")
 
     pbar = None if disable_tqdm else tqdm(total=len(input_requests))
 
+    # This can be used once the minimum Python version is 3.10 or higher,
+    # and it will simplify the code in limited_request_func.
+    #    semaphore = (asyncio.Semaphore(max_concurrency)
+    #                 if max_concurrency else contextlib.nullcontext())
+    semaphore = (asyncio.Semaphore(max_concurrency)
+                 if max_concurrency else None)
+
+    async def limited_request_func(request_func_input, pbar):
+        if semaphore is None:
+            return await request_func(request_func_input=request_func_input,
+                                      pbar=pbar)
+        async with semaphore:
+            return await request_func(request_func_input=request_func_input,
+                                      pbar=pbar)
+
     benchmark_start_time = time.perf_counter()
     tasks: List[asyncio.Task] = []
     async for request in get_request(input_requests, request_rate):
@@ -463,8 +515,8 @@ async def benchmark(
                                               ignore_eos=ignore_eos)
         tasks.append(
             asyncio.create_task(
-                request_func(request_func_input=request_func_input,
-                             pbar=pbar)))
+                limited_request_func(request_func_input=request_func_input,
+                                     pbar=pbar)))
     outputs: List[RequestFuncOutput] = await asyncio.gather(*tasks)
 
     if profile:
@@ -494,6 +546,7 @@ async def benchmark(
         tokenizer=tokenizer,
         selected_percentile_metrics=selected_percentile_metrics,
         selected_percentiles=selected_percentiles,
+        gootput_config_dict=gootput_config_dict,
     )
 
     print("{s:{c}^{n}}".format(s=' Serving Benchmark Result ', n=50, c='='))
@@ -505,6 +558,9 @@ async def benchmark(
                                  metrics.total_output))
     print("{:<40} {:<10.2f}".format("Request throughput (req/s):",
                                     metrics.request_throughput))
+    if gootput_config_dict:
+        print("{:<40} {:<10.2f}".format("Request goodput (req/s):",
+                                        metrics.request_goodput))
     print("{:<40} {:<10.2f}".format("Output token throughput (tok/s):",
                                     metrics.output_throughput))
     print("{:<40} {:<10.2f}".format("Total Token throughput (tok/s):",
@@ -516,6 +572,8 @@ async def benchmark(
         "total_input_tokens": metrics.total_input,
         "total_output_tokens": metrics.total_output,
         "request_throughput": metrics.request_throughput,
+        "request_goodput:":
+        metrics.request_goodput if gootput_config_dict else None,
         "output_throughput": metrics.output_throughput,
         "total_token_throughput": metrics.total_token_throughput,
         "input_lens": [output.prompt_len for output in outputs],
@@ -569,6 +627,41 @@ async def benchmark(
     return result
 
 
+def check_goodput_args(args):
+    # Check and parse goodput arguments
+    gootput_config_dict = {}
+    VALID_NAMES = ["ttft", "tpot", "e2el"]
+    if args.goodput:
+        gootput_config_dict = parse_goodput(args.goodput)
+        for slo_name, slo_val in gootput_config_dict.items():
+            if slo_name not in VALID_NAMES:
+                raise ValueError(
+                    f"Invalid metric name found, {slo_name}: {slo_val}. "
+                    "The service level objective name should be one of "
+                    f"{str(VALID_NAMES)}. ")
+            if slo_val < 0:
+                raise ValueError(
+                    f"Invalid value found, {slo_name}: {slo_val}. "
+                    "The service level objective value should be "
+                    "non-negative.")
+    return gootput_config_dict
+
+
+def parse_goodput(slo_pairs):
+    gootput_config_dict = {}
+    try:
+        for slo_pair in slo_pairs:
+            slo_name, slo_val = slo_pair.split(":")
+            gootput_config_dict[slo_name] = float(slo_val)
+    except ValueError as err:
+        raise argparse.ArgumentTypeError(
+            "Invalid format found for service level objectives. "
+            "Specify service level objectives for goodput as \"KEY:VALUE\" "
+            "pairs, where the key is a metric name, and the value is a "
+            "number in milliseconds.") from err
+    return gootput_config_dict
+
+
 def main(args: argparse.Namespace):
     print(args)
     random.seed(args.seed)
@@ -646,6 +739,7 @@ def main(args: argparse.Namespace):
             dataset_split=args.hf_split,
             num_requests=args.num_prompts,
             tokenizer=tokenizer,
+            random_seed=args.seed,
             fixed_output_len=args.hf_output_len,
         )
 
@@ -662,6 +756,8 @@ def main(args: argparse.Namespace):
     else:
         raise ValueError(f"Unknown dataset: {args.dataset_name}")
 
+    gootput_config_dict = check_goodput_args(args)
+
     benchmark_result = asyncio.run(
         benchmark(
             backend=backend,
@@ -680,6 +776,8 @@ def main(args: argparse.Namespace):
                 float(p) for p in args.metric_percentiles.split(",")
             ],
             ignore_eos=args.ignore_eos,
+            gootput_config_dict=gootput_config_dict,
+            max_concurrency=args.max_concurrency,
         ))
 
     # Save config and results to json
@@ -709,13 +807,16 @@ def main(args: argparse.Namespace):
         # Traffic
         result_json["request_rate"] = (
             args.request_rate if args.request_rate < float("inf") else "inf")
+        result_json["max_concurrency"] = args.max_concurrency
 
         # Merge with benchmark result
         result_json = {**result_json, **benchmark_result}
 
         # Save to file
         base_model_id = model_id.split("/")[-1]
-        file_name = f"{backend}-{args.request_rate}qps-{base_model_id}-{current_dt}.json"  #noqa
+        max_concurrency_str = (f"-concurrency{args.max_concurrency}"
+                               if args.max_concurrency is not None else "")
+        file_name = f"{backend}-{args.request_rate}qps{max_concurrency_str}-{base_model_id}-{current_dt}.json"  #noqa
         if args.result_filename:
             file_name = args.result_filename
         if args.result_dir:
@@ -766,6 +867,19 @@ if __name__ == "__main__":
                         default=None,
                         help="Path to the sharegpt/sonnet dataset. "
                         "Or the huggingface dataset ID if using HF dataset.")
+    parser.add_argument(
+        "--max-concurrency",
+        type=int,
+        default=None,
+        help="Maximum number of concurrent requests. This can be used "
+        "to help simulate an environment where a higher level component "
+        "is enforcing a maximum number of concurrent requests. While the "
+        "--request-rate argument controls the rate at which requests are "
+        "initiated, this argument will control how many are actually allowed "
+        "to execute at a time. This means that when used in combination, the "
+        "actual request rate may be lower than specified with --request-rate, "
+        "if the server is not processing requests fast enough to keep up.")
+
     parser.add_argument(
         "--model",
         type=str,
@@ -879,6 +993,17 @@ if __name__ == "__main__":
         "Default value is \"99\". "
         "Use \"--percentile-metrics\" to select metrics.",
     )
+    parser.add_argument(
+        "--goodput",
+        nargs="+",
+        required=False,
+        help="Specify service level objectives for goodput as \"KEY:VALUE\" "
+        "pairs, where the key is a metric name, and the value is in "
+        "milliseconds. Multiple \"KEY:VALUE\" pairs can be provided, "
+        "separated by spaces. Allowed request level metric names are "
+        "\"ttft\", \"tpot\", \"e2el\". For more context on the definition of "
+        "goodput, refer to DistServe paper: https://arxiv.org/pdf/2401.09670 "
+        "and the blog: https://hao-ai-lab.github.io/blogs/distserve")
 
     # group for dataset specific arguments
     sonnet_group = parser.add_argument_group("sonnet dataset options")