Sync with upstream @ 0.6.1.post2

.buildkite/run-cpu-test.sh

@@ -23,12 +23,10 @@ docker exec cpu-test-avx2 bash -c "python3 examples/offline_inference.py"
 # Run basic model test
 docker exec cpu-test bash -c "
   pip install pytest matplotlib einops transformers_stream_generator
-  pytest -v -s tests/models -m \"not vlm\" --ignore=tests/models/test_embedding.py \
-      --ignore=tests/models/test_oot_registration.py \
-      --ignore=tests/models/test_registry.py \
-      --ignore=tests/models/test_fp8.py \
-      --ignore=tests/models/test_jamba.py \
-      --ignore=tests/models/test_danube3_4b.py" # Mamba and Danube3-4B on CPU is not supported
+  pytest -v -s tests/models/decoder_only/language \
+    --ignore=tests/models/test_fp8.py \
+    --ignore=tests/models/decoder_only/language/test_jamba.py \
+    --ignore=tests/models/decoder_only/language/test_danube3_4b.py" # Mamba and Danube3-4B on CPU is not supported
 
 # Run compressed-tensor test
 docker exec cpu-test bash -c "

.buildkite/test-pipeline.yaml

@@ -94,7 +94,6 @@ steps:
   - pytest -v -s entrypoints/test_chat_utils.py
   - pytest -v -s entrypoints/offline_mode # Needs to avoid interference with other tests
 
-
 - label: Distributed Tests (4 GPUs) # 10min
   working_dir: "/vllm-workspace/tests"
   num_gpus: 4
@@ -164,30 +163,13 @@ steps:
     - python3 tensorize_vllm_model.py --model facebook/opt-125m serialize --serialized-directory /tmp/ --suffix v1 && python3 tensorize_vllm_model.py --model facebook/opt-125m deserialize --path-to-tensors /tmp/vllm/facebook/opt-125m/v1/model.tensors
     - python3 offline_inference_encoder_decoder.py
 
-- label: Models Test # 1hr10min
-  source_file_dependencies:
-  - vllm/
-  - tests/models
-  commands:
-    - pip install -e ./plugins/vllm_add_dummy_model
-    - pytest -v -s models/test_oot_registration.py # it needs a clean process
-    - pytest -v -s models -m \"not vlm\" --ignore=models/test_oot_registration.py
-
 - label: torch compile integration test
   source_file_dependencies:
   - vllm/
   commands:
     - pytest -v -s ./compile/test_full_graph.py
     - pytest -v -s ./compile/test_wrapper.py
 
-
-- label: Vision Language Models Test # 42min
-  #mirror_hardwares: [amd]
-  source_file_dependencies:
-  - vllm/
-  commands:
-    - pytest -v -s models -m vlm
-
 - label: Prefix Caching Test # 7min
   #mirror_hardwares: [amd]
   source_file_dependencies:
@@ -286,6 +268,45 @@ steps:
   commands:
     - pytest -v -s tool_use
 
+#####  models test  #####
+
+- label: Basic Models Test # 3min
+  source_file_dependencies:
+  - vllm/
+  - tests/models
+  commands:
+    - pip install -e ./plugins/vllm_add_dummy_model
+    - 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 # 1h3min
+  #mirror_hardwares: [amd]
+  source_file_dependencies:
+  - vllm/
+  - tests/models/decoder_only/language
+  commands:
+    - pytest -v -s models/decoder_only/language
+
+- label: Decoder-only Multi-Modal Models Test # 56min
+  #mirror_hardwares: [amd]
+  source_file_dependencies:
+  - vllm/
+  - tests/models/decoder_only/audio_language
+  - tests/models/decoder_only/vision_language
+  commands:
+    - pytest -v -s models/decoder_only/audio_language
+    - pytest -v -s models/decoder_only/vision_language
+
+- label: Other Models Test # 5min
+  #mirror_hardwares: [amd]
+  source_file_dependencies:
+  - vllm/
+  - tests/models/embedding/language
+  - tests/models/encoder_decoder/language
+  commands:
+    - pytest -v -s models/embedding/language
+    - pytest -v -s models/encoder_decoder/language
+
 #####  1 GPU test  #####
 #####  multi gpus test  #####
 
@@ -311,11 +332,11 @@ steps:
   - tests/distributed/
   commands:
   - # the following commands are for the first node, with ip 192.168.10.10 (ray environment already set up)
-    - VLLM_TEST_SAME_HOST=0 torchrun --nnodes 2 --nproc-per-node=2 --rdzv_backend=c10d --rdzv_endpoint=192.168.10.10 distributed/test_same_node.py
+    - VLLM_TEST_SAME_HOST=0 torchrun --nnodes 2 --nproc-per-node=2 --rdzv_backend=c10d --rdzv_endpoint=192.168.10.10 distributed/test_same_node.py | grep -q 'Same node test passed'
     - VLLM_MULTI_NODE=1 pytest -v -s distributed/test_multi_node_assignment.py
     - VLLM_MULTI_NODE=1 pytest -v -s distributed/test_pipeline_parallel.py
   - # the following commands are for the second node, with ip 192.168.10.11 (ray environment already set up)
-    - VLLM_TEST_SAME_HOST=0 torchrun --nnodes 2 --nproc-per-node=2 --rdzv_backend=c10d --rdzv_endpoint=192.168.10.10 distributed/test_same_node.py
+    - VLLM_TEST_SAME_HOST=0 torchrun --nnodes 2 --nproc-per-node=2 --rdzv_backend=c10d --rdzv_endpoint=192.168.10.10 distributed/test_same_node.py | grep -q 'Same node test passed'
 
 - label: Distributed Tests (2 GPUs) # 28min
   #mirror_hardwares: [amd]
@@ -328,11 +349,10 @@ steps:
   - vllm/model_executor/models/
   - tests/distributed/
   commands:
-  - VLLM_TEST_SAME_HOST=1 torchrun --nproc-per-node=4 distributed/test_same_node.py
-  - TARGET_TEST_SUITE=L4 pytest -v -s distributed/test_basic_distributed_correctness.py
-  - pytest -v -s distributed/test_basic_distributed_correctness_enc_dec.py
-  - pytest -v -s distributed/test_chunked_prefill_distributed.py
-  - pytest -v -s distributed/test_multimodal_broadcast.py
+  - VLLM_TEST_SAME_HOST=1 torchrun --nproc-per-node=4 distributed/test_same_node.py | grep -q 'Same node test passed'
+  - TARGET_TEST_SUITE=L4 pytest basic_correctness/ -v -s -m distributed_2_gpus
+  # Avoid importing model tests that cause CUDA reinitialization error
+  - pytest models/encoder_decoder/language/test_bart.py models/decoder_only/vision_language/test_broadcast.py -v -s -m distributed_2_gpus
   - pytest -v -s spec_decode/e2e/test_integration_dist_tp2.py
   - pip install -e ./plugins/vllm_add_dummy_model
   - pytest -v -s distributed/test_distributed_oot.py

docs/source/getting_started/cpu-installation.rst

@@ -59,6 +59,20 @@ Build from source
     $ pip install wheel packaging ninja "setuptools>=49.4.0" numpy
     $ pip install -v -r requirements-cpu.txt --extra-index-url https://download.pytorch.org/whl/cpu
 
+- Third, build and install oneDNN library from source:
+
+.. code-block:: console
+
+    $ git clone -b rls-v3.5 https://github.com/oneapi-src/oneDNN.git
+    $ cmake -B ./oneDNN/build -S ./oneDNN -G Ninja -DONEDNN_LIBRARY_TYPE=STATIC \ 
+        -DONEDNN_BUILD_DOC=OFF \ 
+        -DONEDNN_BUILD_EXAMPLES=OFF \ 
+        -DONEDNN_BUILD_TESTS=OFF \ 
+        -DONEDNN_BUILD_GRAPH=OFF \ 
+        -DONEDNN_ENABLE_WORKLOAD=INFERENCE \ 
+        -DONEDNN_ENABLE_PRIMITIVE=MATMUL
+    $ cmake --build ./oneDNN/build --target install --config Release
+
 - Finally, build and install vLLM CPU backend: 
 
 .. code-block:: console

docs/source/getting_started/installation.rst

@@ -26,6 +26,10 @@ You can install vLLM using pip:
     $ # Install vLLM with CUDA 12.1.
     $ pip install vllm
 
+.. note::
+
+    Although we recommend using ``conda`` to create and manage Python environments, it is highly recommended to use ``pip`` to install vLLM. This is because ``pip`` can install ``torch`` with separate library packages like ``NCCL``, while ``conda`` installs ``torch`` with statically linked ``NCCL``. This can cause issues when vLLM tries to use ``NCCL``. See `this issue <https://github.com/vllm-project/vllm/issues/8420>`_ for more details.
+
 .. note::
 
     As of now, vLLM's binaries are compiled with CUDA 12.1 and public PyTorch release versions by default.
@@ -34,7 +38,7 @@ You can install vLLM using pip:
     .. code-block:: console
 
         $ # Install vLLM with CUDA 11.8.
-        $ export VLLM_VERSION=0.4.0
+        $ export VLLM_VERSION=0.6.1.post1
         $ export PYTHON_VERSION=310
         $ pip install https://github.com/vllm-project/vllm/releases/download/v${VLLM_VERSION}/vllm-${VLLM_VERSION}+cu118-cp${PYTHON_VERSION}-cp${PYTHON_VERSION}-manylinux1_x86_64.whl --extra-index-url https://download.pytorch.org/whl/cu118
 
@@ -48,7 +52,7 @@ You can install vLLM using pip:
 
     .. code-block:: console
 
-        $ export VLLM_VERSION=0.5.4 # vLLM's main branch version is currently set to latest released tag
+        $ export VLLM_VERSION=0.6.1.post1 # vLLM's main branch version is currently set to latest released tag
         $ pip install https://vllm-wheels.s3.us-west-2.amazonaws.com/nightly/vllm-${VLLM_VERSION}-cp38-abi3-manylinux1_x86_64.whl
         $ # You can also access a specific commit
         $ # export VLLM_COMMIT=...
@@ -80,11 +84,11 @@ You can also build and install vLLM from source:
 
 .. tip::
 
-    Building from source requires quite a lot compilation. If you are building from source for multiple times, it is beneficial to cache the compilation results. For example, you can install `ccache <https://github.com/ccache/ccache>`_ via either `conda install ccache` or `apt install ccache` . As long as `which ccache` command can find the `ccache` binary, it will be used automatically by the build system. After the first build, the subsequent builds will be much faster.
+    Building from source requires quite a lot compilation. If you are building from source for multiple times, it is beneficial to cache the compilation results. For example, you can install `ccache <https://github.com/ccache/ccache>`_ via either ``conda install ccache`` or ``apt install ccache`` . As long as ``which ccache`` command can find the ``ccache`` binary, it will be used automatically by the build system. After the first build, the subsequent builds will be much faster.
 
 .. tip::
     To avoid your system being overloaded, you can limit the number of compilation jobs
-    to be run simultaneously, via the environment variable `MAX_JOBS`. For example:
+    to be run simultaneously, via the environment variable ``MAX_JOBS``. For example:
 
     .. code-block:: console
 
@@ -99,7 +103,7 @@ You can also build and install vLLM from source:
         $ # Use `--ipc=host` to make sure the shared memory is large enough.
         $ docker run --gpus all -it --rm --ipc=host nvcr.io/nvidia/pytorch:23.10-py3
 
-    If you don't want to use docker, it is recommended to have a full installation of CUDA Toolkit. You can download and install it from `the official website <https://developer.nvidia.com/cuda-toolkit-archive>`_. After installation, set the environment variable `CUDA_HOME` to the installation path of CUDA Toolkit, and make sure that the `nvcc` compiler is in your `PATH`, e.g.:
+    If you don't want to use docker, it is recommended to have a full installation of CUDA Toolkit. You can download and install it from `the official website <https://developer.nvidia.com/cuda-toolkit-archive>`_. After installation, set the environment variable ``CUDA_HOME`` to the installation path of CUDA Toolkit, and make sure that the ``nvcc`` compiler is in your ``PATH``, e.g.:
 
     .. code-block:: console
 

docs/source/models/supported_models.rst

@@ -342,7 +342,7 @@ Note that, as an inference engine, vLLM does not introduce new models. Therefore
 
 We have the following levels of testing for models:
 
-1. **Strict Consistency**: We compare the output of the model with the output of the model in the HuggingFace Transformers library under greedy decoding. This is the most stringent test. Please refer to `test_models.py <https://github.com/vllm-project/vllm/blob/main/tests/models/test_models.py>`_ and `test_big_models.py <https://github.com/vllm-project/vllm/blob/main/tests/models/test_big_models.py>`_ for the models that have passed this test.
+1. **Strict Consistency**: We compare the output of the model with the output of the model in the HuggingFace Transformers library under greedy decoding. This is the most stringent test. Please refer to `models tests <https://github.com/vllm-project/vllm/blob/main/tests/models>`_ for the models that have passed this test.
 2. **Output Sensibility**: We check if the output of the model is sensible and coherent, by measuring the perplexity of the output and checking for any obvious errors. This is a less stringent test.
 3. **Runtime Functionality**: We check if the model can be loaded and run without errors. This is the least stringent test. Please refer to `functionality tests <https://github.com/vllm-project/vllm/tree/main/tests>`_ and `examples <https://github.com/vllm-project/vllm/tree/main/examples>`_ for the models that have passed this test.
 4. **Community Feedback**: We rely on the community to provide feedback on the models. If a model is broken or not working as expected, we encourage users to raise issues to report it or open pull requests to fix it. The rest of the models fall under this category.

pyproject.toml

@@ -85,5 +85,6 @@ skip_gitignore = true
 [tool.pytest.ini_options]
 markers = [
     "skip_global_cleanup",
-    "vlm: run tests for vision language models only",
+    "core_model: run this model test in each PR instead of just daily",
+    "distributed_2_gpus: run this test only in distributed tests for 2 GPUs",
 ]

requirements-common.txt

@@ -7,7 +7,8 @@ py-cpuinfo
 transformers >= 4.43.2  # Required for Chameleon and Llama 3.1 hotfox.
 tokenizers >= 0.19.1  # Required for Llama 3.
 protobuf # Required by LlamaTokenizer.
-fastapi >= 0.114.1
+fastapi < 0.113.0; python_version < '3.9'
+fastapi >= 0.114.1; python_version >= '3.9'
 aiohttp
 openai >= 1.40.0 # Ensure modern openai package (ensure types module present)
 uvicorn[standard]

tests/async_engine/test_async_llm_engine.py

@@ -159,7 +159,8 @@ def should_do_global_cleanup_after_test(request) -> bool:
 
 
 @pytest.mark.asyncio(scope="module")
-async def test_asyncio_run(async_engine):
+@pytest.mark.parametrize("stop", [None, ["a stop string"]])
+async def test_asyncio_run(async_engine, stop):
 
     scheduler_config = await async_engine.get_scheduler_config()
     num_scheduler_steps = scheduler_config.num_scheduler_steps
@@ -169,6 +170,7 @@ async def run(prompt: str):
             temperature=0,
             max_tokens=32,
             min_tokens=32,
+            stop=stop,
         )
 
         output_count = 0
@@ -203,7 +205,8 @@ async def run(prompt: str):
 
 
 @pytest.mark.asyncio(scope="module")
-async def test_output_kinds(async_engine):
+@pytest.mark.parametrize("stop", [None, ["a stop string"]])
+async def test_output_kinds(async_engine, stop):
     """Test that output_kind works as expected and that
     results are equivalent across different kinds."""
 
@@ -214,6 +217,7 @@ async def test_output_kinds(async_engine):
         temperature=0,
         max_tokens=32,
         min_tokens=32,
+        stop=stop,
     )
 
     async def run(prompt: str, kind: RequestOutputKind):
@@ -229,6 +233,8 @@ async def run(prompt: str, kind: RequestOutputKind):
             final_output = output
 
         assert final_output is not None
+        assert final_output.finished
+
         return (final_output.prompt_token_ids,
                 final_output.outputs[0].token_ids,
                 final_output.outputs[0].text, output_count)
@@ -241,16 +247,18 @@ async def run_deltas(prompt: str):
         output_tokens: List[int] = []
         output_text = ""
         output_count = 0
+        final_output = None
         async for output in async_engine.generate(prompt,
                                                   params,
                                                   request_id=uid()):
             token_ids = output.outputs[0].token_ids
             text = output.outputs[0].text
+            final_output = output
 
             # Ensure we get prompt ids iff we haven't yet received output tokens
             if output_tokens:
                 assert 1 <= len(token_ids) <= num_scheduler_steps
-                assert text
+                assert stop or text
                 assert not output.prompt_token_ids
             else:
                 assert output.prompt_token_ids
@@ -260,6 +268,10 @@ async def run_deltas(prompt: str):
             output_text += text
 
             output_count += 1
+
+        assert final_output is not None
+        assert final_output.finished
+
         return prompt_tokens, output_tokens, output_text, output_count
 
     results = await asyncio.gather(
@@ -291,14 +303,16 @@ async def run_deltas(prompt: str):
 
 
 @pytest.mark.asyncio(scope="module")
-async def test_cancellation(async_engine):
+@pytest.mark.parametrize("stop", [None, ["a stop string"]])
+async def test_cancellation(async_engine, stop):
     scheduler_config = await async_engine.get_scheduler_config()
     num_scheduler_steps = scheduler_config.num_scheduler_steps
 
     sampling_params = SamplingParams(
         temperature=0,
         min_tokens=13,
         max_tokens=13,
+        stop=stop,
     )
 
     stop_at = 5 if num_scheduler_steps == 1 else 1
@@ -319,7 +333,8 @@ async def test_cancellation(async_engine):
 
 
 @pytest.mark.asyncio(scope="module")
-async def test_delayed_generator(async_engine):
+@pytest.mark.parametrize("stop", [None, ["a stop string"]])
+async def test_delayed_generator(async_engine, stop):
     scheduler_config = await async_engine.get_scheduler_config()
 
     if scheduler_config.num_scheduler_steps != 1:
@@ -329,6 +344,7 @@ async def test_delayed_generator(async_engine):
         temperature=0,
         min_tokens=10,
         max_tokens=10,
+        stop=stop,
     )
 
     stream = async_engine.generate("test3", sampling_params, request_id=uid())