merge v0.4.0

.buildkite/download-images.sh

+#!/bin/bash
+
+set -ex
+set -o pipefail
+
+(which wget && which curl) || (apt-get update && apt-get install -y wget curl)
+
+# aws s3 sync s3://air-example-data-2/vllm_opensource_llava/ images/
+mkdir -p images
+cd images
+wget https://air-example-data-2.s3.us-west-2.amazonaws.com/vllm_opensource_llava/stop_sign_pixel_values.pt
+wget https://air-example-data-2.s3.us-west-2.amazonaws.com/vllm_opensource_llava/stop_sign_image_features.pt
+wget https://air-example-data-2.s3.us-west-2.amazonaws.com/vllm_opensource_llava/cherry_blossom_pixel_values.pt
+wget https://air-example-data-2.s3.us-west-2.amazonaws.com/vllm_opensource_llava/cherry_blossom_image_features.pt
+wget https://air-example-data-2.s3.us-west-2.amazonaws.com/vllm_opensource_llava/stop_sign.jpg
+wget https://air-example-data-2.s3.us-west-2.amazonaws.com/vllm_opensource_llava/cherry_blossom.jpg
+
+cd -

.buildkite/run-amd-test.sh

+# This script build the ROCm docker image and run the API server inside the container.
+# It serves a sanity check for compilation and basic model usage.
+set -ex
+
+# Print ROCm version
+rocminfo
+
+# Try building the docker image
+docker build -t rocm -f Dockerfile.rocm .
+
+# Setup cleanup
+remove_docker_container() { docker rm -f rocm || true; }
+trap remove_docker_container EXIT
+remove_docker_container
+
+# Run the image
+docker run --device /dev/kfd --device /dev/dri --network host --name rocm rocm python3 -m vllm.entrypoints.api_server &
+
+# Wait for the server to start
+wait_for_server_to_start() {
+    timeout=300
+    counter=0
+
+    while [ "$(curl -s -o /dev/null -w ''%{http_code}'' localhost:8000/health)" != "200" ]; do
+        sleep 1
+        counter=$((counter + 1))
+        if [ $counter -ge $timeout ]; then
+            echo "Timeout after $timeout seconds"
+            break
+        fi
+    done
+}
+wait_for_server_to_start
+
+# Test a simple prompt
+curl -X POST -H "Content-Type: application/json" \
+    localhost:8000/generate \
+    -d '{"prompt": "San Francisco is a"}'

.buildkite/run-benchmarks.sh

@@ -23,8 +23,9 @@ wget https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered/r
 # wait for server to start, timeout after 600 seconds
 timeout 600 bash -c 'until curl localhost:8000/v1/models; do sleep 1; done' || exit 1
 python3 benchmarks/benchmark_serving.py \
-    --backend openai \
-    --dataset ./ShareGPT_V3_unfiltered_cleaned_split.json \
+    --backend vllm \
+    --dataset-name sharegpt \
+    --dataset-path ./ShareGPT_V3_unfiltered_cleaned_split.json \
     --model meta-llama/Llama-2-7b-chat-hf \
     --num-prompts 20 \
     --endpoint /v1/completions \
@@ -48,7 +49,9 @@ sed -n '$p' benchmark_throughput.txt >> benchmark_results.md # last line
 echo "### Serving Benchmarks" >> benchmark_results.md
 sed -n '1p' benchmark_serving.txt >> benchmark_results.md # first line
 echo "" >> benchmark_results.md
-tail -n 13 benchmark_serving.txt >> benchmark_results.md # last 13 lines
+echo '```' >> benchmark_results.md
+tail -n 20 benchmark_serving.txt >> benchmark_results.md # last 20 lines
+echo '```' >> benchmark_results.md
 
 # upload the results to buildkite
 /workspace/buildkite-agent annotate --style "info" --context "benchmark-results" < benchmark_results.md

.buildkite/test-pipeline.yaml

@@ -12,45 +12,73 @@ steps:
   command: pytest -v -s async_engine
 
 - label: Basic Correctness Test
-  command: pytest -v -s --forked basic_correctness
+  command: pytest -v -s basic_correctness
+
+- label: Core Test
+  command: pytest -v -s core
 
 - label: Distributed Comm Ops Test
-  command: pytest -v -s --forked test_comm_ops.py
+  command: pytest -v -s test_comm_ops.py
   working_dir: "/vllm-workspace/tests/distributed"
   num_gpus: 2 # only support 1 or 2 for now.
 
-- label: Distributed Correctness Test
-  command: pytest -v -s --forked test_basic_distributed_correctness.py
+- label: Distributed Tests
   working_dir: "/vllm-workspace/tests/distributed"
   num_gpus: 2 # only support 1 or 2 for now.
+  commands:
+  - pytest -v -s test_pynccl.py
+  - TEST_DIST_MODEL=facebook/opt-125m pytest -v -s test_basic_distributed_correctness.py
+  - TEST_DIST_MODEL=meta-llama/Llama-2-7b-hf pytest -v -s test_basic_distributed_correctness.py
 
 - label: Engine Test
-  command: pytest -v -s engine
+  command: pytest -v -s engine tokenization test_sequence.py test_config.py
 
 - label: Entrypoints Test
   command: pytest -v -s entrypoints
 
-- label: Kernels Test
-  command: pytest -v -s kernels
-  soft_fail: true
+- label: Examples Test
+  working_dir: "/vllm-workspace/examples"
+  commands:
+    # install aws cli for llava_example.py
+    - pip install awscli
+    - python3 offline_inference.py
+    - python3 offline_inference_with_prefix.py
+    - python3 llm_engine_example.py
+    - python3 llava_example.py
+
+- label: Kernels Test %N
+  command: pytest -v -s kernels --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
+  parallelism: 4
 
 - label: Models Test
   commands:
-    - pytest -v -s models --forked
-  soft_fail: true
+    - bash ../.buildkite/download-images.sh
+    - pytest -v -s models --ignore=models/test_llava.py --ignore=models/test_mistral.py
+
+- label: Llava Test
+  commands:
+    - bash ../.buildkite/download-images.sh
+    - pytest -v -s models/test_llava.py
 
 - label: Prefix Caching Test
   commands:
     - pytest -v -s prefix_caching
 
 - label: Samplers Test
-  command: pytest -v -s samplers --forked
+  command: pytest -v -s samplers
+
+- label: LogitsProcessor Test
+  command: pytest -v -s test_logits_processor.py
 
 - label: Worker Test
   command: pytest -v -s worker
 
-- label: LoRA Test
-  command: pytest -v -s lora --forked
+- label: Speculative decoding tests
+  command: pytest -v -s spec_decode
+
+- label: LoRA Test %N
+  command: pytest -v -s lora --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
+  parallelism: 4
 
 - label: Metrics Test
   command: pytest -v -s metrics

.buildkite/test-template.j2

@@ -3,6 +3,11 @@
 {% set default_working_dir = "/vllm-workspace/tests" %}
 
 steps:
+  - label: "AMD Test"
+    agents:
+      queue: amd
+    command: bash .buildkite/run-amd-test.sh
+
   - label: ":docker: build image"
     commands:
       - "docker build --build-arg max_jobs=16 --tag {{ docker_image }} --target test --progress plain ."
@@ -20,6 +25,9 @@ steps:
     agents:
       queue: kubernetes
     soft_fail: {{ step.soft_fail or false }}
+    {% if step.parallelism %}
+    parallelism: {{ step.parallelism }}
+    {% endif %}
     retry:
       automatic:
         - exit_status: -1  # Agent was lost
@@ -45,6 +53,8 @@ steps:
                     nvidia.com/gpu: "{{ step.num_gpus or default_num_gpu }}"
                 {% endif %}
                 env:
+                  - name: VLLM_USAGE_SOURCE
+                    value: ci-test
                   - name: HF_TOKEN
                     valueFrom:
                       secretKeyRef:

.github/ISSUE_TEMPLATE/100-documentation.yml

+name: 📚 Documentation
+description: Report an issue related to https://docs.vllm.ai/
+title: "[Doc]: "
+labels: ["documentation"]
+
+body:
+- type: textarea
+  attributes:
+    label: 📚 The doc issue
+    description: >
+      A clear and concise description of what content in https://docs.vllm.ai/ is an issue.
+  validations:
+    required: true
+- type: textarea
+  attributes:
+    label: Suggest a potential alternative/fix
+    description: >
+      Tell us how we could improve the documentation in this regard.
+- type: markdown
+  attributes:
+    value: >
+      Thanks for contributing 🎉!

.github/ISSUE_TEMPLATE/200-installation.yml

+name: 🛠️ Installation
+description: Report an issue here when you hit errors during installation.
+title: "[Installation]: "
+labels: ["installation"]
+
+body:
+- type: markdown
+  attributes:
+    value: >
+      #### Before submitting an issue, please make sure the issue hasn't been already addressed by searching through [the existing and past issues](https://github.com/vllm-project/vllm/issues?q=is%3Aissue+sort%3Acreated-desc+).
+- type: textarea
+  attributes:
+    label: Your current environment
+    description: |
+      Please run the following and paste the output below.
+      ```sh
+      wget https://raw.githubusercontent.com/vllm-project/vllm/main/collect_env.py
+      # For security purposes, please feel free to check the contents of collect_env.py before running it.
+      python collect_env.py
+      ```
+    value: |
+      ```text
+      The output of `python collect_env.py`
+      ```
+  validations:
+    required: true
+- type: textarea
+  attributes:
+    label: How you are installing vllm
+    description: |
+      Paste the full command you are trying to execute.
+    value: |
+      ```sh
+      pip install -vvv vllm
+      ```
+- type: markdown
+  attributes:
+    value: >
+      Thanks for contributing 🎉!

.github/ISSUE_TEMPLATE/300-usage.yml

+name: 💻 Usage
+description: Raise an issue here if you don't know how to use vllm.
+title: "[Usage]: "
+labels: ["usage"]
+
+body:
+- type: markdown
+  attributes:
+    value: >
+      #### Before submitting an issue, please make sure the issue hasn't been already addressed by searching through [the existing and past issues](https://github.com/vllm-project/vllm/issues?q=is%3Aissue+sort%3Acreated-desc+).
+- type: textarea
+  attributes:
+    label: Your current environment
+    description: |
+      Please run the following and paste the output below.
+      ```sh
+      wget https://raw.githubusercontent.com/vllm-project/vllm/main/collect_env.py
+      # For security purposes, please feel free to check the contents of collect_env.py before running it.
+      python collect_env.py
+      ```
+    value: |
+      ```text
+      The output of `python collect_env.py`
+      ```
+  validations:
+    required: true
+- type: textarea
+  attributes:
+    label: How would you like to use vllm
+    description: |
+      A detailed description of how you want to use vllm.
+    value: |
+      I want to run inference of a [specific model](put link here). I don't know how to integrate it with vllm.
+- type: markdown
+  attributes:
+    value: >
+      Thanks for contributing 🎉!

.github/ISSUE_TEMPLATE/400-bug report.yml

+name: 🐛 Bug report
+description: Raise an issue here if you find a bug.
+title: "[Bug]: "
+labels: ["bug"]
+
+body:
+- type: markdown
+  attributes:
+    value: >
+      #### Before submitting an issue, please make sure the issue hasn't been already addressed by searching through [the existing and past issues](https://github.com/vllm-project/vllm/issues?q=is%3Aissue+sort%3Acreated-desc+).
+- type: textarea
+  attributes:
+    label: Your current environment
+    description: |
+      Please run the following and paste the output below.
+      ```sh
+      wget https://raw.githubusercontent.com/vllm-project/vllm/main/collect_env.py
+      # For security purposes, please feel free to check the contents of collect_env.py before running it.
+      python collect_env.py
+      ```
+    value: |
+      ```text
+      The output of `python collect_env.py`
+      ```
+  validations:
+    required: true
+- type: textarea
+  attributes:
+    label: 🐛 Describe the bug
+    description: |
+      Please provide a clear and concise description of what the bug is.
+
+      If relevant, add a minimal example so that we can reproduce the error by running the code. It is very important for the snippet to be as succinct (minimal) as possible, so please take time to trim down any irrelevant code to help us debug efficiently. We are going to copy-paste your code and we expect to get the same result as you did: avoid any external data, and include the relevant imports, etc. For example:
+
+      ```python
+      from vllm import LLM, SamplingParams
+
+      prompts = [
+          "Hello, my name is",
+          "The president of the United States is",
+          "The capital of France is",
+          "The future of AI is",
+      ]
+      sampling_params = SamplingParams(temperature=0.8, top_p=0.95)
+
+      llm = LLM(model="facebook/opt-125m")
+
+      outputs = llm.generate(prompts, sampling_params)
+
+      # Print the outputs.
+      for output in outputs:
+          prompt = output.prompt
+          generated_text = output.outputs[0].text
+          print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
+      ```
+
+      If the code is too long (hopefully, it isn't), feel free to put it in a public gist and link it in the issue: https://gist.github.com.
+
+      Please also paste or describe the results you observe instead of the expected results. If you observe an error, please paste the error message including the **full** traceback of the exception. It may be relevant to wrap error messages in ```` ```triple quotes blocks``` ````.
+    placeholder: |
+      A clear and concise description of what the bug is.
+
+      ```python
+      # Sample code to reproduce the problem
+      ```
+
+      ```
+      The error message you got, with the full traceback.
+      ```
+  validations:
+    required: true
+- type: markdown
+  attributes:
+    value: >
+      ⚠️ Please separate bugs of `transformers` implementation or usage from bugs of `vllm`. If you think anything is wrong with the models' output:
+
+      - Try the counterpart of `transformers` first. If the error appears, please go to [their issues](https://github.com/huggingface/transformers/issues?q=is%3Aissue+is%3Aopen+sort%3Aupdated-desc).
+
+      - If the error only appears in vllm, please provide the detailed script of how you run `transformers` and `vllm`, also highlight the difference and what you expect.
+
+      Thanks for contributing 🎉!

.github/ISSUE_TEMPLATE/500-feature request.yml

+name: 🚀 Feature request
+description: Submit a proposal/request for a new vllm feature
+title: "[Feature]: "
+labels: ["feature request"]
+
+body:
+- type: markdown
+  attributes:
+    value: >
+      #### Before submitting an issue, please make sure the issue hasn't been already addressed by searching through [the existing and past issues](https://github.com/vllm-project/vllm/issues?q=is%3Aissue+sort%3Acreated-desc+).
+- type: textarea
+  attributes:
+    label: 🚀 The feature, motivation and pitch
+    description: >
+      A clear and concise description of the feature proposal. Please outline the motivation for the proposal. Is your feature request related to a specific problem? e.g., *"I'm working on X and would like Y to be possible"*. If this is related to another GitHub issue, please link here too.
+  validations:
+    required: true
+- type: textarea
+  attributes:
+    label: Alternatives
+    description: >
+      A description of any alternative solutions or features you've considered, if any.
+- type: textarea
+  attributes:
+    label: Additional context
+    description: >
+      Add any other context or screenshots about the feature request.
+- type: markdown
+  attributes:
+    value: >
+      Thanks for contributing 🎉!

.github/ISSUE_TEMPLATE/600-new model.yml

+name: 🤗 Support request for a new model from huggingface
+description: Submit a proposal/request for a new model from huggingface
+title: "[New Model]: "
+labels: ["new model"]
+
+body:
+- type: markdown
+  attributes:
+    value: >
+      #### Before submitting an issue, please make sure the issue hasn't been already addressed by searching through [the existing and past issues](https://github.com/vllm-project/vllm/issues?q=is%3Aissue+sort%3Acreated-desc+).
+
+      #### We also highly recommend you read https://docs.vllm.ai/en/latest/models/adding_model.html first to understand how to add a new model.
+- type: textarea
+  attributes:
+    label: The model to consider.
+    description: >
+      A huggingface url, pointing to the model, e.g. https://huggingface.co/openai-community/gpt2 .
+  validations:
+    required: true
+- type: textarea
+  attributes:
+    label: The closest model vllm already supports.
+    description: >
+      Here is the list of models already supported by vllm: https://github.com/vllm-project/vllm/tree/main/vllm/model_executor/models . Which model is the most similar to the model you want to add support for?
+- type: textarea
+  attributes:
+    label: What's your difficulty of supporting the model you want?
+    description: >
+      For example, any new operators or new architecture?
+- type: markdown
+  attributes:
+    value: >
+      Thanks for contributing 🎉!

.github/ISSUE_TEMPLATE/700-performance discussion.yml

+name: ⚡ Discussion on the performance of vllm
+description: Submit a proposal/discussion about the performance of vllm
+title: "[Performance]: "
+labels: ["performance"]
+
+body:
+- type: markdown
+  attributes:
+    value: >
+      #### Before submitting an issue, please make sure the issue hasn't been already addressed by searching through [the existing and past issues](https://github.com/vllm-project/vllm/issues?q=is%3Aissue+sort%3Acreated-desc+).
+- type: textarea
+  attributes:
+    label: Proposal to improve performance
+    description: >
+      How do you plan to improve vllm's performance?
+  validations:
+    required: false
+- type: textarea
+  attributes:
+    label: Report of performance regression
+    description: >
+      Please provide detailed description of performance comparison to confirm the regression. You may want to run the benchmark script at https://github.com/vllm-project/vllm/tree/main/benchmarks .
+  validations:
+    required: false
+- type: textarea
+  attributes:
+    label: Misc discussion on performance
+    description: >
+      Anything about the performance.
+  validations:
+    required: false
+- type: textarea
+  attributes:
+    label: Your current environment (if you think it is necessary)
+    description: |
+      Please run the following and paste the output below.
+      ```sh
+      wget https://raw.githubusercontent.com/vllm-project/vllm/main/collect_env.py
+      # For security purposes, please feel free to check the contents of collect_env.py before running it.
+      python collect_env.py
+      ```
+    value: |
+      ```text
+      The output of `python collect_env.py`
+      ```
+  validations:
+    required: false
+- type: markdown
+  attributes:
+    value: >
+      Thanks for contributing 🎉!

.github/ISSUE_TEMPLATE/800-misc discussion.yml

+name: 🎲 Misc/random discussions that do not fit into the above categories.
+description: Submit a discussion as you like. Note that developers are heavily overloaded and we mainly rely on community users to answer these issues.
+title: "[Misc]: "
+labels: ["misc"]
+
+body:
+- type: markdown
+  attributes:
+    value: >
+      #### Before submitting an issue, please make sure the issue hasn't been already addressed by searching through [the existing and past issues](https://github.com/vllm-project/vllm/issues?q=is%3Aissue+sort%3Acreated-desc+).
+- type: textarea
+  attributes:
+    label: Anything you want to discuss about vllm.
+    description: >
+      Anything you want to discuss about vllm.
+  validations:
+    required: true
+- type: markdown
+  attributes:
+    value: >
+      Thanks for contributing 🎉!

.github/ISSUE_TEMPLATE/config.yml

+blank_issues_enabled: false

.github/PULL_REQUEST_TEMPLATE.md

+FILL IN THE PR DESCRIPTION HERE
+
+FIX #xxxx (*link existing issues this PR will resolve*)
+
+**BEFORE SUBMITTING, PLEASE READ THE CHECKLIST BELOW AND FILL IN THE DESCRIPTION ABOVE**
+
+---
+
+<details>
+<!-- inside this <details> section, markdown rendering does not work, so we use raw html here. -->
+<summary><b> PR Checklist (Click to Expand) </b></summary>
+
+<p>Thank you for your contribution to vLLM! Before submitting the pull request, please ensure the PR meets the following criteria. This helps vLLM maintain the code quality and improve the efficiency of the review process.</p>
+
+<h3>PR Title and Classification</h3>
+<p>Only specific types of PRs will be reviewed. The PR title is prefixed appropriately to indicate the type of change. Please use one of the following:</p>
+<ul>
+    <li><code>[Bugfix]</code> for bug fixes.</li>
+    <li><code>[CI/Build]</code> for build or continuous integration improvements.</li>
+    <li><code>[Doc]</code> for documentation fixes and improvements.</li>
+    <li><code>[Model]</code> for adding a new model or improving an existing model. Model name should appear in the title.</li>
+    <li><code>[Frontend]</code> For changes on the vLLM frontend (e.g., OpenAI API server, <code>LLM</code> class, etc.) </li>
+    <li><code>[Kernel]</code> for changes affecting CUDA kernels or other compute kernels.</li>
+    <li><code>[Core]</code> for changes in the core vLLM logic (e.g., <code>LLMEngine</code>, <code>AsyncLLMEngine</code>, <code>Scheduler</code>, etc.)</li>
+    <li><code>[Hardware][Vendor]</code> for hardware-specific changes. Vendor name should appear in the prefix (e.g., <code>[Hardware][AMD]</code>).</li>
+    <li><code>[Misc]</code> for PRs that do not fit the above categories. Please use this sparingly.</li>
+</ul>
+<p><strong>Note:</strong> If the PR spans more than one category, please include all relevant prefixes.</p>
+
+<h3>Code Quality</h3>
+
+<p>The PR need to meet the following code quality standards:</p>
+
+<ul>
+    <li>We adhere to <a href="https://google.github.io/styleguide/pyguide.html">Google Python style guide</a> and <a href="https://google.github.io/styleguide/cppguide.html">Google C++ style guide</a>.</li>
+    <li>Pass all linter checks. Please use <a href="https://github.com/vllm-project/vllm/blob/main/format.sh"><code>format.sh</code></a> to format your code.</li>
+    <li>The code need to be well-documented to ensure future contributors can easily understand the code.</li>
+    <li>Include sufficient tests to ensure the project to stay correct and robust. This includes both unit tests and integration tests.</li>
+    <li>Please add documentation to <code>docs/source/</code> if the PR modifies the user-facing behaviors of vLLM. It helps vLLM user understand and utilize the new features or changes.</li>
+</ul>
+
+<h3>Notes for Large Changes</h3>
+<p>Please keep the changes as concise as possible. For major architectural changes (>500 LOC excluding kernel/data/config/test), we would expect a GitHub issue (RFC) discussing the technical design and justification. Otherwise, we will tag it with <code>rfc-required</code> and might not go through the PR.</p>
+
+<h3>What to Expect for the Reviews</h3>
+
+<p>The goal of the vLLM team is to be a <i>transparent reviewing machine</i>. We would like to make the review process transparent and efficient and make sure no contributor feel confused or frustrated. However, the vLLM team is small, so we need to prioritize some PRs over others. Here is what you can expect from the review process: </p>
+
+<ul>
+    <li> After the PR is submitted, the PR will be assigned to a reviewer. Every reviewer will pick up the PRs based on their expertise and availability.</li>
+    <li> After the PR is assigned, the reviewer will provide status update every 2-3 days. If the PR is not reviewed within 7 days, please feel free to ping the reviewer or the vLLM team.</li>
+    <li> After the review, the reviewer will put an <code> action-required</code> label on the PR if there are changes required. The contributor should address the comments and ping the reviewer to re-review the PR.</li>
+    <li> Please respond to all comments within a reasonable time frame. If a comment isn't clear or you disagree with a suggestion, feel free to ask for clarification or discuss the suggestion.
+ </li>
+</ul>
+
+<h3>Thank You</h3>
+
+<p> Finally, thank you for taking the time to read these guidelines and for your interest in contributing to vLLM. Your contributions make vLLM a great tool for everyone! </p>
+
+
+</details>
+
+

.github/workflows/ruff.yml

@@ -25,10 +25,13 @@ jobs:
     - name: Install dependencies
       run: |
         python -m pip install --upgrade pip
-        pip install ruff==0.1.5 codespell==2.2.6 tomli==2.0.1
+        pip install ruff==0.1.5 codespell==2.2.6 tomli==2.0.1 isort==5.13.2
     - name: Analysing the code with ruff
       run: |
-        ruff vllm tests
+        ruff .
     - name: Spelling check with codespell
       run: |
         codespell --toml pyproject.toml
+    - name: Run isort
+      run: |
+        isort . --check-only

.yapfignore

+collect_env.py

CMakeLists.txt

+cmake_minimum_required(VERSION 3.21)
+
+project(vllm_extensions LANGUAGES CXX)
+
+message(STATUS "Build type: ${CMAKE_BUILD_TYPE}")
+
+include(${CMAKE_CURRENT_LIST_DIR}/cmake/utils.cmake)
+
+#
+# Supported python versions.  These versions will be searched in order, the
+# first match will be selected.  These should be kept in sync with setup.py.
+#
+set(PYTHON_SUPPORTED_VERSIONS "3.8" "3.9" "3.10" "3.11")
+
+# Supported NVIDIA architectures.
+set(CUDA_SUPPORTED_ARCHS "7.0;7.5;8.0;8.6;8.9;9.0")
+
+# Supported AMD GPU architectures.
+set(HIP_SUPPORTED_ARCHS "gfx908;gfx90a;gfx942;gfx926;gfx928;gfx1100")
+
+#
+# Supported/expected torch versions for CUDA/ROCm.
+#
+# Currently, having an incorrect pytorch version results in a warning
+# rather than an error.
+#
+# Note: the CUDA torch version is derived from pyproject.toml and various
+# requirements.txt files and should be kept consistent.  The ROCm torch
+# versions are derived from Dockerfile.rocm
+#
+set(TORCH_SUPPORTED_VERSION_CUDA "2.1.2")
+set(TORCH_SUPPORTED_VERSION_ROCM_5X "2.0.1")
+set(TORCH_SUPPORTED_VERSION_ROCM_6X "2.1.1")
+
+#
+# Try to find python package with an executable that exactly matches
+# `VLLM_PYTHON_EXECUTABLE` and is one of the supported versions.
+#
+if (VLLM_PYTHON_EXECUTABLE)
+  find_python_from_executable(${VLLM_PYTHON_EXECUTABLE} "${PYTHON_SUPPORTED_VERSIONS}")
+else()
+  message(FATAL_ERROR
+    "Please set VLLM_PYTHON_EXECUTABLE to the path of the desired python version"
+    " before running cmake configure.")
+endif()
+
+#
+# Update cmake's `CMAKE_PREFIX_PATH` with torch location.
+#
+append_cmake_prefix_path("torch" "torch.utils.cmake_prefix_path")
+
+# Ensure the 'nvcc' command is in the PATH
+find_program(NVCC_EXECUTABLE nvcc)
+if (CUDA_FOUND AND NOT NVCC_EXECUTABLE)
+    message(FATAL_ERROR "nvcc not found")
+endif()
+
+#
+# Import torch cmake configuration.
+# Torch also imports CUDA (and partially HIP) languages with some customizations,
+# so there is no need to do this explicitly with check_language/enable_language,
+# etc.
+#
+find_package(Torch REQUIRED)
+
+#
+# Normally `torch.utils.cpp_extension.CUDAExtension` would add
+# `libtorch_python.so` for linking against an extension. Torch's cmake
+# configuration does not include this library (presumably since the cmake
+# config is used for standalone C++ binaries that link against torch).
+# The `libtorch_python.so` library defines some of the glue code between
+# torch/python via pybind and is required by VLLM extensions for this
+# reason. So, add it by manually with `find_library` using torch's
+# installed library path.
+#
+find_library(torch_python_LIBRARY torch_python PATHS
+  "${TORCH_INSTALL_PREFIX}/lib")
+
+#
+# Set up GPU language and check the torch version and warn if it isn't
+# what is expected.
+#
+if (NOT HIP_FOUND AND CUDA_FOUND)
+  set(VLLM_GPU_LANG "CUDA")
+
+  if (NOT Torch_VERSION VERSION_EQUAL ${TORCH_SUPPORTED_VERSION_CUDA})
+    message(WARNING "Pytorch version ${TORCH_SUPPORTED_VERSION_CUDA} "
+      "expected for CUDA build, saw ${Torch_VERSION} instead.")
+  endif()
+elseif(HIP_FOUND)
+  set(VLLM_GPU_LANG "HIP")
+
+  # Importing torch recognizes and sets up some HIP/ROCm configuration but does
+  # not let cmake recognize .hip files. In order to get cmake to understand the
+  # .hip extension automatically, HIP must be enabled explicitly.
+  enable_language(HIP)
+
+  # ROCm 5.x
+  if (ROCM_VERSION_DEV_MAJOR EQUAL 5 AND
+      NOT Torch_VERSION VERSION_EQUAL ${TORCH_SUPPORTED_VERSION_ROCM_5X})
+    message(WARNING "Pytorch version ${TORCH_SUPPORTED_VERSION_ROCM_5X} "
+      "expected for ROCMm 5.x build, saw ${Torch_VERSION} instead.")
+  endif()
+
+  # ROCm 6.x
+  if (ROCM_VERSION_DEV_MAJOR EQUAL 6 AND
+      NOT Torch_VERSION VERSION_EQUAL ${TORCH_SUPPORTED_VERSION_ROCM_6X})
+    message(WARNING "Pytorch version ${TORCH_SUPPORTED_VERSION_ROCM_6X} "
+      "expected for ROCMm 6.x build, saw ${Torch_VERSION} instead.")
+  endif()
+else()
+  message(FATAL_ERROR "Can't find CUDA or HIP installation.")
+endif()
+
+#
+# Override the GPU architectures detected by cmake/torch and filter them by
+# the supported versions for the current language.
+# The final set of arches is stored in `VLLM_GPU_ARCHES`.
+#
+override_gpu_arches(VLLM_GPU_ARCHES
+  ${VLLM_GPU_LANG}
+  "${${VLLM_GPU_LANG}_SUPPORTED_ARCHS}")
+
+#
+# Query torch for additional GPU compilation flags for the given
+# `VLLM_GPU_LANG`.
+# The final set of arches is stored in `VLLM_GPU_FLAGS`.
+#
+get_torch_gpu_compiler_flags(VLLM_GPU_FLAGS ${VLLM_GPU_LANG})
+
+#
+# Set nvcc parallelism.
+#
+if(NVCC_THREADS AND VLLM_GPU_LANG STREQUAL "CUDA")
+  list(APPEND VLLM_GPU_FLAGS "--threads=${NVCC_THREADS}")
+endif()
+
+#
+# Define extension targets
+#
+
+#
+# _C extension
+#
+
+set(VLLM_EXT_SRC
+  "csrc/cache_kernels.cu"
+  "csrc/attention/attention_kernels.cu"
+  "csrc/pos_encoding_kernels.cu"
+  "csrc/activation_kernels.cu"
+  "csrc/layernorm_kernels.cu"
+  "csrc/quantization/squeezellm/quant_cuda_kernel.cu"
+  "csrc/quantization/gptq/q_gemm.cu"
+  "csrc/cuda_utils_kernels.cu"
+  "csrc/moe_align_block_size_kernels.cu"
+  "csrc/pybind.cpp")
+
+if(VLLM_GPU_LANG STREQUAL "CUDA")
+  list(APPEND VLLM_EXT_SRC
+    "csrc/quantization/awq/gemm_kernels.cu"
+    "csrc/quantization/marlin/marlin_cuda_kernel.cu"
+    "csrc/custom_all_reduce.cu")
+endif()
+
+define_gpu_extension_target(
+  _C
+  DESTINATION vllm
+  LANGUAGE ${VLLM_GPU_LANG}
+  SOURCES ${VLLM_EXT_SRC}
+  COMPILE_FLAGS ${VLLM_GPU_FLAGS}
+  ARCHITECTURES ${VLLM_GPU_ARCHES}
+  WITH_SOABI)
+
+#
+# _moe_C extension
+#
+
+set(VLLM_MOE_EXT_SRC
+  "csrc/moe/moe_ops.cpp"
+  "csrc/moe/topk_softmax_kernels.cu")
+
+define_gpu_extension_target(
+  _moe_C
+  DESTINATION vllm
+  LANGUAGE ${VLLM_GPU_LANG}
+  SOURCES ${VLLM_MOE_EXT_SRC}
+  COMPILE_FLAGS ${VLLM_GPU_FLAGS}
+  ARCHITECTURES ${VLLM_GPU_ARCHES}
+  WITH_SOABI)
+
+#
+# _punica_C extension
+#
+
+set(VLLM_PUNICA_EXT_SRC
+  "csrc/punica/bgmv/bgmv_bf16_bf16_bf16.cu"
+  "csrc/punica/bgmv/bgmv_bf16_bf16_fp16.cu"
+  "csrc/punica/bgmv/bgmv_bf16_fp16_bf16.cu"
+  "csrc/punica/bgmv/bgmv_bf16_fp16_fp16.cu"
+  "csrc/punica/bgmv/bgmv_bf16_fp32_bf16.cu"
+  "csrc/punica/bgmv/bgmv_bf16_fp32_fp16.cu"
+  "csrc/punica/bgmv/bgmv_fp16_bf16_bf16.cu"
+  "csrc/punica/bgmv/bgmv_fp16_bf16_fp16.cu"
+  "csrc/punica/bgmv/bgmv_fp16_fp16_bf16.cu"
+  "csrc/punica/bgmv/bgmv_fp16_fp16_fp16.cu"
+  "csrc/punica/bgmv/bgmv_fp16_fp32_bf16.cu"
+  "csrc/punica/bgmv/bgmv_fp16_fp32_fp16.cu"
+  "csrc/punica/bgmv/bgmv_fp32_bf16_bf16.cu"
+  "csrc/punica/bgmv/bgmv_fp32_bf16_fp16.cu"
+  "csrc/punica/bgmv/bgmv_fp32_fp16_bf16.cu"
+  "csrc/punica/bgmv/bgmv_fp32_fp16_fp16.cu"
+  "csrc/punica/bgmv/bgmv_fp32_fp32_bf16.cu"
+  "csrc/punica/bgmv/bgmv_fp32_fp32_fp16.cu"
+  "csrc/punica/punica_ops.cc")
+
+#
+# Copy GPU compilation flags+update for punica
+#
+set(VLLM_PUNICA_GPU_FLAGS ${VLLM_GPU_FLAGS})
+list(REMOVE_ITEM VLLM_PUNICA_GPU_FLAGS
+  "-D__CUDA_NO_HALF_OPERATORS__"
+  "-D__CUDA_NO_HALF_CONVERSIONS__"
+  "-D__CUDA_NO_BFLOAT16_CONVERSIONS__"
+  "-D__CUDA_NO_HALF2_OPERATORS__")
+
+#
+# Filter out CUDA architectures < 8.0 for punica.
+#
+if (${VLLM_GPU_LANG} STREQUAL "CUDA")
+  set(VLLM_PUNICA_GPU_ARCHES)
+  foreach(ARCH ${VLLM_GPU_ARCHES})
+    string_to_ver(CODE_VER ${ARCH})
+    if (CODE_VER GREATER_EQUAL 8.0)
+      list(APPEND VLLM_PUNICA_GPU_ARCHES ${ARCH})
+    endif()
+  endforeach()
+  message(STATUS "Punica target arches: ${VLLM_PUNICA_GPU_ARCHES}")
+endif()
+
+if (VLLM_PUNICA_GPU_ARCHES)
+  define_gpu_extension_target(
+    _punica_C
+    DESTINATION vllm
+    LANGUAGE ${VLLM_GPU_LANG}
+    SOURCES ${VLLM_PUNICA_EXT_SRC}
+    COMPILE_FLAGS ${VLLM_PUNICA_GPU_FLAGS}
+    ARCHITECTURES ${VLLM_PUNICA_GPU_ARCHES}
+    WITH_SOABI)
+else()
+  message(WARNING "Unable to create _punica_C target because none of the "
+    "requested architectures (${VLLM_GPU_ARCHES}) are supported, i.e. >= 8.0")
+endif()
+
+#
+# Add the `default` target which detects which extensions should be
+# built based on platform/architecture.  This is the same logic that
+# setup.py uses to select which extensions should be built and should
+# be kept in sync.
+#
+# The `default` target makes direct use of cmake easier since knowledge
+# of which extensions are supported has been factored in, e.g.
+#
+# mkdir build && cd build
+# cmake -G Ninja -DVLLM_PYTHON_EXECUTABLE=`which python3` -DCMAKE_LIBRARY_OUTPUT_DIRECTORY=../vllm ..
+# cmake --build . --target default
+#
+add_custom_target(default)
+
+if(VLLM_GPU_LANG STREQUAL "CUDA" OR VLLM_GPU_LANG STREQUAL "HIP")
+  message(STATUS "Enabling C extension.")
+  add_dependencies(default _C)
+endif()
+
+if(VLLM_GPU_LANG STREQUAL "CUDA")
+  message(STATUS "Enabling moe extension.")
+  add_dependencies(default _moe_C)
+
+  # Enable punica if -DVLLM_INSTALL_PUNICA_KERNELS=ON or
+  # VLLM_INSTALL_PUNICA_KERNELS is set in the environment and
+  # there are supported target arches.
+  if (VLLM_PUNICA_GPU_ARCHES AND
+      (ENV{VLLM_INSTALL_PUNICA_KERNELS} OR VLLM_INSTALL_PUNICA_KERNELS))
+    message(STATUS "Enabling punica extension.")
+    add_dependencies(default _punica_C)
+  endif()
+endif()

CONTRIBUTING.md

@@ -45,31 +45,9 @@ pytest tests/
 If you encounter a bug or have a feature request, please check our issues page first to see if someone else has already reported it.
 If not, please file a new issue, providing as much relevant information as possible.
 
-### Coding Style Guide
+### Pull Requests & Code Reviews
 
-In general, we adhere to [Google Python style guide](https://google.github.io/styleguide/pyguide.html) and [Google C++ style guide](https://google.github.io/styleguide/cppguide.html).
-
-We include a formatting script [`format.sh`](./format.sh) to format the code.
-
-### Pull Requests
-
-When submitting a pull request:
-
-1. Make sure your code has been rebased on top of the latest commit on the main branch.
-2. Ensure code is properly formatted by running [`format.sh`](./format.sh).
-3. Include a detailed description of the changes in the pull request.
-Explain why you made the changes you did.
-If your pull request fixes an open issue, please include a reference to it in the description.
-
-### Code Reviews
-
-All submissions, including submissions by project members, require a code review.
-To make the review process as smooth as possible, please:
-
-1. Keep your changes as concise as possible.
-If your pull request involves multiple unrelated changes, consider splitting it into separate pull requests.
-2. Respond to all comments within a reasonable time frame.
-If a comment isn't clear or you disagree with a suggestion, feel free to ask for clarification or discuss the suggestion.
+Please check the PR checklist in the [PR template](.github/PULL_REQUEST_TEMPLATE.md) for detailed guide for contribution.
 
 ### Thank You
 

Dockerfile

@@ -35,9 +35,14 @@ COPY requirements-build.txt requirements-build.txt
 RUN --mount=type=cache,target=/root/.cache/pip \
     pip install -r requirements-build.txt
 
+# install compiler cache to speed up compilation leveraging local or remote caching
+RUN apt-get update -y && apt-get install -y ccache
+
 # copy input files
 COPY csrc csrc
 COPY setup.py setup.py
+COPY cmake cmake
+COPY CMakeLists.txt CMakeLists.txt
 COPY requirements.txt requirements.txt
 COPY pyproject.toml pyproject.toml
 COPY vllm/__init__.py vllm/__init__.py
@@ -54,9 +59,27 @@ ENV NVCC_THREADS=$nvcc_threads
 # make sure punica kernels are built (for LoRA)
 ENV VLLM_INSTALL_PUNICA_KERNELS=1
 
-RUN python3 setup.py build_ext --inplace
+ENV CCACHE_DIR=/root/.cache/ccache
+RUN --mount=type=cache,target=/root/.cache/ccache \
+    python3 setup.py build_ext --inplace
 #################### EXTENSION Build IMAGE ####################
 
+#################### FLASH_ATTENTION Build IMAGE ####################
+FROM dev as flash-attn-builder
+# max jobs used for build
+ARG max_jobs=2
+ENV MAX_JOBS=${max_jobs}
+# flash attention version
+ARG flash_attn_version=v2.5.6
+ENV FLASH_ATTN_VERSION=${flash_attn_version}
+
+WORKDIR /usr/src/flash-attention-v2
+
+# Download the wheel or build it if a pre-compiled release doesn't exist
+RUN pip --verbose wheel flash-attn==${FLASH_ATTN_VERSION} \
+    --no-build-isolation --no-deps --no-cache-dir
+
+#################### FLASH_ATTENTION Build IMAGE ####################
 
 #################### TEST IMAGE ####################
 # image to run unit testing suite
@@ -68,6 +91,9 @@ WORKDIR /vllm-workspace
 # ADD is used to preserve directory structure
 ADD . /vllm-workspace/
 COPY --from=build /workspace/vllm/*.so /vllm-workspace/vllm/
+# Install flash attention (from pre-built wheel)
+RUN --mount=type=bind,from=flash-attn-builder,src=/usr/src/flash-attention-v2,target=/usr/src/flash-attention-v2 \
+    pip install /usr/src/flash-attention-v2/*.whl --no-cache-dir
 # ignore build dependencies installation because we are using pre-complied extensions
 RUN rm pyproject.toml
 RUN --mount=type=cache,target=/root/.cache/pip VLLM_USE_PRECOMPILED=1 pip install . --verbose
@@ -76,7 +102,7 @@ RUN --mount=type=cache,target=/root/.cache/pip VLLM_USE_PRECOMPILED=1 pip instal
 
 #################### RUNTIME BASE IMAGE ####################
 # We used base cuda image because pytorch installs its own cuda libraries.
-# However cupy depends on cuda libraries so we had to switch to the runtime image
+# However pynccl depends on cuda libraries so we had to switch to the runtime image
 # In the future it would be nice to get a container with pytorch and cuda without duplicating cuda
 FROM nvidia/cuda:12.1.0-runtime-ubuntu22.04 AS vllm-base
 
@@ -88,6 +114,11 @@ WORKDIR /workspace
 COPY requirements.txt requirements.txt
 RUN --mount=type=cache,target=/root/.cache/pip \
     pip install -r requirements.txt
+
+# Install flash attention (from pre-built wheel)
+RUN --mount=type=bind,from=flash-attn-builder,src=/usr/src/flash-attention-v2,target=/usr/src/flash-attention-v2 \
+    pip install /usr/src/flash-attention-v2/*.whl --no-cache-dir
+
 #################### RUNTIME BASE IMAGE ####################
 
 
@@ -96,10 +127,12 @@ RUN --mount=type=cache,target=/root/.cache/pip \
 FROM vllm-base AS vllm-openai
 # install additional dependencies for openai api server
 RUN --mount=type=cache,target=/root/.cache/pip \
-    pip install accelerate
+    pip install accelerate hf_transfer modelscope
 
 COPY --from=build /workspace/vllm/*.so /workspace/vllm/
 COPY vllm vllm
 
+ENV VLLM_USAGE_SOURCE production-docker-image
+
 ENTRYPOINT ["python3", "-m", "vllm.entrypoints.openai.api_server"]
 #################### OPENAI API SERVER ####################