Add vllm

.azure-pipelines/azure-pipelines-linux.yml

@@ -77,6 +77,13 @@ jobs:
       done
     displayName: Manage disk space
 
+  - script: |
+      sudo fallocate -l 10GiB /swapfile || true
+      sudo chmod 600 /swapfile || true
+      sudo mkswap /swapfile || true
+      sudo swapon /swapfile || true
+    displayName: Create swap file
+
   - script: |
       # sudo pip install --upgrade pip
       sudo pip install setuptools shyaml
@@ -121,6 +128,13 @@ jobs:
       done
     displayName: Manage disk space
 
+  - script: |
+      sudo fallocate -l 10GiB /swapfile || true
+      sudo chmod 600 /swapfile || true
+      sudo mkswap /swapfile || true
+      sudo swapon /swapfile || true
+    displayName: Create swap file
+
   - script: |
       # sudo pip install --upgrade pip
       sudo pip install setuptools shyaml

conda-forge.yml

@@ -3,3 +3,6 @@ azure:
   settings_win:
     pool:
       vmImage: windows-2022
+  free_disk_space: true
+  settings_linux:
+    swapfile_size: 10GiB

recipes/vllm/recipe.yaml

@@ -0,0 +1,108 @@
+context:
+  version: 0.7.0
+
+package:
+  name: vllm
+  version: ${{ version }}
+
+source:
+- url: https://pypi.org/packages/source/v/vllm/vllm-${{ version }}.tar.gz
+  sha256: 91521ca3e7629e2a40a4fceea3290cbde34c3ea6cb291ee9f84ad2ce752e65be
+
+build:
+  number: 0
+  script: |
+    python use_existing_torch.py
+    VERBOSE=1 ${{ PYTHON }} -m pip install . --no-build-isolation
+  python:
+    entry_points:
+    - vllm  =  vllm.scripts:main
+  skip: cuda_compiler_version == "None"
+
+requirements:
+  host:
+  - python
+  - pip
+  - setuptools
+  - setuptools-scm
+  - ninja
+  - packaging
+  - wheel
+  - jinja2
+  - pytorch * cuda*
+  - cmake
+  - git
+  - ${{ stdlib('c') }}
+  - ${{ compiler('c') }}
+  - ${{ compiler('cxx') }}
+  - ${{ compiler('cuda') }}
+  run:
+  - python
+  - psutil
+  - sentencepiece
+  - numpy <2.0.0
+  - requests >=2.26.0
+  - tqdm
+  - blake3
+  - py-cpuinfo
+  - transformers >=4.45.2
+  - tokenizers >=0.19.1
+  - protobuf
+  - fastapi !=0.113.*,!=0.114.0,>=0.107.0
+  - aiohttp
+  - openai >=1.52.0
+  - uvicorn
+  - pydantic >=2.9
+  - prometheus_client >=0.18.0
+  - pillow
+  - prometheus-fastapi-instrumentator >=7.0.0
+  - tiktoken >=0.6.0
+  - lm-format-enforcer <0.11,>=0.10.9
+  - outlines ==0.1.11
+  - lark ==1.2.2
+  # - xgrammar >=0.1.6 # platform_machine == "x86_64"
+  - typing_extensions >=4.10
+  - filelock >=3.16.1
+  - partial-json-parser
+  - pyzmq
+  - msgspec
+  - gguf ==0.10.0
+  - importlib-metadata
+  - mistral_common >=1.5.0
+  - pyyaml
+  # - six >=1.16.0 # python_version > "3.11"
+  # - setuptools >=74.1.1 # python_version > "3.11"
+  - einops
+  - compressed-tensors ==0.8.1
+  - depyf ==0.18.0
+  - cloudpickle
+  - ray-default >=2.9
+  - nvidia-ml-py >=12.560.30
+  - torch ==2.5.1
+  - torchvision ==0.20.1
+  # - xformers ==0.0.28.post3 # platform_system == "Linux" and platform_machine == "x86_64"
+  # - tensorizer >=2.9.0 # extra == "tensorizer"
+  # - runai-model-streamer  # extra == "runai"
+  # - runai-model-streamer-s3  # extra == "runai"
+  # - boto3  # extra == "runai"
+  # - librosa  # extra == "audio"
+  # - soundfile  # extra == "audio"
+  # - decord  # extra == "video"
+
+tests:
+- python:
+    imports:
+    - vllm
+    pip_check: true
+
+about:
+  homepage: https://github.com/vllm-project/vllm
+  summary: A high-throughput and memory-efficient inference and serving engine for LLMs
+  description:  Easy, fast, and cheap LLM serving for everyone
+  license: Apache-2.0
+  license_file: LICENSE
+  documentation: https://vllm.readthedocs.io/en/latest/
+
+extra:
+  recipe-maintainers:
+    - maresb