Commit e58aa1c6 authored by Conroy Cheers's avatar Conroy Cheers
Browse files

python312Packages.vllm: 0.9.0.1 -> 0.9.1

parent 5b0471b7

pkgs/development/python-modules/vllm/0005-drop-intel-reqs.patch

+17 −4
Original line number Diff line number Diff line 
From 7511784ceb9252091a9d63ac6b54dcc67dd2b262 Mon Sep 17 00:00:00 2001

From: Conroy Cheers <conroy@corncheese.org>

Date: Fri, 13 Jun 2025 17:42:10 +1000

Subject: [PATCH] drop intel reqs



---

 requirements/cpu.txt | 3 ---

 1 file changed, 3 deletions(-)



diff --git a/requirements/cpu.txt b/requirements/cpu.txt

index 121330158..d41918883 100644

index d7b0fc6d8..be2df751b 100644

--- a/requirements/cpu.txt

+++ b/requirements/cpu.txt

@@ -20,7 +20,3 @@ datasets # for benchmark scripts

 

@@ -24,8 +24,5 @@ datasets # for benchmark scripts

 # cpu cannot use triton 3.3.0

 triton==3.2.0; platform_machine == "x86_64"

-

 

-# Intel Extension for PyTorch, only for x86_64 CPUs

-intel-openmp==2024.2.1; platform_machine == "x86_64"

-intel_extension_for_pytorch==2.7.0; platform_machine == "x86_64"

 py-libnuma; platform_system != "Darwin"

 psutil; platform_system != "Darwin"

-- 

2.49.0

pkgs/development/python-modules/vllm/default.nix

+36 −27
Original line number Diff line number Diff line
@@ -3,6 +3,7 @@ 
  stdenv,

  python,

  buildPythonPackage,

  pythonAtLeast,

  fetchFromGitHub,

  fetchpatch,

  symlinkJoin,
@@ -67,6 +68,7 @@ 
  opentelemetry-exporter-otlp,

  bitsandbytes,

  flashinfer,

  py-libnuma,



  # internal dependency - for overriding in overlays

  vllm-flash-attn ? null,
@@ -246,16 +248,19 @@ in 


buildPythonPackage rec {

  pname = "vllm";

  version = "0.9.0.1";

  version = "0.9.1";

  pyproject = true;



  # https://github.com/vllm-project/vllm/issues/12083

  disabled = pythonAtLeast "3.13";



  stdenv = torch.stdenv;



  src = fetchFromGitHub {

    owner = "vllm-project";

    repo = "vllm";

    tag = "v${version}";

    hash = "sha256-gNe/kdsDQno8Fd6mo29feWmbyC0c2+kljlVxY4v7R9U=";

    hash = "sha256-sp7rDpewTPXTVRBJHJMj+8pJDS6wAu0/OTJZwbPPqKc=";

  };



  patches = [
@@ -264,14 +269,18 @@ buildPythonPackage rec { 
      url = "https://github.com/vllm-project/vllm/commit/6a5d7e45f52c3a13de43b8b4fa9033e3b342ebd2.patch";

      hash = "sha256-KYthqu+6XwsYYd80PtfrMMjuRV9+ionccr7EbjE4jJE=";

    })

    (fetchpatch {

      name = "fall-back-to-gloo-when-nccl-unavailable.patch";

      url = "https://github.com/vllm-project/vllm/commit/aa131a94410683b0a02e74fed2ce95e6c2b6b030.patch";

      hash = "sha256-jNlQZQ8xiW85JWyBjsPZ6FoRQsiG1J8bwzmQjnaWFBg=";

    })

    ./0002-setup.py-nix-support-respect-cmakeFlags.patch

    ./0003-propagate-pythonpath.patch

    ./0004-drop-lsmod.patch

    ./0005-drop-intel-reqs.patch

  ];



  postPatch =

    ''

  postPatch = ''

    # pythonRelaxDeps does not cover build-system

    substituteInPlace pyproject.toml \

      --replace-fail "torch ==" "torch >="
@@ -286,11 +295,6 @@ buildPythonPackage rec { 
    # Pass build environment PYTHONPATH to vLLM's Python configuration scripts

    substituteInPlace CMakeLists.txt \

      --replace-fail '$PYTHONPATH' '$ENV{PYTHONPATH}'

    ''

    + lib.optionalString (nccl == null) ''

      # On platforms where NCCL is not supported (e.g. Jetson), substitute Gloo (provided by Torch)

      substituteInPlace vllm/distributed/parallel_state.py \

        --replace-fail '"nccl"' '"gloo"'

  '';



  nativeBuildInputs =
@@ -362,7 +366,6 @@ buildPythonPackage rec { 
      outlines

      pandas

      prometheus-fastapi-instrumentator

      psutil

      py-cpuinfo

      pyarrow

      pydantic
@@ -392,9 +395,15 @@ buildPythonPackage rec { 
      opentelemetry-api

      opentelemetry-exporter-otlp

      bitsandbytes

      # vLLM needs Torch's compiler to be present in order to use torch.compile

      torch.stdenv.cc

    ]

    ++ uvicorn.optional-dependencies.standard

    ++ aioprometheus.optional-dependencies.starlette

    ++ lib.optionals stdenv.targetPlatform.isLinux [

      py-libnuma

      psutil

    ]

    ++ lib.optionals cudaSupport [

      cupy

      pynvml
@@ -404,11 +413,11 @@ buildPythonPackage rec { 
  dontUseCmakeConfigure = true;

  cmakeFlags =

    [

    ]

    ++ lib.optionals cudaSupport [

      (lib.cmakeFeature "FETCHCONTENT_SOURCE_DIR_CUTLASS" "${lib.getDev cutlass}")

      (lib.cmakeFeature "FLASH_MLA_SRC_DIR" "${lib.getDev flashmla}")

      (lib.cmakeFeature "VLLM_FLASH_ATTN_SRC_DIR" "${lib.getDev vllm-flash-attn'}")

    ]

    ++ lib.optionals cudaSupport [

      (lib.cmakeFeature "TORCH_CUDA_ARCH_LIST" "${gpuTargetString}")

      (lib.cmakeFeature "CUTLASS_NVCC_ARCHS_ENABLED" "${cudaPackages.flags.cmakeCudaArchitecturesString}")

      (lib.cmakeFeature "CUDA_TOOLKIT_ROOT_DIR" "${symlinkJoin {