python312Packages.vllm: 0.9.0.1 -> 0.9.1 (#416385)

{

  lib,

  stdenv,

  buildPythonPackage,

  fetchFromGitHub,

  pythonOlder,

  setuptools,

  numactl,

}:

buildPythonPackage rec {

  pname = "py-libnuma";

  version = "1.2";

  pyproject = true;

  src = fetchFromGitHub {

    owner = "eedalong";

    repo = "pynuma";

    rev = "66cab0e008b850a04cfec5c4fb3f50bf28e3d488";

    hash = "sha256-ALYCcdN5eXrVWsTRwkHCwo4xsLMs/du3mUl1xSlo5iU=";

  };

  postPatch = ''

    substituteInPlace numa/__init__.py \

      --replace-fail \

        'LIBNUMA = CDLL(find_library("numa"))' \

        'LIBNUMA = CDLL("${numactl}/lib/libnuma${stdenv.hostPlatform.extensions.sharedLibrary}")'

  '';

  build-system = [ setuptools ];

  dependencies = [

    numactl

  ];

  # Tests write NUMA configuration, which may be persistent until reboot.

  doCheck = false;

  pythonImportsCheck = [ "numa" ];

  meta = {

    description = "Python3 Interface to numa Linux library";

    homepage = "https://github.com/eedalong/pynuma";

    platforms = lib.platforms.linux;

    license = lib.licenses.mit;

  };

}

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

  stdenv,

  python,

  buildPythonPackage,

  pythonAtLeast,

  fetchFromGitHub,

  fetchpatch,

  symlinkJoin,

  opentelemetry-exporter-otlp,

  bitsandbytes,

  flashinfer,

  py-libnuma,

  # internal dependency - for overriding in overlays

  vllm-flash-attn ? null,

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 = [

      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 >="

    # 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 =

      outlines

      pandas

      prometheus-fastapi-instrumentator

      psutil

      py-cpuinfo

      pyarrow

      pydantic

      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

  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 {

From c01e056ee845ae973ec36cc50125492ef8c02c12 Mon Sep 17 00:00:00 2001

From: Conroy Cheers <conroy@corncheese.org>

Date: Thu, 12 Jun 2025 17:45:27 +1000

Subject: [PATCH] [Fix] find nanobind from Python module

---

 cpp/nanobind/CMakeLists.txt | 4 ++++

 pyproject.toml              | 2 +-

 2 files changed, 5 insertions(+), 1 deletion(-)

diff --git a/cpp/nanobind/CMakeLists.txt b/cpp/nanobind/CMakeLists.txt

index 8ea5622..02500ac 100644

--- a/cpp/nanobind/CMakeLists.txt

+++ b/cpp/nanobind/CMakeLists.txt

@@ -3,6 +3,10 @@ find_package(

   COMPONENTS Interpreter Development.Module

   REQUIRED

 )

+

+execute_process(

+  COMMAND "${Python_EXECUTABLE}" -m nanobind --cmake_dir

+  OUTPUT_STRIP_TRAILING_WHITESPACE OUTPUT_VARIABLE nanobind_DIR)

 find_package(nanobind CONFIG REQUIRED)

 # Compile this source file seperately. Nanobind suggests to optimize bindings code for size, but

diff --git a/pyproject.toml b/pyproject.toml

index 11fae7d..d2078b1 100644

--- a/pyproject.toml

+++ b/pyproject.toml

@@ -44,7 +44,7 @@ provider = "scikit_build_core.metadata.regex"

 input = "python/xgrammar/version.py"

 [build-system]

-requires = ["scikit-build-core>=0.10.0", "nanobind==2.5.0"]

+requires = ["scikit-build-core>=0.10.0", "nanobind>=2.5.0"]

 build-backend = "scikit_build_core.build"

 [tool.scikit-build]

-- 

2.49.0

  # build-system

  cmake,

  ninja,

  pybind11,

  nanobind,

  scikit-build-core,

  # dependencies

buildPythonPackage rec {

  pname = "xgrammar";

  version = "0.1.14";

  version = "0.1.19";

  pyproject = true;

  src = fetchFromGitHub {

    repo = "xgrammar";

    tag = "v${version}";

    fetchSubmodules = true;

    hash = "sha256-ohsoc3g5XUp9vSXxyOGj20wXzCXZC02ktHYVQjDqNeM=";

    hash = "sha256-0b2tJx1D/2X/uosbthHfevUpTCBtuSKNlxOKyidTotA=";

  };

  patches = [

    ./0001-fix-find-nanobind-from-python-module.patch

  ];

  build-system = [

    cmake

    ninja

    pybind11

    nanobind

    scikit-build-core

  ];

  dontUseCmakeConfigure = true;

    writableTmpDirAsHomeHook

  ];

  NIX_CFLAGS_COMPILE = toString [

    # xgrammar hardcodes -flto=auto while using static linking, which can cause linker errors without this additional flag.

    "-ffat-lto-objects"

  ];

  disabledTests = [

    # You are trying to access a gated repo.

    "test_grammar_compiler"