@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / d627a3d837
Fork 0

A high-throughput and memory-efficient inference and serving engine for LLMs

Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git

Synced Nov 17, 01:45 UTC

[Misc] Upgrade to `torch==2.3.0` (#4454)

Michael Goin committed Apr 29, 2024 at 20:05 UTC d627a3d837976a23f89ba808f5ef6908fdb65bfa
7 files changed +8 -8
.github/workflows/publish.yml
+1 -1
@@ -49,7 +49,7 @@ jobs:
49 matrix:
50 os: ['ubuntu-20.04']
51 python-version: ['3.8', '3.9', '3.10', '3.11']
52 - pytorch-version: ['2.2.1'] # Must be the most recent version that meets requirements-cuda.txt.
52 + pytorch-version: ['2.3.0'] # Must be the most recent version that meets requirements-cuda.txt.
53 cuda-version: ['11.8', '12.1']
54
55 steps:
CMakeLists.txt
+1 -1
@@ -31,7 +31,7 @@ set(HIP_SUPPORTED_ARCHS "gfx906;gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx11
31 # requirements.txt files and should be kept consistent. The ROCm torch
32 # versions are derived from Dockerfile.rocm
33 #
34 -set(TORCH_SUPPORTED_VERSION_CUDA "2.2.1")
34 +set(TORCH_SUPPORTED_VERSION_CUDA "2.3.0")
35 set(TORCH_SUPPORTED_VERSION_ROCM_5X "2.0.1")
36 set(TORCH_SUPPORTED_VERSION_ROCM_6X "2.1.1")
37
Dockerfile
+1 -1
@@ -85,7 +85,7 @@ FROM dev as flash-attn-builder
85 ARG max_jobs=2
86 ENV MAX_JOBS=${max_jobs}
87 # flash attention version
88 -ARG flash_attn_version=v2.5.6
88 +ARG flash_attn_version=v2.5.8
89 ENV FLASH_ATTN_VERSION=${flash_attn_version}
90
91 WORKDIR /usr/src/flash-attention-v2
pyproject.toml
+1 -1
@@ -5,7 +5,7 @@ requires = [
5 "ninja",
6 "packaging",
7 "setuptools >= 49.4.0",
8 - "torch == 2.2.1",
8 + "torch == 2.3.0",
9 "wheel",
10 ]
11 build-backend = "setuptools.build_meta"
requirements-build.txt
+1 -1
@@ -3,5 +3,5 @@ cmake>=3.21
3 ninja
4 packaging
5 setuptools>=49.4.0
6 -torch==2.2.1
6 +torch==2.3.0
7 wheel
requirements-cpu.txt
+1 -1
@@ -2,5 +2,5 @@
2 -r requirements-common.txt
3
4 # Dependencies for x86_64 CPUs
5 -torch == 2.2.1+cpu
5 +torch == 2.3.0+cpu
6 triton >= 2.2.0 # FIXME(woosuk): This is a hack to avoid import error.
\ No newline at end of file
requirements-cuda.txt
+2 -2
@@ -5,5 +5,5 @@
5 ray >= 2.9
6 nvidia-ml-py # for pynvml package
7 vllm-nccl-cu12>=2.18,<2.19 # for downloading nccl library
8 -torch == 2.2.1
9 -xformers == 0.0.25 # Requires PyTorch 2.2.1
8 +torch == 2.3.0
9 +xformers == 0.0.26.post1 # Requires PyTorch 2.3.0