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, 05:30 UTC
Pin PyTorch & xformers versions (#2155)
Woosuk Kwon committed
Dec 17, 2023 at 01:46 UTC
b0a1d667b04dcce8e54201cb429a57155bf9670c
4 files changed
+5
-5
.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: ['2.1.1']
52
+ pytorch-version: ['2.1.2'] # Must be the most recent version that meets requirements.txt.
53
cuda-version: ['11.8', '12.1']
54
55
steps:
pyproject.toml
+1
-1
@@ -4,7 +4,7 @@ requires = [
4
"ninja",
5
"packaging",
6
"setuptools >= 49.4.0",
7
- "torch >= 2.1.1",
7
+ "torch == 2.1.2",
8
"wheel",
9
]
10
build-backend = "setuptools.build_meta"
requirements-build.txt
+1
-1
@@ -2,5 +2,5 @@
2
ninja
3
packaging
4
setuptools>=49.4.0
5
-torch>=2.1.0
5
+torch==2.1.2
6
wheel
\ No newline at end of file
requirements.txt
+2
-2
@@ -5,9 +5,9 @@ pandas # Required for Ray data.
5
pyarrow # Required for Ray data.
6
sentencepiece # Required for LLaMA tokenizer.
7
numpy
8
-torch >= 2.1.1
8
+torch == 2.1.2
9
transformers >= 4.36.0 # Required for Mixtral.
10
-xformers >= 0.0.23 # Required for CUDA 12.1.
10
+xformers == 0.0.23 # Required for CUDA 12.1.
11
fastapi
12
uvicorn[standard]
13
pydantic == 1.10.13 # Required for OpenAI server.