@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / 15859f2357
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 16, 04:30 UTC

[[Misc]Upgrade bitsandbytes to the latest version 0.45.0 (#11201)

Jee Jee Li committed Dec 15, 2024 at 11:03 UTC 15859f2357059ef488405e5336d2c6e5d246687b
5 files changed +8 -8
Dockerfile
+1 -1
@@ -218,7 +218,7 @@ FROM vllm-base AS vllm-openai
218
219 # install additional dependencies for openai api server
220 RUN --mount=type=cache,target=/root/.cache/pip \
221 - pip install accelerate hf_transfer 'modelscope!=1.15.0' 'bitsandbytes>=0.44.0' timm==0.9.10
221 + pip install accelerate hf_transfer 'modelscope!=1.15.0' 'bitsandbytes>=0.45.0' timm==0.9.10
222
223 ENV VLLM_USAGE_SOURCE production-docker-image
224
docs/source/quantization/bnb.rst
+1 -1
@@ -11,7 +11,7 @@ Below are the steps to utilize BitsAndBytes with vLLM.
11
12 .. code-block:: console
13
14 - $ pip install bitsandbytes>=0.44.0
14 + $ pip install bitsandbytes>=0.45.0
15
16 vLLM reads the model's config file and supports both in-flight quantization and pre-quantized checkpoint.
17
requirements-test.in
+1 -1
@@ -25,7 +25,7 @@ datamodel_code_generator # required for minicpm3 test
25 lm-eval[api]==0.4.4 # required for model evaluation test
26
27 # quantization
28 -bitsandbytes>=0.44.0
28 +bitsandbytes>=0.45.0
29 buildkite-test-collector==0.1.9
30
31 numpy < 2.0.0
requirements-test.txt
+1 -1
@@ -37,7 +37,7 @@ audioread==3.0.1
37 # via librosa
38 awscli==1.35.23
39 # via -r requirements-test.in
40 -bitsandbytes==0.44.1
40 +bitsandbytes>=0.45.0
41 # via -r requirements-test.in
42 black==24.10.0
43 # via datamodel-code-generator
vllm/model_executor/layers/quantization/bitsandbytes.py
+4 -4
@@ -145,12 +145,12 @@ class BitsAndBytesLinearMethod(LinearMethodBase):
145 def __init__(self, quant_config: BitsAndBytesConfig):
146 try:
147 import bitsandbytes
148 - if bitsandbytes.__version__ < "0.44.0":
148 + if bitsandbytes.__version__ < "0.45.0":
149 raise ImportError("bitsandbytes version is wrong. Please "
150 - "install bitsandbytes>=0.44.0.")
150 + "install bitsandbytes>=0.45.0.")
151 except ImportError as err:
152 - raise ImportError("Please install bitsandbytes>=0.44.0 via "
153 - "`pip install bitsandbytes>=0.44.0` to use "
152 + raise ImportError("Please install bitsandbytes>=0.45.0 via "
153 + "`pip install bitsandbytes>=0.45.0` to use "
154 "bitsandbytes quantizer.") from err
155
156 self.quant_config = quant_config