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 14, 22:35 UTC
Move flashinfer-python to optional extra `vllm[flashinfer]` (#21959)
Signed-off-by: mgoin <mgoin64@gmail.com>
Michael Goin committed
Jul 31, 2025 at 21:02 UTC
0bd409cf01c37bbc99a5d3c70c4954da2113aba8
2 files changed
+4
-4
requirements/cuda.txt
+1
-3
@@ -11,6 +11,4 @@ torchaudio==2.7.1
11
# These must be updated alongside torch
12
torchvision==0.22.1 # Required for phi3v processor. See https://github.com/pytorch/vision?tab=readme-ov-file#installation for corresponding version
13
# https://github.com/facebookresearch/xformers/releases/tag/v0.0.31
14
-xformers==0.0.31; platform_system == 'Linux' and platform_machine == 'x86_64' # Requires PyTorch >= 2.7
15
-# FlashInfer should be updated together with the Dockerfile
16
-flashinfer_python==0.2.9rc2
\ No newline at end of file
14
+xformers==0.0.31; platform_system == 'Linux' and platform_machine == 'x86_64' # Requires PyTorch >= 2.7
\ No newline at end of file
setup.py
+3
-1
@@ -671,7 +671,9 @@ setup(
671
["runai-model-streamer >= 0.13.3", "runai-model-streamer-s3", "boto3"],
672
"audio": ["librosa", "soundfile",
673
"mistral_common[audio]"], # Required for audio processing
674
- "video": [] # Kept for backwards compatibility
674
+ "video": [], # Kept for backwards compatibility
675
+ # FlashInfer should be updated together with the Dockerfile
676
+ "flashinfer": ["flashinfer-python==0.2.9rc2"],
677
},
678
cmdclass=cmdclass,
679
package_data=package_data,