Spaces:
Running
on
Zero
Running
on
Zero
Use precompiled flash_attn wheel
Browse files- requirements.txt +3 -2
requirements.txt
CHANGED
|
@@ -26,5 +26,6 @@ onnxruntime-gpu==1.22.0
|
|
| 26 |
insightface==0.7.3
|
| 27 |
transformers==4.52.0
|
| 28 |
huggingface_hub
|
| 29 |
-
ninja
|
| 30 |
-
flash_attn
|
|
|
|
|
|
| 26 |
insightface==0.7.3
|
| 27 |
transformers==4.52.0
|
| 28 |
huggingface_hub
|
| 29 |
+
ninja
|
| 30 |
+
# 使用预编译的 flash_attn wheel (torch 2.6.0 + CUDA 12.4 + Python 3.10)
|
| 31 |
+
https://github.com/Dao-AILab/flash-attention/releases/download/v2.8.1/flash_attn-2.8.1+cu124torch2.6cxx11abiFALSE-cp310-cp310-linux_x86_64.whl
|