Feature(MInference): fix env

app.py

@@ -2,21 +2,21 @@ import subprocess
 import os
 # Install flash attention, skipping CUDA build if necessary
 
-os.environ["CPATH"] = "$CPATH:/usr/local/cuda/include"
-os.environ["CUDA_HOME"] = "/usr/local/cuda"
-os.environ["PATH"] = "$PATH:$CUDA_HOME/bin"
-os.environ["LIBRARY_PATH"] = "$LIBRARY_PATH:/usr/local/cuda/lib64"
+# os.environ["CPATH"] = "$CPATH:/usr/local/cuda/include"
+# os.environ["CUDA_HOME"] = "/usr/local/cuda"
+# os.environ["PATH"] = "$PATH:$CUDA_HOME/bin"
+# os.environ["LIBRARY_PATH"] = "$LIBRARY_PATH:/usr/local/cuda/lib64"
 
 subprocess.run(
     "/home/user/.pyenv/shims/pip install flash-attn --no-build-isolation",
     env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"},
     shell=True,
 )
-subprocess.run(
-    "/home/user/.pyenv/shims/pip install pycuda==2023.1",
-    env={"CPATH": "$CPATH:/usr/local/cuda/include", "LIBRARY_PATH": "$LIBRARY_PATH:/usr/local/cuda/lib64"},
-    shell=True,
-)
+# subprocess.run(
+#     "/home/user/.pyenv/shims/pip install pycuda==2023.1",
+#     env={"CPATH": "$CPATH:/usr/local/cuda/include", "LIBRARY_PATH": "$LIBRARY_PATH:/usr/local/cuda/lib64"},
+#     shell=True,
+# )
 
 import gradio as gr
 import os

requirements.txt

@@ -2,4 +2,5 @@ triton==2.1.0
 accelerate
 transformers
 wheel
-setuptools
+setuptools
+pycuda==2023.1