req ref

app.py

@@ -2,20 +2,6 @@ import logging
 # Отключаем спам от httpx
 logging.getLogger("httpx").setLevel(logging.WARNING)
 
-# First, install flash-attention
-import subprocess
-import os
-subprocess.run("pip install -q moviepy", shell=True)
-subprocess.run("pip install -q imageio-ffmpeg", shell=True)
-# Install flash-attention with specific environment variable
-subprocess.run(
-    "pip install flash-attn --no-build-isolation",
-    env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"},
-    shell=True,
-)
-
-from moviepy.editor import VideoFileClip
-import tempfile
 import gradio as gr
 from transformers import pipeline
 import torch
@@ -26,6 +12,19 @@ from speechbrain.inference import EncoderClassifier
 import torchaudio
 import numpy as np
 from sklearn.cluster import AgglomerativeClustering
+import tempfile
+import os
+import subprocess
+from moviepy.editor import VideoFileClip
+
+# Initialize Whisper with flash attention
+pipe = pipeline(
+    "automatic-speech-recognition",
+    model="openai/whisper-large-v3-turbo",
+    torch_dtype=torch.float16,
+    device="cuda:0",
+    model_kwargs={"attn_implementation": "flash_attention_2"},
+)
 
 # Initialize Whisper with flash attention
 pipe = pipeline(

requirements.txt

@@ -1,4 +1,12 @@
 transformers[torch]
 speechbrain
 scikit-learn
-moviepy
+moviepy
+imageio-ffmpeg
+gradio
+numpy
+torchaudio
+flash-attn
+ffmpeg-python
+--extra-index-url https://download.pytorch.org/whl/cu118
+torch