Update app.py

app.py

@@ -3,6 +3,13 @@ import numpy as np
 import torch
 import typing
 import torchaudio
+import streamlit as st
+import subprocess
+import json
+import requests
+import gc
+import pandas as pd
+from datetime import timedelta
 
 # --- CRITICAL ENVIRONMENT FIXES ---
 # 1. Fix for Hugging Face millicore OMP_NUM_THREADS error
@@ -16,21 +23,23 @@ try:
 except Exception:
     pass
 
-# 3. GLOBAL PYTORCH 2.6+ SECURITY BYPASS (MONKEYPATCH)
-_original_torch_load = torch.load
-def patched_torch_load(*args, **kwargs):
-    # Debug print to see what's being loaded
-    # print(f"DEBUG: Intercepted torch.load call. Target: {args[0] if args else 'Unknown'}")
-    
-    # FORCE Disable security check
-    kwargs['weights_only'] = False 
-    return _original_torch_load(*args, **kwargs)
+# 3. GLOBAL PYTORCH SECURITY BYPASS (One-Time Patch)
+if not hasattr(torch.load, "_is_patched"):
+    print("DEBUG: Applying Monkeypatch to torch.load")
+    _original_torch_load = torch.load
 
-torch.load = patched_torch_load
-print("DEBUG: torch.load has been monkeypatched to allow all globals.")
+    def patched_torch_load(*args, **kwargs):
+        # FORCE Disable security check
+        kwargs['weights_only'] = False 
+        return _original_torch_load(*args, **kwargs)
+    
+    # Mark as patched to prevent recursion loop on Streamlit reruns
+    patched_torch_load._is_patched = True
+    torch.load = patched_torch_load
+else:
+    print("DEBUG: torch.load is already patched. Skipping.")
 
 # 4. EXPLICIT SAFE GLOBALS WHITELIST
-# Even with the monkeypatch, we add these to be double-safe against internal calls
 try:
     safe_list = [
         typing.Any,
@@ -44,16 +53,16 @@ try:
     elif hasattr(np, 'core') and hasattr(np.core, 'multiarray'):
         safe_list.append(np.core.multiarray.scalar)
 
-    # OmegaConf (Crucial for Pyannote/WhisperX config loading)
+    # OmegaConf
     try:
         from omegaconf.listconfig import ListConfig
         from omegaconf.dictconfig import DictConfig
         from omegaconf.base import ContainerMetadata, Metadata, Node
         safe_list.extend([ListConfig, DictConfig, ContainerMetadata, Metadata, Node])
     except ImportError:
-        print("Warning: Could not import omegaconf for whitelisting.")
+        pass
 
-    # Pyannote internals (if available)
+    # Pyannote internals
     try:
         from pyannote.audio.core.task import Specifications, Problem, Resolution
         from pyannote.audio.core.model import Model
@@ -70,17 +79,11 @@ except Exception as e:
 if not hasattr(np, 'NaN'):
     np.NaN = np.nan
 
-import streamlit as st
-import subprocess
-import json
-import requests
+# Import whisperx AFTER patching
 import whisperx
-import gc
-import pandas as pd
-from datetime import timedelta
 
 # --- Configuration & Tokens ---
-HARDCODED_HF_TOKEN = "PASTE_YOUR_HF_TOKEN_HERE"
+HARDCODED_HF_TOKEN = "PASTE_YOUR_HF_TOKEN_HERE" 
 HARDCODED_GEMINI_KEY = ""
 
 ENV_HF_TOKEN = os.environ.get("HF_TOKEN", "")
@@ -117,7 +120,6 @@ def generate_cmx_edl(edl_title, segments, source_name, fps=25):
     return "\n".join(edl_lines)
 
 def call_gemini_for_edl(transcript_data, story_prompt, api_key):
-    """Sends diarized, word-level transcript to Gemini Senior Editor."""
     if not api_key:
         st.error("Gemini API Key is missing.")
         return None
@@ -163,7 +165,6 @@ with st.sidebar:
     st.header("Model Settings")
     model_size = st.selectbox("Whisper Model", ["large-v2", "medium", "base"], index=0)
     
-    # --- Language Option ---
     language_map = {
         "Auto-Detect": None,
         "English": "en",
@@ -179,7 +180,10 @@ with st.sidebar:
     num_speakers = st.number_input("Speakers (0=Auto)", min_value=0, value=0)
     
     st.divider()
-    st.info("API Keys are managed via Environment Secrets.")
+    if ACTIVE_HF_TOKEN == "PASTE_YOUR_HF_TOKEN_HERE":
+        st.warning("⚠️ HF_TOKEN not set in Secrets!")
+    else:
+        st.success("✅ HF_TOKEN Loaded")
 
 uploaded_file = st.file_uploader("Upload Video/Audio Clip", type=["mp4", "m4a", "wav", "mp3", "mov"])
 
@@ -234,17 +238,7 @@ if uploaded_file:
                         
                         # 3. Diarize
                         st.write("🗣️ **Diarizing Speakers...**")
-                        # Pass token for gated diarization models
-                        # Try to bypass the torch security default if necessary
-                        try:
-                            diarize_model = whisperx.DiarizationPipeline(use_auth_token=ACTIVE_HF_TOKEN, device=device)
-                        except Exception as e:
-                            # Catch generic loading errors and try to print detail or retry
-                            if "Weights only load failed" in str(e) or "Unsupported global" in str(e):
-                                st.warning("⚠️ Security restriction encountered. Re-attempting load with implicit overrides.")
-                                diarize_model = whisperx.DiarizationPipeline(use_auth_token=ACTIVE_HF_TOKEN, device=device)
-                            else:
-                                raise e
+                        diarize_model = whisperx.DiarizationPipeline(use_auth_token=ACTIVE_HF_TOKEN, device=device)
                         
                         diarize_kwargs = {}
                         if num_speakers > 0:
@@ -270,6 +264,9 @@ if uploaded_file:
 
                     except Exception as e:
                         st.error(f"Processing Error: {e}")
+                        # Clean up temp files
+                        if os.path.exists("temp_input"): os.remove("temp_input")
+                        if os.path.exists("temp_audio.wav"): os.remove("temp_audio.wav")
                         st.stop()
 
     if "transcript" in st.session_state: