fix: 5h fixed timeout per file, check total feasibility first

app.py

@@ -661,16 +661,15 @@ def gradio_main():
         yield _log_text(), gr.Button(visible=False), gr.Button(visible=True), gr.File()
 
         if audio_to_caption and use_lm_caption and _server_ok():
-            # --- Mode: GGUF LM captioning (slow, best quality) ---
-            # 5h total budget — check if feasible first
+            # --- Mode: GGUF LM captioning (best quality, 5h timeout per file) ---
             LM_TIMEOUT = 18000  # 5h per file
             est_total = int(total_dur * 7 + len(audio_to_caption) * 600)
             if est_total > LM_TIMEOUT:
-                _log(f"[WARN] Estimated {est_total // 60} min for LM captioning "
-                     f"— exceeds 5h, switching to fast captioning")
+                _log(f"[WARN] Estimated {est_total // 60} min exceeds 5h, switching to fast captioning")
                 use_lm_caption = False
+                yield _log_text(), gr.Button(visible=False), gr.Button(visible=True), gr.File()
             else:
-                _log(f"[INFO] LM captioning {len(audio_to_caption)} files...")
+                _log(f"[INFO] LM captioning {len(audio_to_caption)} files (5h timeout per file)...")
                 yield _log_text(), gr.Button(visible=False), gr.Button(visible=True), gr.File()
                 for audio_fname, full_path, sidecar_json in audio_to_caption:
                     if _training_cancel.is_set():
@@ -681,15 +680,15 @@ def gradio_main():
                         full_path, timeout=LM_TIMEOUT,
                         cancel_check=lambda: _training_cancel.is_set(),
                     )
-                if caption_data:
-                    bpm_s = caption_data.get("bpm", "?")
-                    key_s = caption_data.get("keyscale", caption_data.get("key", "?"))
-                    _log(f"  {audio_fname}: OK (BPM={bpm_s}, key={key_s})")
-                    with open(sidecar_json, "w") as cj:
-                        json.dump(caption_data, cj)
-                else:
-                    _log(f"  {audio_fname}: LM failed, will use fast captioning")
-                yield _log_text(), gr.Button(visible=False), gr.Button(visible=True), gr.File()
+                    if caption_data:
+                        bpm_s = caption_data.get("bpm", "?")
+                        key_s = caption_data.get("keyscale", caption_data.get("key", "?"))
+                        _log(f"  {audio_fname}: OK (BPM={bpm_s}, key={key_s})")
+                        with open(sidecar_json, "w") as cj:
+                            json.dump(caption_data, cj)
+                    else:
+                        _log(f"  {audio_fname}: LM failed")
+                    yield _log_text(), gr.Button(visible=False), gr.Button(visible=True), gr.File()
 
         if audio_to_caption and not use_lm_caption:
             # --- Mode: Fast captioning (CLAP + Whisper + librosa) ---