Upload app.py with huggingface_hub

app.py

@@ -18,13 +18,12 @@ from threading import Thread, Lock
 from huggingface_hub import snapshot_download
 
 # 🛡️ 1. SILENCE VERBOSE LOGGING
-# Minimize "steps" in logs as requested by user
 logging.getLogger("transformers").setLevel(logging.ERROR)
 logging.getLogger("TTS").setLevel(logging.ERROR)
 os.environ["CT2_VERBOSE"] = "0"
 os.environ["KMP_WARNINGS"] = "0"
 
-# 🛠️ 2. CRITICAL COMPATIBILITY MONKEYPATCHES
+# 🛠️ 2. COMPATIBILITY PATCHES
 if "torchaudio.backend" not in sys.modules:
     backend = types.ModuleType("torchaudio.backend")
     common = types.ModuleType("torchaudio.backend.common")
@@ -78,8 +77,8 @@ except ImportError:
             if f is None: return lambda x: x
             return f
 
-# FORCE BUILD TRIGGER: 11:45:00 Jan 21 2026
-# v93: Silent Local-Only Mode. Forces instant retrieval after warmup.
+# FORCE BUILD TRIGGER: 11:55:00 Jan 21 2026
+# v94: Startup Event Warmup + Fix Port Conflict. Final Stabilization.
 
 os.environ["COQUI_TOS_AGREED"] = "1"
 MODELS = {"stt": None, "translate": None, "tts": None, "denoiser": None}
@@ -88,47 +87,34 @@ WARMUP_STATUS = {"complete": False, "in_progress": False, "error": None}
 WARMUP_LOCK = Lock()
 
 def activate_gpu_models(action):
-    """v93: Silent Instant Activation"""
+    """v94: Direct GPU Activation"""
     global MODELS, WARMUP_STATUS
-    
-    # Force local-only if warmup is done
     local_only = WARMUP_STATUS["complete"]
     
-    # 1. Faster-Whisper GPU Activation
+    # 1. Faster-Whisper
     if action in ["stt", "s2st"]:
         stt_ready = False
         try: stt_ready = MODELS["stt"] is not None and MODELS["stt"].model.device == "cuda"
         except: pass
-        
         if not stt_ready:
-            print(f"🎙️ [v93] Activating Whisper (Local Mode={local_only})...")
-            # Explicitly clear any CPU crumbs to prevent CUDA conflicts
-            if MODELS["stt"]: del MODELS["stt"]
-            gc.collect(); torch.cuda.empty_cache()
-            
-            MODELS["stt"] = WhisperModel(
-                "large-v3", 
-                device="cuda", 
-                compute_type="float16",
-                local_files_only=local_only
-            )
+            print(f"🎙️ [v94] Activating Whisper (Local={local_only})...")
+            if MODELS["stt"]: del MODELS["stt"]; gc.collect(); torch.cuda.empty_cache()
+            MODELS["stt"] = WhisperModel("large-v3", device="cuda", compute_type="float16", local_files_only=local_only)
 
-    # 2. XTTS-v2 GPU Activation
+    # 2. XTTS-v2
     if action in ["tts", "s2st"]:
-        tts_on_gpu = False
+        tts_ready = False
         try:
-            current_dev = str(next(MODELS["tts"].synthesizer.tts_model.parameters()).device)
-            tts_on_gpu = "cuda" in current_dev
+            curr = str(next(MODELS["tts"].synthesizer.tts_model.parameters()).device)
+            tts_ready = "cuda" in curr
         except: pass
-
-        if MODELS["tts"] is None or not tts_on_gpu:
-            print(f"🔊 [v93] Activating XTTS-v2 (Local Mode={local_only})...")
+        if MODELS["tts"] is None or not tts_ready:
+            print(f"🔊 [v94] Activating XTTS-v2 (Local={local_only})...")
             if MODELS["tts"] is None:
                 MODELS["tts"] = TTS(model_name="tts_models/multilingual/multi-dataset/xtts_v2", gpu=True)
-            else:
-                MODELS["tts"].to("cuda")
+            else: MODELS["tts"].to("cuda")
 
-    # 3. Helpers (Always Local)
+    # 3. Helpers
     if MODELS["denoiser"] is None:
         try: MODELS["denoiser"] = init_df()
         except: pass
@@ -136,38 +122,26 @@ def activate_gpu_models(action):
     chatterbox_utils.load_chatterbox(device="cuda" if torch.cuda.is_available() else "cpu")
 
 def warmup_task():
-    """Silent Background Warmup (v93)"""
+    """Silent Background Warmup (Threaded)"""
     global WARMUP_STATUS
     with WARMUP_LOCK:
         if WARMUP_STATUS["complete"] or WARMUP_STATUS["in_progress"]: return
         WARMUP_STATUS["in_progress"] = True
     
-    # We load them to CPU RAM first to ensure weights are in OS page cache
-    print("\n🔥 --- SILENT WARMUP STARTED (v93) ---")
+    print("\n🔥 --- SILENT WARMUP STARTED (v94) ---")
     start = time.time()
     try:
-        # 1. Faster-Whisper
-        print("📥 Pre-loading Whisper to System RAM...")
         MODELS["stt"] = WhisperModel("large-v3", device="cpu", compute_type="int8")
-        
-        # 2. XTTS-v2
-        print("📥 Pre-loading XTTS-v2 to System RAM...")
         MODELS["tts"] = TTS(model_name="tts_models/multilingual/multi-dataset/xtts_v2", gpu=False)
-        
-        # 3. Chatterbox
         chatterbox_utils.warmup_chatterbox()
-        
         WARMUP_STATUS["complete"] = True
-        print(f"✅ --- SYSTEM OPTIMIZED: INSTANT RETRIEVAL READY ({time.time()-start:.2f}s) --- \n")
+        print(f"✅ --- SYSTEM READY ({time.time()-start:.2f}s) --- \n")
     except Exception as e:
         print(f"❌ Warmup fail: {e}")
         WARMUP_STATUS["error"] = str(e)
     finally:
         WARMUP_STATUS["in_progress"] = False
 
-def start_background_warmup():
-    Thread(target=warmup_task, daemon=True).start()
-
 def _stt_logic(request_dict):
     audio_bytes = base64.b64decode(request_dict.get("file"))
     lang = request_dict.get("lang")
@@ -187,7 +161,6 @@ def _tts_logic(text, lang, speaker_wav_b64):
     XTTS_MAP = {"en": "en", "de": "de", "fr": "fr", "es": "es", "it": "it", "pl": "pl", "pt": "pt", "tr": "tr", "ru": "ru", "nl": "nl", "cs": "cs", "ar": "ar", "hu": "hu", "ko": "ko", "hi": "hi", "zh": "zh-cn"}
     clean_lang = lang.strip().lower().split('-')[0]
     mapped_lang = XTTS_MAP.get(clean_lang) or ("zh-cn" if clean_lang == "zh" else None)
-    
     if mapped_lang:
         speaker_wav_path = None
         if speaker_wav_b64:
@@ -203,7 +176,6 @@ def _tts_logic(text, lang, speaker_wav_b64):
         finally:
             if speaker_wav_path and "default_speaker" not in speaker_wav_path and os.path.exists(speaker_wav_path): os.unlink(speaker_wav_path)
             if 'output_path' in locals() and os.path.exists(output_path): os.unlink(output_path)
-            
     try:
         temp_ref = None
         if speaker_wav_b64:
@@ -218,9 +190,8 @@ def _tts_logic(text, lang, speaker_wav_b64):
 @spaces.GPU(duration=150)
 def core_process(request_dict):
     action = request_dict.get("action")
-    t0 = time.time()
-    # v93: Optimized logs (less "steps")
-    print(f"--- [v93] 🚀 GPU SESSION START: {action} ---")
+    t1 = time.time()
+    print(f"--- [v94] 🚀 GPU SESSION START: {action} ---")
     activate_gpu_models(action)
     try:
         if action == "stt": res = _stt_logic(request_dict)
@@ -234,12 +205,19 @@ def core_process(request_dict):
         elif action == "health": res = {"status": "awake"}
         else: res = {"error": f"Unknown action: {action}"}
     finally:
-        print(f"--- [v93] ✨ END: {action} ({time.time()-t0:.2f}s) ---")
+        print(f"--- [v94] ✨ END: {action} ({time.time()-t1:.2f}s) ---")
         gc.collect()
         if torch.cuda.is_available(): torch.cuda.empty_cache()
     return res
 
 app = FastAPI()
+
+@app.on_event("startup")
+async def startup_event():
+    """Ensure warmup starts regardless of entry point (v94)"""
+    print("🚀 App Startup Event: Launching Background Warmup")
+    Thread(target=warmup_task, daemon=True).start()
+
 @app.post("/api/v1/process")
 async def api_process(request: Request):
     try: return core_process(await request.json())
@@ -247,12 +225,7 @@ async def api_process(request: Request):
 
 @app.get("/health")
 def health(): 
-    return {
-        "status": "ok", 
-        "optimized": WARMUP_STATUS["complete"],
-        "gpu_available": torch.cuda.is_available(),
-        "time": time.ctime()
-    }
+    return {"status": "ok", "optimized": WARMUP_STATUS["complete"], "time": time.ctime()}
 
 @app.post("/api/v1/clear_cache")
 async def clear_cache():
@@ -273,6 +246,7 @@ def gradio_fn(req_json):
 demo = gr.Interface(fn=gradio_fn, inputs="text", outputs="text", title="🚀 AI Engine")
 app = gr.mount_gradio_app(app, demo, path="/")
 
+# Note: if __name__ == "__main__" is skipped if launched via 'uvicorn app:app'
 if __name__ == "__main__":
-    start_background_warmup()
+    print("🛠️ Manual Start detected")
     uvicorn.run(app, host="0.0.0.0", port=7860, log_level="error")