Upload app.py with huggingface_hub

app.py

@@ -1,5 +1,6 @@
 from fastapi import FastAPI, Request, Response
 from fastapi.responses import StreamingResponse
+from contextlib import asynccontextmanager
 import gradio as gr
 import uvicorn
 import base64
@@ -17,10 +18,12 @@ import logging
 from threading import Thread, Lock
 from huggingface_hub import snapshot_download
 
-# 🛡️ 1. SILENCE LOGS
+# 🛡️ 1. SILENCE LOGS & WARNINGS
 logging.getLogger("transformers").setLevel(logging.ERROR)
 logging.getLogger("TTS").setLevel(logging.ERROR)
+logging.getLogger("onnxruntime").setLevel(logging.ERROR)
 os.environ["CT2_VERBOSE"] = "0"
+os.environ["ORT_LOGGING_LEVEL"] = "3" # Silence ONNX discovery noise
 
 # 🛠️ 2. COMPATIBILITY PATCHES
 if "torchaudio.backend" not in sys.modules:
@@ -76,8 +79,8 @@ except ImportError:
             if f is None: return lambda x: x
             return f
 
-# FORCE BUILD TRIGGER: 12:40:00 Jan 21 2026
-# v98: Peak Performance Re-enabled (4-Workers).
+# FORCE BUILD TRIGGER: 12:45:00 Jan 21 2026
+# v99: Fix Queue Loop. Health on CPU. Re-enable 4 Workers. Lifespan events.
 
 os.environ["COQUI_TOS_AGREED"] = "1"
 MODELS = {"stt": None, "translate": None, "tts": None, "denoiser": None}
@@ -86,17 +89,17 @@ WARMUP_STATUS = {"complete": False, "in_progress": False}
 WARMUP_LOCK = Lock()
 
 def activate_gpu_models(action):
-    """v98: High-Parallelism Peak Mode"""
+    """v99: Optimized GPU Session Activation"""
     global MODELS, WARMUP_STATUS
     local_only = WARMUP_STATUS["complete"]
     
-    # 1. Faster-Whisper: Peak Performance with 4 workers
+    # 1. Faster-Whisper: Peak Performance Requested
     if action in ["stt", "s2st"]:
         stt_on_gpu = False
         try: stt_on_gpu = MODELS["stt"] is not None and MODELS["stt"].model.device == "cuda"
         except: pass
         if not stt_on_gpu:
-            print(f"🎙️ [v98] Peak Performance Activation: Whisper (GPU) with 4-Workers...")
+            print(f"🎙️ [v99] Activating Whisper (GPU: 4-Workers)...")
             try:
                 if MODELS["stt"]: del MODELS["stt"]
                 gc.collect(); torch.cuda.empty_cache()
@@ -104,11 +107,11 @@ def activate_gpu_models(action):
                     "large-v3", 
                     device="cuda", 
                     compute_type="int8_float16", 
-                    num_workers=4, # RESTORED (User request)
+                    num_workers=4, 
                     local_files_only=local_only
                 )
             except Exception as e:
-                print(f"⚠️ Whisper GPU Peak failed: {e}. Falling back to standard mode.")
+                print(f"⚠️ GPU Init failed: {e}. Falling back to standard.")
                 MODELS["stt"] = WhisperModel("large-v3", device="cuda", compute_type="float16", local_files_only=local_only)
 
     # 2. XTTS-v2
@@ -119,12 +122,12 @@ def activate_gpu_models(action):
             tts_on_gpu = "cuda" in curr
         except: pass
         if MODELS["tts"] is None or not tts_on_gpu:
-            print(f"🔊 [v98] Activating XTTS-v2 (GPU)...")
+            print(f"🔊 [v99] Activating XTTS-v2 (GPU)...")
             if MODELS["tts"] is None:
                 MODELS["tts"] = TTS(model_name="tts_models/multilingual/multi-dataset/xtts_v2", gpu=True)
             else: MODELS["tts"].to("cuda")
 
-    # 3. Chatterbox Accelerated
+    # 3. Chatterbox GPU-Mode
     chatterbox_utils.load_chatterbox(device="cuda" if torch.cuda.is_available() else "cpu")
 
     # 4. Helpers
@@ -134,9 +137,9 @@ def activate_gpu_models(action):
     if MODELS["translate"] is None: MODELS["translate"] = "active"
 
 def release_gpu_models():
-    """v98: Clean Idle (Resident RAM)"""
+    """v99: Persistence in RAM (CPU)"""
     global MODELS
-    print("🧹 [v98] Session complete. Releasing GPU...")
+    print("🧹 [v99] Releasing GPU. Engines staying WARM in RAM.")
     try:
         if MODELS["stt"] and MODELS["stt"].model.device == "cuda":
             del MODELS["stt"]
@@ -150,19 +153,20 @@ def release_gpu_models():
     if torch.cuda.is_available(): torch.cuda.empty_cache()
 
 def warmup_task():
-    """Silent Power-Warmup"""
+    """V99: Resident RAM Pre-loading"""
     global WARMUP_STATUS
     with WARMUP_LOCK:
         if WARMUP_STATUS["complete"] or WARMUP_STATUS["in_progress"]: return
         WARMUP_STATUS["in_progress"] = True
-    print("\n🔥 --- SILENT POWER-WARMUP STARTED (v98) ---")
+    print("\n🔥 --- V99: PEAK WARMUP STARTED ---")
     try:
         MODELS["stt"] = WhisperModel("large-v3", device="cpu", compute_type="int8")
         MODELS["tts"] = TTS(model_name="tts_models/multilingual/multi-dataset/xtts_v2", gpu=False)
         chatterbox_utils.warmup_chatterbox()
         WARMUP_STATUS["complete"] = True
-        print(f"✅ --- PEAK READY -- \n")
-    except: pass
+        print(f"✅ --- SYSTEM READY --- \n")
+    except Exception as e:
+        print(f"❌ Warmup fail: {e}")
     finally: WARMUP_STATUS["in_progress"] = False
 
 def _stt_logic(request_dict):
@@ -171,7 +175,6 @@ def _stt_logic(request_dict):
     with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as f:
         f.write(audio_bytes); temp_path = f.name
     try:
-        # Multi-worker Power (4 workers)
         segments, _ = MODELS["stt"].transcribe(temp_path, language=lang, beam_size=1)
         return {"text": " ".join([s.text for s in segments]).strip()}
     finally:
@@ -215,7 +218,7 @@ def _tts_logic(text, lang, speaker_wav_b64):
 def core_process(request_dict):
     action = request_dict.get("action")
     t1 = time.time()
-    print(f"--- [v98] 🚀 PEAK GPU SESSION: {action} ---")
+    print(f"--- [v99] 🚀 GPU SESSION: {action} ---")
     activate_gpu_models(action)
     try:
         if action == "stt": res = _stt_logic(request_dict)
@@ -226,21 +229,31 @@ def core_process(request_dict):
             translated = _translate_logic(stt_res.get("text", ""), request_dict.get("target_lang"))
             tts_res = _tts_logic(translated, request_dict.get("target_lang"), request_dict.get("speaker_wav"))
             res = {"text": stt_res.get("text"), "translated": translated, "audio": tts_res.get("audio")}
-        elif action == "health": res = {"status": "awake"}
         else: res = {"error": f"Unknown action: {action}"}
     finally:
-        print(f"--- [v98] ✨ PEAK-DONE: {action} ({time.time()-t1:.2f}s) ---")
+        print(f"--- [v99] ✨ SUCCESS: {action} ({time.time()-t1:.2f}s) ---")
         release_gpu_models()
     return res
 
-app = FastAPI()
-@app.on_event("startup")
-async def startup_event():
+@asynccontextmanager
+async def lifespan(app: FastAPI):
+    # Startup: Background Warmup
     Thread(target=warmup_task, daemon=True).start()
+    yield
+    # Shutdown logic (optional)
+    pass
+
+app = FastAPI(lifespan=lifespan)
 
 @app.post("/api/v1/process")
 async def api_process(request: Request):
-    try: return core_process(await request.json())
+    try:
+        req_data = await request.json()
+        action = req_data.get("action")
+        # 🔥 V99 CRITICAL FIX: Handle 'health' on CPU to prevent GPU queue loops
+        if action == "health":
+            return {"status": "awake", "warm": WARMUP_STATUS["complete"]}
+        return core_process(req_data)
     except Exception as e: return {"error": str(e)}
 
 @app.get("/health")