Upload app.py with huggingface_hub

app.py

@@ -1,9 +1,9 @@
-# 🚀 V106: ZEROGPU PRIORITY IMPORT
+# 🚀 V107: ZEROGPU HARDENING
 # Must be first to patch environment correctly
 try:
     import spaces
 except ImportError:
-    print("⚠️ 'spaces' not installed. ZeroGPU features disabled.")
+    print("⚠️ 'spaces' not installed.")
     class spaces:
         @staticmethod
         def GPU(duration=60, f=None):
@@ -30,7 +30,7 @@ import logging
 from threading import Thread, Lock
 from huggingface_hub import snapshot_download
 
-# 🛡️ 1. SILENCE LOGS & WARNINGS (v106: Verified Silence)
+# 🛡️ 1. SILENCE LOGS & WARNINGS (v107: Stability Milestone)
 logging.getLogger("transformers").setLevel(logging.ERROR)
 logging.getLogger("TTS").setLevel(logging.ERROR)
 logging.getLogger("onnxruntime").setLevel(logging.ERROR)
@@ -84,8 +84,8 @@ from TTS.api import TTS
 from df.enhance import init_df
 import deep_translator
 
-# FORCE BUILD TRIGGER: 14:45:00 Jan 21 2026
-# v106: Import Priority + Explicit Queue.
+# FORCE BUILD TRIGGER: 17:10:00 Jan 21 2026
+# v107: Whisper int8 stability. Gradio 5.9.1.
 
 os.environ["COQUI_TOS_AGREED"] = "1"
 MODELS = {"stt": None, "translate": None, "tts": None, "denoiser": None}
@@ -94,7 +94,7 @@ WARMUP_STATUS = {"complete": False, "in_progress": False}
 WARMUP_LOCK = Lock()
 
 def activate_gpu_models(action):
-    """v106: Stable Activation"""
+    """v107: Safe Hardware Activation"""
     global MODELS, WARMUP_STATUS
     local_only = WARMUP_STATUS["complete"]
     
@@ -103,17 +103,19 @@ def activate_gpu_models(action):
         try: stt_on_gpu = MODELS["stt"] is not None and MODELS["stt"].model.device == "cuda"
         except: pass
         if not stt_on_gpu:
-            print(f"🎙️ [v106] Activating Whisper (GPU: int8_float16)...")
+            print(f"🎙️ [v107] Activating Whisper (GPU: int8 Protocol)...")
             try:
                 if MODELS["stt"]: del MODELS["stt"]
                 gc.collect(); torch.cuda.empty_cache()
                 time.sleep(0.5)
+                # v107: Using 'int8' for guaranteed stability on H200 MIG. 
+                # Removed device_index to allow driver-level discovery.
                 MODELS["stt"] = WhisperModel(
                     "large-v3", 
                     device="cuda", 
-                    device_index=0,
-                    compute_type="int8_float16", 
+                    compute_type="int8", 
                     num_workers=1,
+                    cpu_threads=1,
                     local_files_only=local_only
                 )
             except Exception as e:
@@ -127,7 +129,7 @@ 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"🔊 [v106] Activating XTTS-v2 (GPU)...")
+            print(f"🔊 [v107] Activating XTTS-v2 (GPU)...")
             try:
                 if MODELS["tts"] is None:
                     MODELS["tts"] = TTS(model_name="tts_models/multilingual/multi-dataset/xtts_v2", gpu=True)
@@ -142,9 +144,9 @@ def activate_gpu_models(action):
     if MODELS["translate"] is None: MODELS["translate"] = "active"
 
 def release_gpu_models():
-    """v106: Resident Cleanup"""
+    """v107: GPU Cleanup"""
     global MODELS
-    print("🧹 [v106] Releasing GPU resources.")
+    print("🧹 [v107] Releasing GPU resources.")
     try:
         if MODELS["stt"] and MODELS["stt"].model.device == "cuda":
             del MODELS["stt"]
@@ -159,18 +161,18 @@ def release_gpu_models():
     time.sleep(0.5)
 
 def warmup_task():
-    """Silent Warmup (v106)"""
+    """Silent Warmup (v107)"""
     global WARMUP_STATUS
     with WARMUP_LOCK:
         if WARMUP_STATUS["complete"] or WARMUP_STATUS["in_progress"]: return
         WARMUP_STATUS["in_progress"] = True
-    print("\n🔥 --- V106: ZEROGPU RECOVERY WARMUP ---")
+    print("\n🔥 --- V107: ZEROGPU RECOVERY 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"✅ --- SYSTEM READY: v106 --- \n")
+        print(f"✅ --- SYSTEM READY: v107 --- \n")
     except: pass
     finally: WARMUP_STATUS["in_progress"] = False
 
@@ -222,7 +224,7 @@ def _tts_logic(text, lang, speaker_wav_b64):
 def core_process(request_dict):
     action = request_dict.get("action")
     t1 = time.time()
-    print(f"--- [v106] 🚀 GPU SESSION: {action} ---")
+    print(f"--- [v107] 🚀 GPU SESSION: {action} ---")
     activate_gpu_models(action)
     try:
         if action == "stt": res = _stt_logic(request_dict)
@@ -235,7 +237,7 @@ def core_process(request_dict):
             res = {"text": stt_res.get("text"), "translated": translated, "audio": tts_res.get("audio")}
         else: res = {"error": f"Unknown action: {action}"}
     finally:
-        print(f"--- [v106] ✨ SUCCESS: {action} ({time.time()-t1:.2f}s) ---")
+        print(f"--- [v107] ✨ SUCCESS: {action} ({time.time()-t1:.2f}s) ---")
         release_gpu_models()
     return res
 
@@ -244,7 +246,7 @@ async def lifespan(app: FastAPI):
     Thread(target=warmup_task, daemon=True).start()
     yield
 
-# 🚀 PRIMARY FASTAPI
+# 🚀 STEP 1: DEFINE FASTAPI
 app = FastAPI(lifespan=lifespan)
 app.add_middleware(CORSMiddleware, allow_origins=["*"], allow_methods=["*"], allow_headers=["*"])
 
@@ -258,7 +260,7 @@ async def api_process(request: Request):
     except Exception as e: return {"error": str(e)}
 
 @app.get("/health")
-def health(): return {"status": "ok", "warm": WARMUP_STATUS["complete"], "v": "106"}
+def health(): return {"status": "ok", "warm": WARMUP_STATUS["complete"], "v": "107"}
 
 @app.post("/api/v1/clear_cache")
 async def clear_cache_api():
@@ -267,18 +269,17 @@ async def clear_cache_api():
         return {"status": "success"}
     except: return {"status": "error"}
 
+# 🚀 STEP 2: DEFINE GRADIO
 def gradio_fn(req_json):
     try: return json.dumps(core_process(json.loads(req_json)))
     except Exception as e: return json.dumps({"error": str(e)})
 
-# 🚀 GRADIO WITH QUEUE
-demo = gr.Interface(fn=gradio_fn, inputs="text", outputs="text", title="🚀 AI Engine v106")
-# 🔥 V106: EXPLICIT QUEUE. ZeroGPU needs this.
+demo = gr.Interface(fn=gradio_fn, inputs="text", outputs="text", title="🚀 AI Engine v107")
 demo.queue()
 
 # MOUNT
 app = gr.mount_gradio_app(app, demo, path="/")
 
 if __name__ == "__main__":
-    print("🚀 [v106] Starting Unified Server (ZeroGPU Fixed)...")
+    print("🚀 [v107] Starting Unified Server (ZeroGPU Recovery)...")
     uvicorn.run(app, host="0.0.0.0", port=7860, log_level="error")