Changes to be committed:

modified: Dockerfile
modified: create_app.py

Dockerfile

@@ -13,4 +13,4 @@ COPY --chown=user . /app
 
 RUN python -m spacy download en_core_web_sm
 
-CMD ["gunicorn", "--bind", "0.0.0.0:7860", "--timeout", "360", "--workers", "1", "--log-level", "debug", "app:app"]
+CMD ["gunicorn", "--bind", "0.0.0.0:7860", "--timeout", "360", "app:app"]

create_app.py

@@ -19,32 +19,28 @@ MODEL_SESSION = None
 def load_models():
     global MODELS_LOADED, LONGFORMER_TOKENIZER, LONGFORMER_MODEL, QWEN_TOKENIZER, QWEN_MODEL
     device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-    print(f"CUDA Available: {torch.cuda.is_available()}")  
-    print(f"CUDA Device Name: {torch.cuda.get_device_name(0)}")  
-    print(f"Free GPU Memory: {torch.cuda.memory_reserved(0)/1e9:.2f} GB")  
     print("DEIVCE=", device)
-    print("WHY DONT PRINT")
     if not MODELS_LOADED:
-        LONGFORMER_TOKENIZER = LongformerTokenizer.from_pretrained('allenai/longformer-base-4096', device='auto')
-        config = LongformerConfig.from_json_file("Longformer_checkpoint/config.json")
-        LONGFORMER_MODEL = CustomLongformerForSequenceClassification(config).from_pretrained('SFM2001/LongFormerScorer')
-        LONGFORMER_MODEL = LONGFORMER_MODEL.to(device)
-        LONGFORMER_MODEL.eval()
         print("START TO GET QWEN")
-        model_name = 'Qwen/Qwen3-1.7B'
+        model_name = 'Qwen/Qwen3-0.6B'
         QWEN_TOKENIZER = AutoTokenizer.from_pretrained(model_name, device='auto')
         QWEN_TOKENIZER.pad_token_id = QWEN_TOKENIZER.eos_token_id
         print("QWEN TOKENIZER LOADED")
         try:
-            QWEN_MODEL = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.float16, device_map="auto", trust_remote_code=True)
+            QWEN_MODEL = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto")
             QWEN_MODEL = QWEN_MODEL.to(device)
-            print("QWEN MODEL LOADED")  # Typo fixed from your original code ("LOADED" vs. "LOADED")
+            print("QWEN MODEL LOADED", flush=True)
         except Exception as e:
             print(f"ERROR LOADING QWEN MODEL: {str(e)}")
             raise  # Re-raise to see full traceback
         # QWEN_MODEL = AutoModelForCausalLM.from_pretrained(model_name).half()
         # QWEN_MODEL = QWEN_MODEL.to(device)
         print("QWEN MODEL LOADED")
+        LONGFORMER_TOKENIZER = LongformerTokenizer.from_pretrained('allenai/longformer-base-4096', device='auto')
+        config = LongformerConfig.from_json_file("Longformer_checkpoint/config.json")
+        LONGFORMER_MODEL = CustomLongformerForSequenceClassification(config).from_pretrained('SFM2001/LongFormerScorer')
+        LONGFORMER_MODEL = LONGFORMER_MODEL.to(device)
+        LONGFORMER_MODEL.eval()
         MODELS_LOADED = True
     print("LOAD ENDED")