Update app.py

app.py

@@ -40,49 +40,42 @@ def load_model(model_type):
     manage_resources()
     
     try:
-        # Set lower precision to reduce memory usage
-        torch_dtype = torch.float32
-        if torch.cuda.is_available():
-            device = "cuda"
-        else:
-            device = "cpu"
-            torch_dtype = torch.float32  # Use float32 for CPU
-        
+        # For CPU-only environment, don't use device_map
         if model_type == "summarize":
             base_model = AutoModelForSeq2SeqLM.from_pretrained(
                 "facebook/bart-large-cnn",
                 cache_dir="./models",
-                torch_dtype=torch_dtype,
+                torch_dtype=torch.float32,
                 low_cpu_mem_usage=True
             )
             model = PeftModel.from_pretrained(
                 base_model, 
                 "pendar02/results",
-                device_map=device,
-                torch_dtype=torch_dtype
+                torch_dtype=torch.float32
             )
             tokenizer = AutoTokenizer.from_pretrained(
                 "facebook/bart-large-cnn",
                 cache_dir="./models"
             )
         else:  # question_focused
-            base_model = AutoModelForSeq2SeqLM.from_pretrained(
+            base_model = AutoModelForSeq2SeqLation_model = AutoModelForSeq2SeqLM.from_pretrained(
                 "GanjinZero/biobart-base",
                 cache_dir="./models",
-                torch_dtype=torch_dtype,
+                torch_dtype=torch.float32,
                 low_cpu_mem_usage=True
             )
             model = PeftModel.from_pretrained(
                 base_model, 
                 "pendar02/biobart-finetune",
-                device_map=device,
-                torch_dtype=torch_dtype
+                torch_dtype=torch.float32
             )
             tokenizer = AutoTokenizer.from_pretrained(
                 "GanjinZero/biobart-base",
                 cache_dir="./models"
             )
         
+        # Ensure model is on CPU
+        model = model.cpu()
         model.eval()
         return model, tokenizer
     except Exception as e: