debug: attention

src/chat_logic.py

@@ -27,11 +27,19 @@ class ChatProcessor:
                 skip_special_tokens=True
             )
 
-            input_ids = pipe.tokenizer(prompt, return_tensors="pt").input_ids.to(pipe.model.device)
+            # Get full tokenizer output
+            tokenized_inputs = pipe.tokenizer(prompt, return_tensors="pt")
             
-            # Prepare generation kwargs
+            # Determine model device
+            device = pipe.model.device
+            
+            # Move all tensors to the correct device
+            inputs_on_device = {k: v.to(device) for k, v in tokenized_inputs.items()}
+            
+            # Prepare generation kwargs with attention_mask
             generate_kwargs = {
-                "input_ids": input_ids,
+                "input_ids": inputs_on_device["input_ids"],
+                "attention_mask": inputs_on_device["attention_mask"],
                 "max_new_tokens": max_new_tokens,
                 "temperature": temperature,
                 "top_p": top_p,