Update app.py

app.py

@@ -84,13 +84,23 @@ class ChatBot:
                 eos_token_id=[self.tokenizer.eos_token_id] + stop_token_ids,
             )
 
-        # Decode and clean response
-        response = self.tokenizer.decode(
-            outputs[0][len(inputs.input_ids[0]) :], skip_special_tokens=True
-        )
-        response = THINK_TAG_PATTERN.sub("", response).strip()
-        print("------------------------------------------------------------------------------------RESPONSE------------------------------------------------------------------------------------")
-        print(response)
+        # Decode and clean response, with multiple debugs
+        # Inspect token IDs
+        gen_ids = outputs.sequences[0][len(inputs.input_ids[0]):]
+        print("\n--- GENERATED TOKEN IDS ---\n", gen_ids.tolist())
+        
+        # Decode without skipping specials
+        raw_output = self.tokenizer.decode(gen_ids, skip_special_tokens=False)
+        print("\n--- RAW DECODED OUTPUT ---\n", repr(raw_output))
+        
+        # Show first generated token decoded individually
+        if len(gen_ids) > 0:
+            first_token = self.tokenizer.decode([gen_ids[0]])
+            print(f"\n--- FIRST TOKEN --- '{first_token}' ---")
+        
+        # Clean as usual
+        response = THINK_TAG_PATTERN.sub("", raw_output).strip()
+        print("\n--- CLEANED RESPONSE ---\n", repr(response))
         
 
         return history + [(message, response)]