Update app.py

set based on tokenizer ID and not just text

app.py

@@ -68,7 +68,10 @@ class ChatBot:
 
         # Tokenize
         inputs = self.tokenizer(prompt, return_tensors="pt").to(self.model.device)
-
+        
+        # Set EOS seen in training (per Qwen chat template) 
+        stop_id = tokenizer.convert_tokens_to_ids("<|im_end|>")
+        
         # Generate response
         with torch.no_grad():
             outputs = self.model.generate(
@@ -77,7 +80,8 @@ class ChatBot:
                 temperature=0.6,
                 do_sample=True,
                 return_dict_in_generate=True,
-                pad_token_id=self.tokenizer.eos_token_id
+                pad_token_id=self.tokenizer.eos_token_id,
+                eos_token_id=stop_id
             )
 
         # Decode and clean response, with multiple debugs