debug

app.py

@@ -170,10 +170,9 @@ def get_response(params):
     # RuntimeError: The size of tensor a (2) must match the size of tensor b (3) at non-singleton dimension 0
     generate_kwargs = dict(
         inputs=input_ids,
-        do_sample=do_sample,
-        temperature=temperature,
-        top_p=top_p,
-        max_new_tokens=max_new_tokens,
+        do_sample=False,
+        top_p=1.0,
+        max_new_tokens=512,
         pad_token_id=tokenizer.pad_token_id,
         streamer=streamer,
         # stopping_criteria=[stopping_criteria],