new message format3

app.py

@@ -19,17 +19,19 @@ class StopOnTokens(StoppingCriteria):
 
 
 def predict(message, history):
-    history_transformer_format = history + [[message, ""]]
+    print(history)
+    history_transformer_format = history + [{"role": "user", "content": message}, 
+                                            {"role": "assistant", "content": ""}]
     stop = StopOnTokens()
 
     # messages = "".join(["".join(["<|start_header_id|>user<|end_header_id|>\n"+item[0], 
     #                              "<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n"+item[1]])
     #             for item in history_transformer_format])
-    messages = [{"user": item[0], "content": item[1]} for item in history_transformer_format]
-    print(messages)
+    # messages = [{"role": "user", item[0], "content": item[1]} for item in history_transformer_format]
+    #print(messages)
 
     # model_inputs = tokenizer([messages], return_tensors="pt") # .to("cuda")
-    model_inputs = tokenizer.apply_chat_template(messages, return_tensors="pt")
+    model_inputs = tokenizer.apply_chat_template(history_transformer_format, return_tensors="pt")
     streamer = TextIteratorStreamer(tokenizer, timeout=10., skip_prompt=True, skip_special_tokens=True)
     generate_kwargs = dict(
         model_inputs,