Update code/inference.py

code/inference.py

@@ -15,12 +15,12 @@ def predict_fn(data, load_list):
     char_name = request_inputs["char_name"]
     user_name = request_inputs["user_name"]
     template = open(f"{template}.txt", "r").read()
-    user_input = [
+    user_input = "\n".join([
         "{name}: {message}".format(
             name = char_name if (id["role"] == "AI") else user_name,
             message = id["message"].strip()
         ) for id in messages
-    ]
+    ])
     prompt = template.format(char_name = char_name, user_name = user_name, user_input = user_input)
     input_ids = tokenizer(prompt + f"\n{char_name}:", return_tensors = "pt").to("cuda")
     encoded_output = model.generate(