Update app.py

app.py

@@ -38,10 +38,10 @@ def generate(
     message: str,
     chat_history: list[tuple[str, str]],
     system_prompt: str,
-    max_new_tokens: int = 512,
-    temperature: float = 0.2,
+    max_new_tokens: int = 1024,
+    temperature: float = 0.6,
     top_p: float = 0.9,
-    top_k: int = 10,
+    top_k: int = 50,
     repetition_penalty: float = 1.2,
 ) -> Iterator[str]:
     conversation = []
@@ -49,17 +49,10 @@ def generate(
         conversation.append({"role": "system", "content": system_prompt})
     for user, assistant in chat_history:
         conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
-    prompt = f"""### Instruction:
-                根据巨杉数据库SequoiaDB的相关问题进行回答。
-                
-                ### Input:
-                {message}
-                
-                ### Response:
-            """
-    conversation.append({"role": "user", "content": prompt})
+    conversation.append({"role": "user", "content": message})
+
     chat = tokenizer.apply_chat_template(conversation, tokenize=False)
-    inputs = tokenizer(chat, return_tensors="pt", truncation=True).input_ids.cuda()
+    inputs = tokenizer(chat, return_tensors="pt", add_special_tokens=False).to("cuda")
     if len(inputs) > MAX_INPUT_TOKEN_LENGTH:
         inputs = inputs[-MAX_INPUT_TOKEN_LENGTH:]
         gr.Warning("Trimmed input from conversation as it was longer than {MAX_INPUT_TOKEN_LENGTH} tokens.")