Update app.py

app.py

@@ -1,5 +1,5 @@
 from threading import Thread
-
+import torch
 import gradio as gr
 import spaces
 from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
@@ -17,7 +17,7 @@ BANNER_HTML = """
     </center>
 </h3>
 <p>
-    <center><em>The demo is mainly for academic purposes. Do not use this demo for illegal activities. Default model: <a href="https://huggingface.co/hfl/llama-3-chinese-8b-instruct-v3">hfl/llama-3-chinese-8b-instruct-v3</a></em></center>
+    <center><em>The demo is mainly for academic purposes and users are not expected to use this demo for illegal activities.</em></center>
 </p>
 """
 
@@ -34,7 +34,7 @@ def load_model(version):
         model_name = "hfl/llama-3-chinese-8b-instruct-v3"
     
     tokenizer = AutoTokenizer.from_pretrained(model_name)
-    model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto")
+    model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16, device_map="auto", attn_implementation="flash_attention_2")
     return f"Model {model_name} loaded."
 
 @spaces.GPU(duration=50)
@@ -77,7 +77,7 @@ with gr.Blocks() as demo:
             gr.Text(value=DEFAULT_SYSTEM_PROMPT, label="System Prompt / 系统提示词", render=False),
             gr.Radio(choices=["v1", "v2", "v3"], label="Model Version / 模型版本", value="v3", interactive=False, render=False),
             gr.Slider(minimum=0, maximum=1, step=0.1, value=0.5, label="Temperature / 温度系数", render=False),
-            gr.Slider(minimum=128, maximum=2048, step=1, value=256, label="Max new tokens / 最大生成长度", render=False),
+            gr.Slider(minimum=512, maximum=2048, step=1, value=256, label="Max new tokens / 最大生成长度", render=False),
         ],
         cache_examples=False,
     )