update server

server.py

@@ -32,10 +32,7 @@ raw_model = LlamaForCausalLM(
         shared_input_output_embedding=True,
     )
 )
-ckpt = torch.load(
-    "data/saved_ckpt/instruction_tuning/14001.pt",
-    map_location="cpu",
-)
+ckpt = torch.load("data/saved_ckpt/instruction_tuning/14001.pt", map_location="cpu")
 raw_model.load_state_dict(ckpt)
 raw_model.eval()
 model = raw_model.cuda()
@@ -68,10 +65,8 @@ demo = gr.Interface(
         "Can you explain to me what quantum mechanics is and how it relates to quantum computing?",
         "I'm feeling a bit unwell in my stomach today. Do you have any suggestions for dinner?",
     ],
-).queue(
-    concurrency_count=1,
     title="Open-Llama",
-    description="不基于其他预训练模型，完全使用Open-Llama项目从0开始训练的Instruct-GPT模型，总共训练花费在2w美元以内，训练时间80h。",
+    description="不基于其他预训练模型，完全使用[Open-Llama](https://github.com/Bayes-Song/Open-Llama)项目从0开始训练的Instruct-GPT模型，总训练成本不超过2w美元。由于请求需要经Gradio进行转发，可能出现请求丢失的现象，当长时间无响应（如20s以上）可刷新重试。当前体验服务生成的所有内容都是由人工智能模型生成，我们对其生成内容的准确性、完整性和功能性不做任何保证，并且其生成的内容不代表我们的态度或观点。",
     article="[关于作者](http://home.ustc.edu.cn/~sl9292/resume.html)",
-)
+).queue(concurrency_count=1)
 demo.launch(share=True)