ChatGLM-6B
ChatGLM-6B copied to clipboard
THUDM
[BUG/Help] <title>请问如何基于之前的Ptuning模型，再次进行微调
Is there an existing issue for this?
- [X] I have searched the existing issues
Current Behavior
已经训练了200次，生成了一个模型。再次执行训练时，又从头开始跑了
Expected Behavior
No response
Steps To Reproduce
已经训练了200次，生成了一个模型。再次执行训练时，又从头开始跑了
Environment
- OS:
- Python:
- Transformers:
- PyTorch:
- CUDA Support (`python -c "import torch; print(torch.cuda.is_available())"`) :
Anything else?
No response
这个问题已经解决 谢谢 还有一个问题请教 { "prompt": "请用C#帮我实现", "response": "using System;using System.Net.Http;using System.Text;using System.Threading.Tasks;using Newtonsoft.Json;class Program{static async Task Main(){string url="http://127.0.0.1/hangbanguanjia";var requestParams=new{safd="111"};string response=await SendRequestAsync(url,requestParams);Console.WriteLine("Response: "+response);}static async Task<string> SendRequestAsync(string url,object requestParams){using(HttpClient client=new HttpClient()){var content=new StringContent(JsonConvert.SerializeObject(requestParams),Encoding.UTF8,"application/json");HttpResponseMessage response=await client.PostAsync(url,content);response.EnsureSuccessStatusCode();return await response.Content.ReadAsStringAsync();}}}", "history": [ [ "航班管家的出票地址是多少?", "航班管家的出票地址是http://127.0.0.1/hangbanguanjia,请注意，为了数据安全，ip地址为虚拟ip，请替换成实际ip" ], [ "航班管家的请求参数是？", "航管管家的请求参数类型是application/json，参数为{'safd':'111'}" ] ] } 这个是我微调的数据集 但是当我跑完运行之后 问的问题和想要的完全不符，请问知道怎么回事吗
------------------ 原始邮件 ------------------ 发件人: @.>; 发送时间: 2023年4月17日(星期一) 下午2:29 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [THUDM/ChatGLM-6B] [BUG/Help] <title>请问如何基于之前的Ptuning模型，再次进行微调 (Issue #639)
你好 请问你是用的什么配置运行微调的 我运行微调报内存不足 我用的V100 16G
我用的M40 24G
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>
https://github.com/SHITIANYU-hue/ChatGLM-6B/blob/main/ptuning/main.py#L113
https://github.com/SHITIANYU-hue/ChatGLM-6B/blob/main/ptuning/main.py#L113
对，这段代码我也看到了，通过你上面在do_train加上--ptuning_checkpoint是可以继续微调的对吗？然后微调的结果又再一次覆盖到--output_dir吗？
对的
最后再通过这样运行吗？
MODEL_PATH = "./model/chatglm-6b"
CHECKPOINT_PATH = "./output/adgen-chatglm-6b-pt-128-2e-2/checkpoint-1000"
# 载入Tokenizer
tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH, trust_remote_code=True)
config = AutoConfig.from_pretrained(MODEL_PATH, trust_remote_code=True, pre_seq_len=128)
model = AutoModel.from_pretrained(MODEL_PATH, config=config, trust_remote_code=True).cuda()
prefix_state_dict = torch.load(os.path.join(CHECKPOINT_PATH, "pytorch_model.bin"))
new_prefix_state_dict = {}
for k, v in prefix_state_dict.items():
if k.startswith("transformer.prefix_encoder."):
new_prefix_state_dict[k[len("transformer.prefix_encoder."):]] = v
model.transformer.prefix_encoder.load_state_dict(new_prefix_state_dict)
