mymusise / ChatGLM-Tuning

基于ChatGLM-6B + LoRA的Fintune方案
MIT License
3.73k stars 440 forks source link

如何用中间checkpoint看效果 #184

Open 18335100284 opened 1 year ago

18335100284 commented 1 year ago

在没有全部训练完所有的step情况下,我想用中间的ckpt初步看下效果,例如: model = PeftModel.from_pretrained(model, "./output/checkpoint-1200/") 但是报错如下: ValueError: Can't find 'adapter_config.json' at './output/checkpoint-1200/' 所以在没有训练完所有step生成adapter_config.json之前,如何用中间ckpt看下效果?

suc16 commented 1 year ago

把checkpoint的adapter_model.bin复制到./output/,自己再复制一个adapter_config.json

adapter_config.json参考

{
  "base_model_name_or_path": "models/chatglm-6b",
  "bias": "none",
  "enable_lora": [
    true,
    false,
    true
  ],
  "fan_in_fan_out": false,
  "inference_mode": true,
  "lora_alpha": 32,
  "lora_dropout": 0.1,
  "merge_weights": false,
  "modules_to_save": null,
  "peft_type": "LORA",
  "r": 8,
  "target_modules": [
    "query_key_value"
  ],
  "task_type": "CAUSAL_LM"
}