SpongebBob/Finetune-ChatGLM2-6B

Issues

• 请问可以训练chatGLM2-6B-32K的模型嘛

  #23 opened a year ago by GuocaiL
  0

• 基于school_math_0.25M.json数据集进行微调训练后得到的模型推理效果很差，是什么原因？

  #22 opened a year ago by ivankxt
  0

• 全量微调显存占用

  #1 opened 2 years ago by ACXuLiu
  8

• ChatGLM数据处理中input_ids的padding问题

  #21 opened a year ago by nostalgiaer
  0

• 微调时在runing tokenizer on train dataset这一步卡住

  #20 opened a year ago by underwoodnoble
  1

• 我只有200多条多轮对话的数据，去做全参微调能有效果吗？

  #19 opened a year ago by ymmbb8882ymmbb
  2

• 微调出来的模型可以直接用chatglm2官网发版的web_demo2.py来测吗，会有影响吗？

  #18 opened a year ago by ymmbb8882ymmbb
  1

• 模型训练，出现deepspeed_init() got an unexpected keyword argument 'resume_from_checkpoint'

  #12 opened a year ago by reyvand14
  4

• max_len 768 这个长度在多轮对话中，是单次输入的长度，还是history+这次输入的长度？ 如果是单次输入的长度，那么history的长度在哪里？

  #16 opened a year ago by ymmbb8882ymmbb
  0

• --max_length 762，我想问一下这个是输入的最大长度吗，如果是，那么如果是多轮对话，有长度限制吗？

  #15 opened a year ago by ymmbb8882ymmbb
  0

• timed out

  #6 opened a year ago by LLLLLLoki
  1

• 有考虑 加入lora训练吗

  #9 opened a year ago by nsl2014fm
  1

• 请问这个问题怎么解决OSError: Can't get source for <function apply_rotary_pos_emb at 0x7fef8c15f790>. TorchScript requires source access in order to carry out compilation, make sure original .py files are available.

  #14 opened a year ago by ymmbb8882ymmbb
  0

• 请问全参数微调需要什么配置呢？

  #10 opened a year ago by bigbigwatermalon
  9

• 请问这个项目的依赖是什么，安装的deepspeed版本好像跑不了。

  #11 opened a year ago by Hollywood3
  1

• 训练数据过多时报错Socket Timeout

  #13 opened a year ago by onair1314
  0

• 模型训练，input_ids出现None type

  #7 opened a year ago by Fanshell2333
  2

• 为什么不用新版本的transformer么？

  #8 opened a year ago by sunlei198911
  0

• CUDA out of memory. Tried to allocate 11.63 GiB (GPU 0; 23.69 GiB total capacity; 11.63 GiB already allocated; 11.28 GiB free

  #5 opened a year ago by harbor1981
  1

• 多轮数据下载后要如何处理?

  #4 opened 2 years ago by wac81
  0

• do_eval时报错

  #3 opened 2 years ago by kpg-pgk
  1

• RuntimeError: CUDA error: invalid device ordinal

  #2 opened 2 years ago by zxy333666
  1