Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

微调代码什么时候能够发布 #2

Closed
bltcn opened this issue Oct 27, 2023 · 27 comments
Closed

微调代码什么时候能够发布 #2

bltcn opened this issue Oct 27, 2023 · 27 comments

Comments

@bltcn
Copy link

bltcn commented Oct 27, 2023

关心这个问题,谢谢

@shangzhensen
Copy link

+1

@jlokys
Copy link

jlokys commented Oct 27, 2023

+1支持

@gaojuntian
Copy link

+1

@LZHgrla
Copy link

LZHgrla commented Oct 27, 2023

推荐一手我们团队开发的微调工具库:XTuner
目前已经支持了 ChatGLM3-6B-Base 的微调;同时,在数据集处理逻辑上,我们也进行了精心设计,方便拓展自定义数据。

一键启动

ChatGLM3-6B-Base, QLoRA, open assistant 数据集(显存占用 11GB 左右)

pip install xtuner==0.1.6
xtuner train chatglm3_6b_base_qlora_oasst1_e3

@mockyd
Copy link

mockyd commented Oct 27, 2023

chatGLM2的微调代码适用不?很好奇都是同一个系列的模型,为什么微调代码不能共用呀?

@JamePeng
Copy link
Contributor

@mockyd
Copy link

mockyd commented Oct 28, 2023

魔搭那边支持了:https://github.com/modelscope/swift/tree/main/examples/pytorch/llm/scripts/chatglm3_6b_32k/lora_ddp_ds

想问下大佬,想用多轮对话数据训练chatglm3,应该怎么组织数据呀?魔搭这个文档我没看明白要怎么组织。。。

@bank010
Copy link

bank010 commented Oct 28, 2023

chatGLM2的微调代码适用不?很好奇都是同一个系列的模型,为什么微调代码不能共用呀?

输入格式不一样

@xxw1995
Copy link
Contributor

xxw1995 commented Oct 28, 2023

@leoluopy
Copy link

好东西, mark 一下

@WangRongsheng
Copy link

LLaMA-Factory is all you need: https://github.com/hiyouga/LLaMA-Factory

image

@leoluopy
Copy link

leoluopy commented Oct 29, 2023

@WangRongsheng does LLaMA-Factory support GLM2-6b and using Qlora SFT . in several steps ?

@leoluopy
Copy link

@LZHgrla how to use xtuner in command line to train my custom dataset , mode is QLora . any guide doc link ?

@LZHgrla
Copy link

LZHgrla commented Oct 29, 2023

@LZHgrla how to use xtuner in command line to train my custom dataset , mode is QLora . any guide doc link ?

Single-turn conversation Docs: zh_cn, en

Multi-turn conversation Docs: zh_cn, en

@WangRongsheng
Copy link

@WangRongsheng does LLaMA-Factory support GLM2-6b and using Qlora SFT . in several steps ?

Yes, it can do.

@leoluopy
Copy link

leoluopy commented Oct 29, 2023

@LZHgrla following single turn conversation doc guide : i got this error: NotImplementedError: Loading a dataset cached in a LocalFileSystem is not supported.
any ideas ?? @LZHgrla
leo@leo-System-Product-Name:~/Downloads/mvp/work_dirs$ xtuner -v
10/29 20:58:18 - mmengine - INFO - 0.1.6

@LZHgrla
Copy link

LZHgrla commented Oct 29, 2023

@LZHgrla following single turn conversation doc guide : i got this error: NotImplementedError: Loading a dataset cached in a LocalFileSystem is not supported. any ideas ?? @LZHgrla leo@leo-System-Product-Name:~/Downloads/mvp/work_dirs$ xtuner -v 10/29 20:58:18 - mmengine - INFO - 0.1.6

You can try pip install -U datasets

If you have further questions, please post them on here

@leoluopy
Copy link

@LZHgrla thanks , and i've finally launched up my QLora fine tune .

@yang1111-gif
Copy link

marked

1 similar comment
@zhouao0314
Copy link

marked

@zhangxinfang520
Copy link

推荐一手我们团队开发的微调工具库:XTuner 目前已经支持了 ChatGLM3-6B-Base 的微调;同时,在数据集处理逻辑上,我们也进行了精心设计,方便拓展自定义数据。

一键启动

ChatGLM3-6B-Base, QLoRA, open assistant 数据集(显存占用 11GB 左右)

pip install xtuner==0.1.6
xtuner train chatglm3_6b_base_qlora_oasst1_e3

使用xtuner train 微调chatglm3后 无法生成 adapter_config.json 导致qlora训练后的权重无法使用@LZHgrla

@huwen2117
Copy link

关心这个问题,谢谢

+1

@minghaochen
Copy link

https://github.com/minghaochen/chatglm3-base-tuning

chatglm3发布了,这次还发了base版本的模型,意味着我们可以基于这个base模型去自由地做SFT了。本项目实现了基于base模型的多轮对话SFT。

@yaoxingwei
Copy link

关心这个问题,谢谢

+1

@LZHgrla
Copy link

LZHgrla commented Nov 2, 2023

推荐一手我们团队开发的微调工具库:XTuner 目前已经支持了 ChatGLM3-6B-Base 的微调;同时,在数据集处理逻辑上,我们也进行了精心设计,方便拓展自定义数据。

一键启动

ChatGLM3-6B-Base, QLoRA, open assistant 数据集(显存占用 11GB 左右)

pip install xtuner==0.1.6
xtuner train chatglm3_6b_base_qlora_oasst1_e3

使用xtuner train 微调chatglm3后 无法生成 adapter_config.json 导致qlora训练后的权重无法使用@LZHgrla

我们这边测试并不会出现这个问题,训练后经过转换可以直接获得qlora的adapter权重

@lhtpluto
Copy link

lhtpluto commented Nov 3, 2023

微调代码什么时候能够发布?

@zhangch9
Copy link
Contributor

zhangch9 commented Nov 6, 2023

微调代码已发布,请参考 ChatGLM3-6B 微调示例

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests