Giter VIP home page Giter VIP logo

Comments (15)

EnghishYang avatar EnghishYang commented on July 30, 2024

我也遇到了类似的问题,对于作者的代码去掉peft的部分,是不能直接运行的。

from chatglm-tuning.

mymusise avatar mymusise commented on July 30, 2024

我在colab上的p100也有这个问题,但是换成了T4后就没了 , 不确定是不是硬件问题

from chatglm-tuning.

EnghishYang avatar EnghishYang commented on July 30, 2024

我在colab上的p100也有这个问题,但是换成了T4后就没了 , 不确定是不是硬件问题

大佬是直接去掉peft的部分试的吗

from chatglm-tuning.

bobo-wmdigit avatar bobo-wmdigit commented on July 30, 2024

我在colab上的p100也有这个问题,但是换成了T4后就没了 , 不确定是不是硬件问题

难道因为4090太新了?我已经升级到pytorch 2.0了,想不通还要怎么解决,我如果直接把allow_fp16 = True,会导致loss=nan,脑壳疼。

from chatglm-tuning.

mymusise avatar mymusise commented on July 30, 2024

我在colab上的p100也有这个问题,但是换成了T4后就没了 , 不确定是不是硬件问题

难道因为4090太新了?我已经升级到pytorch 2.0了,想不通还要怎么解决,我如果直接把allow_fp16 = True,会导致loss=nan,脑壳疼。

这种情况可能是环境依赖问题 🤣 我在4090上测过没这问题

from chatglm-tuning.

mymusise avatar mymusise commented on July 30, 2024

另外pytorch2.0我还没试过,目前用的1.13.1

from chatglm-tuning.

bobo-wmdigit avatar bobo-wmdigit commented on July 30, 2024

你跑的是带不带lora的啊?我用你的标准脚本没问题,但是我砍掉了lora那部分,也就是peft的部分,就不行了。。。

from chatglm-tuning.

mymusise avatar mymusise commented on July 30, 2024

你跑的是带不带lora的啊?我用你的标准脚本没问题,但是我砍掉了lora那部分,也就是peft的部分,就不行了。。。

我这样跑下来是没有问题的: https://github.com/mymusise/ChatGLM-Tuning/blob/master/examples/finetune.ipynb

from chatglm-tuning.

mymusise avatar mymusise commented on July 30, 2024

噢,抱歉,你说的是直接finetune原模型,这个我没试过,直接finetune可以参考别的repo:https://github.com/ssbuild/chatglm_finetuning

from chatglm-tuning.

bobo-wmdigit avatar bobo-wmdigit commented on July 30, 2024

嗯,那个repo我看了,话说,你的新版本好像显存占用增加了,我以前可以跑bs = 2现在只能跑1了,同样4090

from chatglm-tuning.

mymusise avatar mymusise commented on July 30, 2024

你跑的是512长度么

from chatglm-tuning.

bobo-wmdigit avatar bobo-wmdigit commented on July 30, 2024

image
大佬看下我的配置对吗?

from chatglm-tuning.

bucm-tcm-tool avatar bucm-tcm-tool commented on July 30, 2024

你跑的是带不带lora的啊?我用你的标准脚本没问题,但是我砍掉了lora那部分,也就是peft的部分,就不行了。。。

试试注释掉finetune.py的第128行:model.lm_head = CastOutputToFloat(model.lm_head)

from chatglm-tuning.

mymusise avatar mymusise commented on July 30, 2024

image 大佬看下我的配置对吗?

模型的配置用官方默认的就可以

你跑的是512长度么

我这块是说训练的样本的最大长度,之前数据预处理脚本有点问题,最大长度限制没起作用,现在修复了,可以拉下最新的代码重跑下数据预处理再试试,我这边同样24G显存跑320长度, bs=2是没问题

from chatglm-tuning.

hjing100 avatar hjing100 commented on July 30, 2024

python fineturn.py的fp16改成bf16试试

from chatglm-tuning.

Related Issues (20)

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google ❤️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.