dvlab-research
Can i use 4 24 G GPU for this project fine tuning
Hi,
Thank you for your work. I am tyring use your wrok on my custom dataset. Can you guide whether the resource of 4GPU x 24G is engouh?
Regards
Hello,
Have you finetuned the model on your custom dataset? Could you please tell me how to train on the basis of the pretrained model？ #159
I succeeded on 2 24g 4090
@mycfhs 我的设备和您一致，请问您有没有复现指南呢～礼貌求（动手能力有点点差刚起步hhh），非常感谢！
I succeeded on 2 24g 4090
@mycfhs 我的设备和您一致，请问您有没有复现指南呢～礼貌求（动手能力有点点差刚起步hhh），非常感谢！
没有指南 有具体的问题可以说一下
方便留个email吗？我整理一下发给您
@mycfhs hello, may i ask that how to infer after train my own model,just use chat.py according to the guidance?
Yes.
发自我的iPhone
------------------ Original ------------------ From: Connor Ye @.> Date: Wed,Apr 16,2025 10:14 PM To: dvlab-research/LISA @.> Cc: YiCheng Yang @.>, Mention @.> Subject: Re: [dvlab-research/LISA] Can i use 4 24 G GPU for this project finetuning (Issue #147)
@mycfhs hello, may i ask that how to infer after train my own model,just use chat.py according to the guidance?
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***> connorye left a comment (dvlab-research/LISA#147)
@mycfhs hello, may i ask that how to infer after train my own model,just use chat.py according to the guidance?
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>
@mycfhs Thanks for the provided information.And I encounter the problem that "LisaModel" object has no attribute 'text_hidden_fcs' after running the chat.py.Did u have the same problem before?
我成功了 2 个 24g 4090
Hi, can you show your training hyperparameter settings?Thank you！
我成功了 2 个 24g 4090
Hi, can you show your training hyperparameter settings?Thank you！
我遇到的问题主要是微调过的模型推理有bug 我把我的训练推理代码放在这里了 https://github.com/mycfhs/lisa-finetune
我成功了 2 个 24g 4090
Hi, can you show your training hyperparameter settings?Thank you！
我遇到的问题主要是微调过的模型推理有bug 我把我的训练推理代码放在这里了 https://github.com/mycfhs/lisa-finetune
如果我想基于他的lisa模型进行微调是不是直接把train_ds.py的version改为lisa模型即可？
试试呗 我也不确定
发自我的iPhone
------------------ Original ------------------ From: fwbc @.> Date: Wed,Jun 25,2025 5:39 PM To: dvlab-research/LISA @.> Cc: YiCheng Yang @.>, Mention @.> Subject: Re: [dvlab-research/LISA] Can i use 4 24 G GPU for this project finetuning (Issue #147)
fwbc left a comment (dvlab-research/LISA#147)
我成功了 2 个 24g 4090
Hi, can you show your training hyperparameter settings?Thank you！
我遇到的问题主要是微调过的模型推理有bug 我把我的训练推理代码放在这里了 https://github.com/mycfhs/lisa-finetune
如果我想基于他的lisa模型进行微调是不是直接把train_ds.py的version改为lisa模型即可？
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>
你用的是我的代码吗？我代码记得把这个bug改了 你去issue搜一下有解决方法
发自我的iPhone
------------------ Original ------------------ From: fwbc @.> Date: Wed,Jun 25,2025 5:42 PM To: dvlab-research/LISA @.> Cc: YiCheng Yang @.>, Mention @.> Subject: Re: [dvlab-research/LISA] Can i use 4 24 G GPU for this project finetuning (Issue #147)
fwbc left a comment (dvlab-research/LISA#147)
试试呗 我也不确定
发自我的iPhone …
'LISAForCausalLM' object has no attribute 'ce_loss_weight'AttributeError: 'LISAForCausalLM' object has no attribute 'ce_loss_weight' 我试了但是报错如下，有点困惑
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>
你用的是我的代码吗？我代码记得把这个bug改了 你去issue搜一下有解决方法 发自我的iPhone …
sorry，我解决了，thanks
请问你在自己数据集上跑通训练了吗？能否交流一下🥹谢谢
具体是什么问题呢？
发自我的iPhone
------------------ Original ------------------ From: Feng Jiang @.> Date: Thu,Jun 26,2025 6:51 PM To: dvlab-research/LISA @.> Cc: YiCheng Yang @.>, Mention @.> Subject: Re: [dvlab-research/LISA] Can i use 4 24 G GPU for this project finetuning (Issue #147)
tumble3eed left a comment (dvlab-research/LISA#147)
你用的是我的代码吗？我代码记得把这个bug改了 你去issue搜一下有解决方法 发自我的iPhone …
sorry，我解决了，thanks
请问你在自己数据集上跑通训练了吗？能否交流一下🥹谢谢
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>