Yu-won Lee comments

Results 230 comments of


                                            Yu-won Lee

when would you show how to fintune the Qwen2.5-VL model?

I've made a code for finetuning Qwen2.5-VL You could use this https://github.com/2U1/Qwen2-VL-Finetune

Finetuning only vision model

Yes 1. For finetuninig only vision module, you could set the `freeze_llm` to true in the `finetune.sh` 2. Also, for the LoRA, you could add the llms layer in the...

@strawhatboy `No label_names provided for model class `PeftModel`. Since `PeftModel` hides base models input arguments, if label_names is not given, label_names can't be set automatically within `Trainer`. Note that empty...

Finetuning only vision model

@strawhatboy I've tested but it works fine for me. I'm not sure what is the reason. I'll test a bit more.

Finetuning only vision model

-9 usually occurs when system memory oom. I think erasing the `model = model.to(training_args.device)` would be better.

The training and inference results are not consistent.

How is it different?

The training and inference results are not consistent.

That's a bit odd. If you are inferncing with the other version of transformers, it may be the issue. I'll update the code for the latest one soon. Also will...

The training and inference results are not consistent.

@black-tea01 I couldn't find the issue for in consistant answering. Sorry for the late response.

AttributeError: 'DeepSpeedCPUAdam' object has no attribute 'ds_opt_adam'

@wangzhiyuan-pixel https://github.com/deepspeedai/DeepSpeed/issues/5659#issuecomment-2234427361 maybe this would work.

Would there any official instructions to fintune this model?

It's not an official one but, if you want some easy use of a finetuning code you could use this. https://github.com/2U1/Qwen2-VL-Finetune Most of case the code would work, but in...