zyandtom
zyandtom
have you solved yet?
also, i found that if deepspeed is enabled, the `max_grad_norm` set in transformers trainer will not be used...hard to know😭
同样的error，请问解决了吗
solved，这个是因为image_processor中生成的temporal_ids没有被输入，需要手动在dataset中生成，或者直接拿到image_processor的输出放进collator。生成过程参考https://github.com/hiyouga/LLaMA-Factory/blob/59f2bf1ea369ca91774b99e8d94a578657be6c7c/src/llamafactory/data/mm_plugin.py#L951
> > solved，这个是因为image_processor中生成的temporal_ids没有被输入，需要手动在dataset中生成，或者直接拿到image_processor的输出放进collator。生成过程参考https://github.com/hiyouga/LLaMA-Factory/blob/59f2bf1ea369ca91774b99e8d94a578657be6c7c/src/llamafactory/data/mm_plugin.py#L951 > > 大佬有没有具体的修改方案代码呀 [@zyandtom](https://github.com/zyandtom) 就在collator里加一个temporal_ids字段给none就好，或者你有需求输入的话参考https://huggingface.co/openbmb/MiniCPM-V-4_5 官方demo实现
> Do you guys notice the finetuned model already has full params for vision-model, then why we load it from siglip again? Does this overwrite the pretrained one? i have...
anybody know how to merge lora weights in the middle ckpts? looks like middle ckpts are not the same as final step
have you solved yet?
probably max_length is too small, GT will be truncated?