Vishaal Udandarao

Ahh just stumbled upon this where the authors responded: https://github.com/microsoft/LoRA/issues/114

Hi @mmaaz60, would you please be able to share your working inference script? It would be really helpful, thanks!

Were you able to solve this issue? I am running into somewhat of a similar issue where I notice that my CPU memory usage increases but is nothing drastic (way...

update: this was because of the itertools.cycle function leading to memory leaks (https://github.com/pytorch/pytorch/issues/23900). Without the cycle function everything works smoothly. This was not a webdataset issue.

Hey @LinB203 did this resolve your issue? I also ran into the same issue where my CLIP training jobs get killed because of OOM, did you try reducing the buffer...

Another question: For fine-tuning from Otter, would you still recommend using the same training pipeline, or would recommend using PEFT methods (I guess you have implemented LoRA support already), do...

Thanks @Luodian , I was finally able to run it on a larger compute node with 4 40GB GPUs with a batch size of 32, but it still was strange...

Thanks for the response @ZhangYuanhan-AI, could you also describe the LAConv and LADD splits?

In my initial experiments, it seems that the format ```python PROMPT = ' Q: Describe the image. A: This is an image of a {}.' ``` is slightly better for...

Hey, unfortunately it seems I have deleted the pickle files and am unable to track them down. Hopefully, the instructions in the report are complete enough for reproducing the features....