Xiangyu Qi

Hey, Based on the log, it seems that you are computing gradients for those tensors that do not set the require grad flags. Could you provide more information about how...

Hi all, Could you try this checkpoint? https://huggingface.co/liuhaotian/llava-llama-2-13b-chat-lightning-preview ------ I checked the llava repository, llava-1.5 was released on Oct 5, which was after the publication of our paper. So, it...

Thank you, Rylan, for helping answer questions & the continuous development of the codebase along this thread!

Hi, when doing adversarial examples attacks, model weights won't be changed. The attacks are conducted by only making perturbations to the inputs.

Yes, if you directly use the pre-trained model from the official repo, the harmfulness score will be high. This is because the visual-instruction tuning will cause the original safety of...

Hi. Thank you for the question. For the optimization process, empirically we found that directly using the harmful corpus without the question part seems easier to optimize than using the...

Hi, if you use .bmp, the image file just saves exactly what the image originally was. If you use .jpg, the image will be going through a lossy compression, and...

Hi, this is a commonly used trick back to https://arxiv.org/abs/1412.6572 My experience is that gradient sign descent usually gives better results, less likely to be stuck at a local optimal.

Hi, the hyperparameters looked good to me. Could you try: 1. Increase the batch_size = 16? Perhaps this will make the optimization more stable. 2. Just try multiple times to...