Wrong LoRA Weights?

[Pompey21]

Hi, I have tried navigating your model via the instructions you have provided but with no success. After completing the set-up, I tried running a simple Text-to-Text inference but with no success. The model outputs repetitive word tokens that do not make sense (image below). After testing the weights for Vicuna and LLaMA separately and realising that this is not the problem I proceeded with checking the code-base. The way I achieved for the model to perform the basic Text-to-Text task was by disabling the LORA weights all together. However, after that I was of course not able to trigger any other modality...

[WenjiaWang0312]

Hi, this is my result. I cannot generate reasonable results as well. Have you solved it?

[Pompey21]

Hi @WenjiaWang0312 , if you want to resolve this issue you must disable the LoRA weights. However, that will only allow you do to text-to-text and not the cross-modality