[Question] Issue with Model Type Mismatch and Download Stuck at 0%

[zxdscsfm]

Question

I encountered the following issue:
Youare using a model of type llama to instantiate a model of type llava_llama. This is not supported for all configurations of models and can yield errors.
Additionally, the download progress is stuck at 0%, and it continuously shows:
Downloading shards: 0%| | 0/2 [00:00<?, ?it/s]
Steps I took:
I set the mirror endpoint in the pretrain.sh script:
export HF_ENDPOINT=https://hf-mirror.com
The model I’m using is lmsys/vicuna-7b-v1.5, and I am attempting to train it.
However, the system is telling me that I am using a llama model to instantiate a llava_llama model, and the download is stuck at 0%.
My Questions:
Why is there a model type mismatch between llama and llava_llama?
What kind of model should I use to ensure compatibility with llava_llama?
Is there an official recommended model or conversion process I should follow to proceed with training?
If anyone has encountered similar issues or knows of any solutions, I would really appreciate your help. Thank you!

[mondalanindya]

Does anyone have a fix for this?