Question on Training Config for UniTok MLLM Pre-training

[hyomin14]

Thank you for your excellent work on UniTok.

I am attempting to reproduce the UniTok MLLM pre-training process as outlined in the configuration files (Issue #20 ).

The configuration specifies the use of a pre-trained multimodal projector checkpoint for initial checkpoint of multi_embedder.mm_projector:
pretrained_mmprojector: Unitok/llama2_mm_projector.bin

However, I am unable to locate this specific file (llama2_mm_projector.bin) either in the repository or through the provided model download links on Hugging Face.

I have two main questions:

1. Where can I find the llama2_mm_projector.bin checkpoint file?
2. Could you briefly detail how this initial weight was trained?

Thank you for your time and assistance.