When will release the training code?

[junwenxiong]

Hi, Author

Thank you for doing such great work! I wanna ask whether the training code will be released?

[wcy1122]

Hi, thanks for your interest. If everything goes well, the training code will be released in two to three weeks. Btw, the training code is very similar to Lyra, you may refer to that repo.

[junwenxiong]

OK, i'm trying the mgm/serve/web_demo.py, but i can't find the "mm_speech_tower": "wcy1122/Dual-Qwen2A-7B-Encoder". Where i can download it?

[wcy1122]

1. You can directly run the code and download the encoder to hf root.
2. You can download https://huggingface.co/wcy1122/Qwen2A-7B-Encoder, and rename the folder to "Dual-Qwen2A-7B-Encoder"

[junwenxiong]

Thanks. I have another question, how to get qwen2.5-VL-3B-Vit weights. I see, you already provided the 7B vit weights, but 7B model is too large to train on my server. I wanna try the 3B model.

[wcy1122]

Hi, you can find the 3B ViT model here https://huggingface.co/wcy1122/Qwen2.5-VL-3B-ViT