Great model! Any news regarding 1B-2B version?

#5
by hongkongbuffer - opened

Hey, I've tested the model, and I'm really impressed with its performance. It works exceptionally well for my use case!

I'm planning to fine-tune the model soon to further improve its performance, particularly for the Chinese language.

I also have a question: Are there any plans to release a smaller model in the 1B–2B parameter range?

Thanks for your great work!

HAL-X org

Hey! Thank you so much for your feedback! We're really glad to hear that the model is working well for your use case.

And yes, we're planning to release a smaller model in the 1B–2B parameter range in the near future.

Thanks for sharing this exciting news! Looking forward to the 1B–2B model.

I have a couple of follow-up questions:

Is there a more specific timeline or ETA for the release of this smaller model?
Regarding the current version's limit of up to 4 images per prompt—is this a hardcoded restriction, or is it simply because the training data contained a maximum of 4 images? Does the underlying architecture natively support multi-image inputs (more than 4)?

Sign up or log in to comment