Release of model weights

[kris-singh]

Hi,

Thanks for open-sourcing the code for MetaQueries. This is really helpful.
Are there plans to release the weights of the models?
Also, if one tries to reproduce the training using 8 A100 80GB GPUs with 25M images from the (CC12M, JourneyDB, and SA-1B), how long should one expect the training to take?

Thanks,
Krish Singh

[xichenpan]

Hi @kris-singh, thank you for your interest! Unfortunately, the model’s release may still be some way off. Let's stay tuned for the progress. With the LlavaOV 0.5B setup, training 25 million images for a single epoch takes about one day on your device.

[SaraGhazanfari]

Hi, Thanks for the nice work!

If anyone performs the training, please consider sharing the resulting weights with us. I'd be more than happy to cite your contribution in the paper I'm currently working on :)

Thanks,
Sara

[eternaldolphin]

Hi @kris-singh, thank you for your interest! Unfortunately, the model’s release may still be some way off. Let's stay tuned for the progress. With the LlavaOV 0.5B setup, training 25 million images for a single epoch takes about one day on your device.

For a 7B qwen + 7B DiT model, how long does it take to train until it can make good pictures, not random mess

[zhangquanwei962]

For a 7B qwen + 7B DiT model, how long does it take to train until it can make good pictures, not random mess

Hi, do you know the answer?

[wxr99]

Hi,

Thanks again for sharing the code and for your detailed previous replies — they’ve been very helpful.

I wanted to kindly ask if there have been any updates on the potential release schedule of the model weights, or if there’s any plan to share partial checkpoints or smaller versions in the meantime.