Junyang Lin
Junyang Lin
Never did this before, but I'll try to save it with torch jit script later. I wonder where the error comes from, would you mind telling me more details, say...
Seemingly the problem comes from here... https://github.com/pytorch/pytorch/issues/70544
Good suggestion! I'll add the huggingface model links to the checkpoint.md, and give more directions in the README
Sorry for missing your information (my bad...) Let me transfer the ckpts to HF models
> Thanks for adding these, I see that you've uploaded the finetuned checkpoints for the Base size  > > Do you plan to upload the fine-tuned checkpoints [for the...
You mean Huge ckpt for HF Transformers? Yes, and btw we are going to update our code for transformers, probably next week.
The checkpoint of OFA-huge for transformers has already been uploaded to [huggingface.co/ofa-sys](https://huggingface.co/OFA-Sys). Also, we updated the code for transformers, check the branch feature/add_transformers and transformers.md for more details.
> Thank you for updating the HF code. With "git clone https://huggingface.co/OFA-Sys/OFA-huge" I get asked for authentication request (username / password) and the code is not publically visible on the...
This might be different from CLIP. CLIP is a two-tower model, and thus it is easy to extract a feature vector for image and text respectively, while OFA is essentially...
Sorry for the late response. 1. Yes, `encoder_out` is the output states of the encoder, and you can use `.shape` to check its shape, and use things like `torch.mean()` to...