Closed zelenooki87 closed 3 months ago
Okay, so I looked into this arch, and I kinda don't want to touch it due to licensing issues.
FeMaSR was published under CC-BY-NC-SA and AdaCode under MIT. MIT is not compatible with CC-BY-NC-SA, so MIT just doesn't work as license for a derivative work of FeMaSR, AFAIK.
I'll make an issue to talk to them about it.
Can you add an option for successful conversion of the AdaCode model to ONNX format for faster inference? I manage to convert from PyTorch to ONNX with FeMaSR, while for AdaCode it is unfeasible. Thank you.
Are you getting an error? We don't control whether models can be converted to onnx, that's entirely on the pytorch/onnx side
I don’t get any error, but after 7-8 minutes I gave up waiting for the model to convert. FeMaSR converts in a few minutes. I’ll try again, I’ll give it a chance for a longer time to convert, maybe that’s the cause.
I apologize, it was just that. It just needed to wait a little longer. Is there a chance to convert the bsr_grl_base.safetensors model to ONNX? ( https://drive.google.com/file/d/1JdzeTFiBVSia7PmSvr5VduwDdLnirxAG/view?usp=sharing ) I get an error when converting it: Error An unexpected error occurred: Error: The application encountered an unexpected error and could not continue. Thanks
Thanks to you developers for all your hard work and a job well done. I'd love to see the adacode architecture in one of the next builds. This project is based on the femasr architecture, so I don't think there will be any problems integrating it as well. Thank you very much! https://github.com/kechunl/AdaCode