Any way to load transformers.CsmForConditionalGeneration weights into csm.Model

[jeffDebug]

If I use the HF transformer API (https://huggingface.co/sesame/csm-1b#fine-tuning--training-%F0%9F%93%89) to finetune the csm-1b model. After I saved the weights of model type transformers.CsmForConditionalGeneration, how can I load the trained model weights back into csm.Model, as defined here in this repo? I'm looking for a simple interface or API to achieve this. Thanks!

[ZackHodari]

The HF implementation is different to csm.Model. There may be code in transformer to convert between the checkpoints (as that's how they went from our impl to their new one)

I'm not sure what your goal is in coming back to use code in this repo if you're setup with the transformer implementation, you should be able to do inference in transformers. If you need to use our generator.py with the HF implementation of CSM, this might require some code changes to generator.