search

XuezheMax / megalodon

Reference implementation of Megalodon 7B model
MIT License
487 stars 50 forks source link

The project is very interesting, thanks for publishing it, now I have a few questions #8

Open NickyDark1 opened 2 weeks ago

NickyDark1 commented 2 weeks ago

How could I implement it with the huggingface models?

google colab for example?

thank you so much

XuezheMax commented 2 weeks ago

Thanks for your question. The PyTorch module for each Megalodon block is here: https://github.com/XuezheMax/megalodon/blob/53cbaa3a3b3e05ea439564bd67cb352012ba6b97/megalodon/model/mega.py#L43 This repo includes all implementation details of the model. However, since some components are implemented by CUDA, you need to also move CUDA code to HuggingFace.