erfanzar / EasyDeL

Accelerate your training with this open-source library. Optimize performance with streamlined training and serving options with JAX. 🚀
https://easydel.readthedocs.io/en/latest/
Apache License 2.0
168 stars 19 forks source link

Pull From Beta Branch (Adding Beta Version of `RLHF` , Fixing falcon bugs and ...) #18

Closed erfanzar closed 9 months ago

erfanzar commented 9 months ago

changes in this version

  1. Adding Beta Version of RLHF
  2. Fixing Falcon And Mpt Weight loading bugs
  3. adding Falcon and Mpt From pretrained to load directly from PyTorch
  4. fixing wrong examples for Training Falcon and Mpt models

Attention 🧬

The RLHF Process is not fully supported for right now, and it's still under heavy development but im trying my best to make it works as good as possible but for now only these models are supported for RLHF

  1. Llama
  2. Llama2
  3. Mpt
  4. Falcon
  5. All Other Models supported in library can be used, but they are not tested yet

and im trying to make a tutorial for that to how to use RLHF in order to train your models