Hi,

I converted Anole-HF (https://huggingface.co/leloy/Anole-7b-v0.1-hf) to Anole-torch using bin_to_pth.py script.

I observe that the {q/k}_norm.weight, {q/k}_norm.bias, self_attn.{q/k}_proj.weight params do not match between the converted Anole-torch (from HF) and the original Anole-torch (Anole-7b-v0.1) shared as a part of this repo.

I believe that this is due to some special operations (permute and t()) that convert_chameleon_weights_to_hf.py performs on the Torch checkpoint to give the HF checkpoint, which the bin_to_pth.py does not revert.

Can you fix the bin_to_pth file for accurate conversion?