Skip to content

Bitnet: directly use scale instead of inverting it twice #13026

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 1 commit into
base: master
Choose a base branch
from
+3 −5
Open

Bitnet: directly use scale instead of inverting it twice #13026

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 3 additions & 5 deletions convert_hf_to_gguf.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2036,11 +2036,9 @@ def weight_quant(self, weight: Tensor) -> Tensor:
dtype = weight.dtype
weight = weight.float()
scale = weight.abs().mean().clamp(min=1e-5)
iscale = 1 / scale
# TODO: multiply by the scale directly instead of inverting it twice
# (this is also unnecessarily doubly inverted upstream)
# ref: https://huggingface.co/1bitLLM/bitnet_b1_58-3B/blob/af89e318d78a70802061246bf037199d2fb97020/utils_quant.py#L10
result = (weight * iscale).round().clamp(-1, 1) / iscale
# Directly use scale instead of inverting it twice
# First round and clamp to -1, 1, then multiply by scale to get back to original range
result = weight.div(scale).round().clamp(-1, 1).mul(scale)
return result.type(dtype)

def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
Expand Down