Skip to content

Conversation

copybara-service[bot]
Copy link
Contributor

Remove f32-qc4w and f32-qc8w

These are unused, and have two serious issues:

  • They expect uint4, not int4.
  • They add the bias before quantization scaling instead of after (inconsistent with the rest of the gemms)

Other work is uncovering and requiring workarounds for these issues, we should just remove these until we can fix them.

These are unused, and have two serious issues:
- They expect uint4, not int4.
- They add the bias before quantization scaling instead of after (inconsistent with the rest of the gemms)

Other work is uncovering and requiring workarounds for these issues, we should just remove these until we can fix them.

PiperOrigin-RevId: 781160003
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

1 participant