Skip to content

Conversation

@ChenZiHong-Gavin
Copy link
Collaborator

This PR replaces the old yes/no entropy loss with a synonym-aware version:

  1. Adds exhaustive English + Chinese synonym lists for “yes” and “no”.
  2. Introduces _normalize_yes_no() that pools probabilities of all synonyms and re-normalizes them to a clean {yes: p, no: 1-p} distribution.
  3. Updates yes_no_loss_entropy() to use this normalized distribution instead of looking only at the top token, making the loss robust to wording variations.

@gemini-code-assist
Copy link

Important

Installation incomplete: to start using Gemini Code Assist, please ask the organization owner(s) to visit the Gemini Code Assist Admin Console and sign the Terms of Services.

@ChenZiHong-Gavin ChenZiHong-Gavin merged commit 26803a8 into main Oct 27, 2025
3 checks passed
@ChenZiHong-Gavin ChenZiHong-Gavin deleted the fix/prob-normalization branch October 27, 2025 08:22
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.

2 participants