docs: clarify configuration for fractional loss weights - #10188
Open
Excelius-Wang wants to merge 1 commit into
Open
Excelius-Wang wants to merge 1 commit into
Excelius-Wang wants to merge 1 commit into
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The custom dataset documentation only tells users to disable binary loss scaling for weights greater than 1, omitting fractional weights such as 0.5. Clarify the requirement for non-binary positive weights in both languages, explain what binary mode preserves, and update the existing JSONL/Python examples to show downweighting alongside upweighting.
Validation: extracted bilingual JSONL and Python examples match; their 0.5 and 2.0 weights survive encoding and padded/padding-free collation using a cached Qwen2.5 tokenizer. The documented Qwen3-8B download was not exercised. Earlier one-step CPU SFT checks on the same upstream revision confirm that default 0.5 and 1.0 weights produce identical loss and gradients, while explicitly disabling binary scaling preserves fractional weighting.
pre-commit run --all-files: all 10 hooks passed.git diff --check: passed. Only the two documentation files changed.