feat(reward): add language consistency reward function #683
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Outdated suggestions cannot be applied.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.
Description
This pull request introduces a new reward function,
lang_consistency
reward, to enhance the linguistic consistency of model outputs during reinforcement learning (e.g., GRPO).Motivation
In multilingual or even monolingual contexts, it's crucial for a model to adhere to the target language specified in a prompt or conversation. This reward function directly incentivizes this behavior by assigning a positive reward (1.0) only when the language of the generated completion matches the expected language. This helps prevent language drift and ensures more reliable and predictable model behavior.
Inspired by the "language consistency reward" concept from the DeepSeek-R1 paper, this function provides a valuable tool for fine-tuning models where strict language adherence is required.