fix: chat_template masking due to truncation, consolidate turn build and keys within field #2123
Conversation
This file contains 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
…and keys within field
8 tasks
winglian
reviewed
Dec 9, 2024
winglian
approved these changes
Dec 9, 2024
djsaunde
pushed a commit
that referenced
this pull request
Dec 16, 2024
…and keys within field (#2123) [skip ci] * fix: chat_template masking due to truncation, consolidate turn build and keys within field * fix: revert roles change * fix: handling of training and training_detail * fix: do not skip setting eos mask even if failed finding turn boundary * fix: truncate reward modelling outputs
djsaunde
pushed a commit
that referenced
this pull request
Dec 17, 2024
…and keys within field (#2123) [skip ci] * fix: chat_template masking due to truncation, consolidate turn build and keys within field * fix: revert roles change * fix: handling of training and training_detail * fix: do not skip setting eos mask even if failed finding turn boundary * fix: truncate reward modelling outputs
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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 fixes cases where masking would fail for
chat_templateusing the new behavior.Changes:
truncationwhile applyingchat_templateas that would cut off conversation and affect masking downstream. We have another section that would drop long conversation already.- Removed defaultroleswithin the prompter to not automatically map roles (in case users intend to keep the original names and wasn't aware we mapped) <- Comment needed for this design choice!roles_to_trainif it exist withinroles. For example,roles_to_train: ["gpt"]but the convo maps it all theassistantleading to everything being ignored.turnacross modules and dropping off system prompt if done.should_trainTODO:
[ ] handlewe support nous tool calling dataset but not hf one for now.tool_callingfor assistant response for tools datasets[] discuss whether to revert removing defaulttemporarily revert for nowroles[ ] update chat_template doc for explicitly specifyingroles[ ] Fix tests to specifyrolesMotivation and Context
How has this been tested?
Test with Puffin dataset locally.
Screenshots (if appropriate)
Types of changes
Social Handles (Optional)