fix: the source part should not participate in loss calculation in SFT stage - #762
Open
xffxff wants to merge 1 commit into
Open
fix: the source part should not participate in loss calculation in SFT stage#762xffxff wants to merge 1 commit into
xffxff wants to merge 1 commit into
Conversation
Copy link
Copy Markdown
Contributor
Sorry, something went wrong.
xffxff
requested review from
RezaYazdaniAminabadi,
ShadenSmith,
arashb,
awan-10,
conglongli,
duli2012,
eltonzheng,
jeffra,
minjiaz,
mrwyattii,
samyam,
tjruwase,
xiaoxiawu-microsoft and
yaozhewei
as code owners
October 10, 2023 11:14
Copy link
Copy Markdown
|
Sorry, something went wrong.
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
Reviewers
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
fix #660
In the SFT stage, it's essential that the source part doesn't contribute to the loss calculation, only the completion part should be considered. To address this issue, I've adjusted the labels for the source part to be set as -100. This specific value, -100, corresponds to the default "ignore index" in the torch.nn.CrossEntropyLoss function. Importantly, both OPT and LLAMA models utilize torch.nn.CrossEntropyLoss for their loss calculations, as seen in OPTForCausalLM and LLamaForCausalLM. As a result, there is no need to make any modifications to the way the loss is computed, as it will automatically handle the source part as intended.
The training loss of opt-350m with Dahoas/rm-static as dataset
