[sft] support DFT #5355
[sft] support DFT #5355
Conversation
There was a problem hiding this comment.
Summary of Changes
Hello @hjh0119, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!
This pull request introduces support for a new loss function, referred to as "DFT loss," into the supervised fine-tuning (SFT) framework. This enhancement allows users to optionally enable this specific loss calculation during model training, potentially improving training dynamics or performance as suggested by the linked research paper.
Highlights
- DFT Loss Integration: Implemented the "DFT loss" calculation within the per_token_loss_func, allowing it to be applied conditionally.
- Configuration Option: Added a new boolean argument --enable_dft_loss to the training arguments, providing a clear way to activate or deactivate this feature.
- Trainer Adaptation: Updated the training pipeline to correctly pass the enable_dft_loss flag to the loss computation, ensuring the new loss is applied when configured.
- Example Script: Provided a new example shell script (dft.sh) demonstrating how to run a full SFT training job with the DFT loss enabled.
Using Gemini Code Assist
The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.
Invoking Gemini
You can request assistance from Gemini at any point in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.
| Feature | Command | Description |
|---|---|---|
| Code Review | /gemini review |
Performs a code review for the current pull request in its current state. |
| Pull Request Summary | /gemini summary |
Provides a summary of the current pull request in its current state. |
| Comment | @gemini-code-assist | Responds in comments when explicitly tagged, both in issue comments and review comments. |
| Help | /gemini help |
Displays a list of available commands. |
Customization
To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.
Limitations & Feedback
Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.
You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.
Footnotes
-
Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩
There was a problem hiding this comment.
Code Review
This pull request introduces support for Direct Preference Tuning (DFT) loss, a technique to align language models with preferences without a separate reward model. The changes include adding a new training argument enable_dft_loss, implementing the DFT loss calculation in swift/plugin/loss.py, and integrating it into the trainer logic in swift/trainers/trainers.py. An example script dft.sh is also added to demonstrate its usage.
My review focuses on the implementation of the DFT loss. I've identified a performance optimization opportunity in the loss calculation logic. The rest of the changes look good and correctly integrate the new feature.
PS: The link to the paper in the PR description seems to have a typo in the year; it should likely be 2408.05629 instead of 2508.05629.
Experiments with DFT vs SFTtraining script: dft.sh Training curves(DFT: red, SFT: green) Evaluation on MATH500Script: https://github.com/yongliang-wu/DFT
|
|
then why i get much worse performace (acc) in llama_factory where use_dft_loss: true |
https://arxiv.org/abs/2508.05629
#5338