PKU-Alignment
[Question] Does PKU-Alignment/Align-DS-V utilize the LLF technique?
Required prerequisites
- [x] I have read the documentation https://align-anything.readthedocs.io.
- [x] I have searched the Issue Tracker and Discussions that this hasn't already been reported. (+1 or comment there if it has.)
- [ ] Consider asking first in a Discussion.
Questions
Has the technical report for Align-DS-V not been released yet? I'm interested in knowing whether the model utilizes the LLF technique.
This repository serves as the official implementation of the paper Align Anything: Training All-Modality Models to Follow Instructions with Language Feedback. Beyond LLF itself, align-anything (this repo) can be used for post-training alignment across various modalities. We provide the relevant code and implementations for this purpose.
Considering resource and time constraints, the currently open-sourced align-ds-v version does not yet incorporate LLF technology. However, our paper has already demonstrated the effectiveness of LLF on the TI2T modality, where employing just 25% of LLF achieves the same effect as 100% binary feedback.
The align-ds-v model series is undergoing continuous updates. We plan to integrate LLF into the training process in future iterations and further refine the models.
