Lightning-AI / Lightning-AI/litgpt
Adding Direct Preference Optimization (DPO) support
Open
Nobody has claimed this yet.
enhancement
- Dominant language
- Python
- Stars
- 13.7k
- Forks
- 1.5k
- Avg merge
- 15h 37m
- Merged PRs (30d)
- 1
Description
Has anyone implemented Direct Preference Optimization (DPO) or is there a plan to add it to lit-gpt?
If not, any advice on the best way to architect it within lit-gpt?
There’s another issue for RLHF support too, that this could be related to for post-SFT fine tuning.
Example DPO implementations in other frameworks:
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
The issue names no repository files, entry points, or tests. Start by reviewing the linked RLHF issue and the referenced DPO implementations to identify how DPO should fit into lit-gpt. Done would require an agreed architecture and a concrete implementation scope, which this issue does not yet define.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- machine-learning
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 20/100