thu-ml / thu-ml/TurboDiffusion

Lora fine tuning

Open
#72 2 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
3.7k
Forks
277
Avg merge
2h 57m
Merged PRs (30d)
2

Description

Hi guys. Very nice work. I would like to know if I have to fine-tune a lora for Wan2.1. What are the correct steps? Can my LORA work on top of these TurboDiffusion weights? I suspect it won't because it was trained on the normal attention, not Spare Linear Attention. What's the best way to make my lora work with turbo diffusion?

  1. Merge my LORA weights into the original Wan2.1 base weights and then treat it as a new base model and train turbo diffusion?
  2. Or should I explore something like orthogonal finetuning?

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

The issue asks whether LoRA fine-tuning for Wan2.1 is compatible with TurboDiffusion weights and discusses merging weights or using orthogonal fine-tuning. No files, tests, or entry points are identified; first seek maintainer guidance on the intended approach and what validation would demonstrate compatibility.

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

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.