OpenImagingLab / OpenImagingLab/FlashVSR
Model fine-tuning with lora
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 1.9k
- Forks
- 152
- PR merge metrics
- No merged PRs in 30d
Description
Thank you for your amazing work. However, we have identified some limitations in certain high-frequency specific scenarios. Therefore, we attempted further fine-tuning (LoRA, scenario-specific datasets) on the basis of your model. Yet, we also found that the top-k Mask of Block-Sparse-Attention used in the inference code has no gradients. We would like to consult you: how did you implement this in Step 2 of the training?
Contributor guide
No contributing guide indexed for this repository
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
Start by locating the inference implementation of the top-k Mask of Block-Sparse-Attention and the Step 2 training path. Compare how gradients are handled in those paths for LoRA fine-tuning. The issue does not define a concrete completion condition, so maintainer clarification is needed.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- machine-learning
- Issue type
- Bug
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 20/100