RosettaCommons / RosettaCommons/RFdiffusion
[Bug?] Missing Padding Mask for Batch Mode
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 3.1k
- Forks
- 644
- PR merge metrics
- No merged PRs in 30d
Description
Hi team, thanks for the great work done for the community. I've learned a lot from reading the code. I have one question regarding the batch mode of RFDiffusion. The released inference code always uses batch_size=1 for generation, but the model actually has a batch dimension. So I assume that RFDiffusion is able to handle batch mode inference and training.
However, as I go deeply into the model implementation, I don't find anything like padding_mask that is required for attention and other computations in batch mode. Without this, the padded token will also be used for generation and representations, which I don't think is correct. Am I missing something here?
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 tracing the released inference code and the model implementation's batch, attention, and representation paths. Determine whether padded tokens affect batch-mode generation or representations; done means the padding-mask behavior is established and the issue's concern is addressed or clearly documented.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- machine-learning
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 35/100