-
Notifications
You must be signed in to change notification settings - Fork 5.3k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[ONNX] Is it possible to load LORAs to be used in the ONNX pipelines? #2493
Comments
AFAIK, nothing has changed in that you'd have to write your own code to "load" LoRAs into an ONNX pipeline, and that would be a massive headache prone to breaking, even to support exclusively the default diffusers style LoRAs. There are "solutions" to convert LoRAs from other formats to diffusers format, then inject into a model before converting to ONNX, but most people would give up before figuring it out, I'd think. |
Yap, that's exactly what I had to do with Textual Inversion embeddings. It worked but the resulting models would have uneven layer sizes and if I tried to convert an 'injected' model back to .ckpt and use it to merge with other models it would fail. Plus, embeddings should really only be loaded on demand - specially LORAs. |
This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread. Please note that issues that do not follow the contributing guidelines are likely to be ignored. |
This is possible, but only if the ONNX model's nodes and initializers have their original names. Running some of the ORT optimization scripts will rename them and makes it very difficult to line things up. I have some notes on the nodes, names, etc in ssube/onnx-web#213 |
This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread. Please note that issues that do not follow the contributing guidelines are likely to be ignored. |
A while ago I've asked about Textual Inversion embeddings and the answer was basically 'no'. There was a way around it but it wasn't an actual solution. However this was some time ago and things might have changed.
So, is it possible now? And if so is there any documentation on this matter I can read?
Sorry this is not an actual issue.
The text was updated successfully, but these errors were encountered: