Skip to content
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

Feature Requests #342

Open
KintCark opened this issue Aug 13, 2024 · 9 comments
Open

Feature Requests #342

KintCark opened this issue Aug 13, 2024 · 9 comments

Comments

@KintCark
Copy link

KintCark commented Aug 13, 2024

Can u add the ability to save the qauntized models to storage that way if we want to use it again we don't have to keep reconverting every time we start an generation.

@grauho
Copy link
Contributor

grauho commented Aug 14, 2024

You should already be able to do that by setting the "-M, --mode" argument to "convert" to save the model as a quantized gguf file at the location specified with "-o, --output".

More information in the docs:
docs/quantization_and_gguf.md

@KintCark
Copy link
Author

It auto saves to the main folder but I tried qauntized aura flow buy it killed when saving output

@grauho
Copy link
Contributor

grauho commented Aug 17, 2024

Interesting, please recompile with --config Debug, re-run with --verbose, and post the output

@KintCark
Copy link
Author

Interesting, please recompile with --config Debug, re-run with --verbose, and post the output

I can't post output termux crashes immediately I got 7gb ram to spare but it loads the tensors but as soon as it trys to save output it crashes. Could someone else try converting aura flow 0.3 please I need q5 or q4 and q8

@grauho
Copy link
Contributor

grauho commented Aug 18, 2024

I'm not familiar with aura flow but I wonder if it's based on a model type that sdcpp doesn't currently support. You could pipe the output to a file to avoid losing it when termux crashes, eg: "./sd etc etc &> foo.txt" to pipe both stdout and stderr to a file.

@KintCark
Copy link
Author

KintCark commented Aug 18, 2024

I'm not familiar with aura flow but I wonder if it's based on a model type that sdcpp doesn't currently support. You could pipe the output to a file to avoid losing it when termux crashes, eg: "./sd etc etc &> foo.txt" to pipe both stdout and stderr to a file.

You're right that has to be what it is it's not supported so it don't work so I can quantize sd3 I bet. Will flux and aura flow be added soon?

@KintCark
Copy link
Author

Gguf used less memory I can run flux q8_0 and t5xxlfp16 in comfyui on my phone so using gguf is better than safetensors

@KintCark
Copy link
Author

Sd3 uses flow so how come flux and aura not auto support?

@grauho
Copy link
Contributor

grauho commented Aug 18, 2024

Sd3 uses flow so how come flux and aura not auto support?

From my understanding of the similarities between Flux and SD3 is that while they share some components, they are not identical in their architecture. So, that's why despite SD3 being available in sdcpp there is no "auto support" for Flux.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants