4eJIoBek1

Strange low cpu usage when sampling on cpu

1

Sampling on CPU is possible but takes about ~5% of overall computations, while 90% is free and not using and idk why.

error when importing custom image to the stylegan latents while working on cpu

2

if you run gradio demo on cpu in windows 11 and import custom image, there will be an error: RuntimeError: Expected all tensors to be on the same device, but...

https://github.com/Anttwo/SuGaR Seems like a good thing to produce meshes from gaussians. I think exporting meshes from gaussians is really feature that i and other people are waiting for.

Just asking:Anybody did img2mesh process but with colmap/metashape on backend

1

Seems like if you create something like 200 renders of image via zero1-to-3-xl it can be possible to feed it into colmap or metashape to produce fine mesh, anybody did...

 This sampling keeps working even if examples are deleted from demo. What i am doing wrong?

Export matches to colmap or other

1

This is a great webui, but kinda useful. Is there/will be in future option to export matching data to colmap (or other similar programs) and batch processing option for using...

https://github.com/Beinabih/Unconditional-MusicGen-Trainer It is not perfect at all, but take a look onto this

Something strange with any non-english prompts with alpaca/gpt4all/vicuna

1

When you try to chat with these on not english language, the answer will hallucinate and on english, but if you use file as prompt, everything works fine(picrelated). This is...

feature request for lora training

1

is it possible to make a simple lora trainer in this gui?

can this one bed executed without flashattention?

3

flash-attention is implemented for only ampere and newer gpus, also it blocks code from using on cpu, which is not good