search

TheLastBen / fast-stable-diffusion

fast-stable-diffusion + DreamBooth
MIT License
7.5k stars 1.31k forks source link

Freeze Textencoder and train unet + embedding at the same time? #1426

Open DarkAlchy opened 1 year ago

DarkAlchy commented 1 year ago

image This is exactly what I am after so I wonder if this could be implemented in the colab?

Thank you.

TheLastBen commented 1 year ago

Training an embedding requires VRAM so does the unet, so I don't see how it would require less resources.

DarkAlchy commented 1 year ago

Training an embedding requires VRAM so does the unet, so I don't see how it would require less resources.

We just @ you on the discord about this which would probably be easier since I am in the dark about this knowing your next question may be "how"? Did not know you were on there so I said hello (just so you know that is me).

78Alpha commented 1 year ago

Would that be similar to what LORA does? With pivotal tuning?

https://github.com/cloneofsimo/lora