deep-floyd / IF

Other
7.7k stars 502 forks source link

finetune #40

Open betterftr opened 1 year ago

betterftr commented 1 year ago

Finetuning code will be released as well? Awesome project btw! Cant wait to train a custom model

pigtigger commented 1 year ago

awesome project! looking forward to the finetuning code

darkman111a commented 1 year ago

Just repurpose the imagen trainer script. Until they release their paper, just use random search and figure the best hyperparameters to you. Currently trying to figure out myself, I'm beyond thirsty lol.

https://github.com/lucidrains/imagen-pytorch/blob/main/imagen_pytorch/trainer.py

baleksey commented 1 year ago

@darkman111a Have you made any progress with fine-tuning the IF model? If so could you please share it? I don't have enough knowledge to modify the Imagen code and would be very appreciated for your help!

alelordelo commented 1 year ago

+1

TheSilverBullets commented 1 year ago

@darkman111a hi, have you sucessfully fine-tuned the IF model? I would appreciate it if you could share your training configs.

oss-roettger commented 1 year ago

I have just published an Open Source version of Textual Inversion to incorporate your own objects, faces, logos or styles into DeepFloyd IF: https://github.com/oss-roettger/T5-Textual-Inversion The clear and short notebook runs within 24 GB CUDA memory. Feedback welcome! Kind regards Hans