Best version for fine tuning.
#5
by
psych0v0yager
- opened
Which version of Guanaco is ideal for fine-tuning?
The blogpost for the paper linked the following colab
https://colab.research.google.com/drive/1VoYNfYDKcKRQRor98Zbf2-9VQTtGJ24k?usp=sharing
In the colab you load a full precision model and convert to 4 bit.
Do you need to load the full precision guacano, convert to 4 bit, then fine tune? Or is there a way to directly finetune the 4 bit model without needing to load the full 16bit precision model.
"The blogpost for the paper" --> can you please add the link to this blogpost?
