Get started¶

Install¶

To install Kernl library, you just have to pip install it:

python3 -m pip install install 'git+https://github.com/ELS-RD/kernl'

Optimize a model¶

Then, in your progam, you have to import the optimization function and apply it to your model:

from transformers import AutoModel
from kernl.model_optimization import optimize_model

model = AutoModel.from_pretrained(model_name).eval().cuda()
optimize_model(model)

That's it, you have your model with Kernl's optimizations !

Beware, Kernl works only on Ampere GPU and with python 3.9.* for now.

Look at the repository README for more informations.