Tiny and Smaller pretrained models!

Hi Adrian, @1adrianb
First, The face_alignment package and model is great. It is very helpful for me to learn facial landmarks detection! But this big model is too slow. I'm very interested the tiny and smaller model. Can you share the tiny or small model?
Thanks.

The issue requests smaller, faster pre-trained models for the face-alignment library. Research would start by examining the existing model architectures in the repository, understanding the training pipeline, and exploring techniques for model compression or distillation. 'Done' would be providing new model files or training scripts that produce a validated, smaller model with comparable accuracy.

索引モデルが issue の本文から書いたものです。