model_664118749_mae_huge.py

Model Overview

A huge-scale implementation of the mae architecture, built for generation tasks.

Architecture

• Architecture: mae
• Scale: huge
• Attention: dilated
• Fusion strategy: tensor fusion
• Task head: generation
• Activation: approx gelu
• Normalization: scalenorm
• Initialization: kaiming normal

Training

• Optimizer: lion
• LR scheduler: cosine

Files

• model_664118749_mae_huge.py — main artifact of this repository

License

See the license field above.