model_588750391_deit_huge.py

Model Overview

A huge-scale implementation of the deit architecture, built for matching tasks.

Architecture

• Architecture: deit
• Scale: huge
• Attention: grouped query
• Fusion strategy: cross attention
• Task head: matching
• Activation: gelu
• Normalization: rmsnorm
• Initialization: kaiming normal

Training

• Optimizer: novograd
• LR scheduler: constant warmup

Files

• model_588750391_deit_huge.py — main artifact of this repository

License

See the license field above.