model_695057586_dino_xlarge.py

Model Overview

A xlarge-scale implementation of the dino architecture, built for multitask tasks.

Architecture

• Architecture: dino
• Scale: xlarge
• Attention: multi query
• Fusion strategy: cross attention
• Task head: multitask
• Activation: gelu tanh
• Normalization: instancenorm
• Initialization: orthogonal

Training

• Optimizer: adafactor
• LR scheduler: onecycle

Files

• model_695057586_dino_xlarge.py — main artifact of this repository

License

See the license field above.