model_217140930_clip_huge.py

Model Overview

A huge-scale implementation of the clip architecture, built for multitask tasks.

Architecture

• Architecture: clip
• Scale: huge
• Attention: multi query
• Fusion strategy: low rank
• Task head: multitask
• Activation: swish
• Normalization: instancenorm
• Initialization: xavier uniform

Training

• Optimizer: adafactor
• LR scheduler: onecycle

Files

• model_217140930_clip_huge.py — main artifact of this repository

License

See the license field above.