model_290829634_coca_huge.py

Model Overview

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

Architecture

• Architecture: coca
• Scale: huge
• Attention: sliding window
• Fusion strategy: low rank
• Task head: matching
• Activation: swish
• Normalization: batchnorm
• Initialization: xavier uniform

Training

• Optimizer: sgd
• LR scheduler: polynomial

Files

• model_290829634_coca_huge.py — main artifact of this repository

License

See the license field above.