metadata
license: mit
tags:
- approx-gelu
- classification
- clip
- co-attention
- cosine
- grouped-query
- groupnorm
- lamb
- xavier-uniform
- xlarge
main.py
Model Overview
A xlarge-scale implementation of the clip architecture, built for classification tasks.
Architecture
- Architecture: clip
- Scale: xlarge
- Attention: grouped query
- Fusion strategy: co attention
- Task head: classification
- Activation: approx gelu
- Normalization: groupnorm
- Initialization: xavier uniform
Training
- Optimizer: lamb
- LR scheduler: cosine
Files
main.py— main artifact of this repository
License
See the license field above.