| ---
|
| license: cc-by-4.0
|
| tags:
|
| - adafactor
|
| - gelu
|
| - grouped-query
|
| - kaiming-normal
|
| - matching
|
| - polynomial
|
| - scalenorm
|
| - small
|
| - tiny-transformer
|
| - tucker
|
| ---
|
|
|
| # predict.py
|
|
|
| ## Model Overview
|
|
|
| A **small**-scale implementation of the **tiny transformer** architecture, built for **matching** tasks.
|
|
|
| ## Architecture
|
|
|
| - **Architecture**: tiny transformer
|
| - **Scale**: small
|
| - **Attention**: grouped query
|
| - **Fusion strategy**: tucker
|
| - **Task head**: matching
|
| - **Activation**: gelu
|
| - **Normalization**: scalenorm
|
| - **Initialization**: kaiming normal
|
|
|
| ## Training
|
|
|
| - **Optimizer**: adafactor
|
| - **LR scheduler**: polynomial
|
|
|
| ## Files
|
|
|
| - `predict.py` — main artifact of this repository
|
|
|
| ## License
|
|
|
| See the license field above.
|
|
|
