| ---
|
| license: bsd-3-clause
|
| tags:
|
| - adafactor
|
| - base
|
| - dino
|
| - flash
|
| - gelu
|
| - kaiming
|
| - multitask
|
| - polynomial
|
| - scalenorm
|
| - tucker
|
| ---
|
|
|
| # train.py
|
|
|
| ## Model Overview
|
|
|
| A **base**-scale implementation of the **dino** architecture, built for **multitask** tasks.
|
|
|
| ## Architecture
|
|
|
| - **Architecture**: dino
|
| - **Scale**: base
|
| - **Attention**: flash
|
| - **Fusion strategy**: tucker
|
| - **Task head**: multitask
|
| - **Activation**: gelu
|
| - **Normalization**: scalenorm
|
| - **Initialization**: kaiming
|
|
|
| ## Training
|
|
|
| - **Optimizer**: adafactor
|
| - **LR scheduler**: polynomial
|
|
|
| ## Files
|
|
|
| - `train.py` — main artifact of this repository
|
|
|
| ## License
|
|
|
| See the license field above.
|
|
|
