| ---
|
| license: cc-by-4.0
|
| tags:
|
| - adamw
|
| - approx-gelu
|
| - batchnorm
|
| - bilinear
|
| - deit
|
| - flash
|
| - generation
|
| - huge
|
| - kaiming
|
| - step
|
| ---
|
|
|
| # model_587404722_deit_huge.py
|
|
|
| ## Model Overview
|
|
|
| A **huge**-scale implementation of the **deit** architecture, built for **generation** tasks.
|
|
|
| ## Architecture
|
|
|
| - **Architecture**: deit
|
| - **Scale**: huge
|
| - **Attention**: flash
|
| - **Fusion strategy**: bilinear
|
| - **Task head**: generation
|
| - **Activation**: approx gelu
|
| - **Normalization**: batchnorm
|
| - **Initialization**: kaiming
|
|
|
| ## Training
|
|
|
| - **Optimizer**: adamw
|
| - **LR scheduler**: step
|
|
|
| ## Files
|
|
|
| - `model_587404722_deit_huge.py` — main artifact of this repository
|
|
|
| ## License
|
|
|
| See the license field above.
|
|
|
