| ---
|
| license: mit
|
| tags:
|
| - approx-gelu
|
| - bilinear
|
| - dilated
|
| - exponential
|
| - huge
|
| - layernorm
|
| - mobilevit
|
| - retrieval
|
| - sgd
|
| - xavier
|
| ---
|
|
|
| # predict.py
|
|
|
| ## Model Overview
|
|
|
| A **huge**-scale implementation of the **mobilevit** architecture, built for **retrieval** tasks.
|
|
|
| ## Architecture
|
|
|
| - **Architecture**: mobilevit
|
| - **Scale**: huge
|
| - **Attention**: dilated
|
| - **Fusion strategy**: bilinear
|
| - **Task head**: retrieval
|
| - **Activation**: approx gelu
|
| - **Normalization**: layernorm
|
| - **Initialization**: xavier
|
|
|
| ## Training
|
|
|
| - **Optimizer**: sgd
|
| - **LR scheduler**: exponential
|
|
|
| ## Files
|
|
|
| - `predict.py` — main artifact of this repository
|
|
|
| ## License
|
|
|
| See the license field above.
|
|
|
