HarutoKondo's picture
Upload 2 files
df2a9ba verified
|
Raw
History Blame Contribute Delete
791 Bytes
---
license: bsd-3-clause
tags:
- adafactor
- cosine
- cross-attention
- dilated
- gelu-tanh
- kaiming
- layernorm
- perceiver
- retrieval
- tiny
---
# model_083084092_perceiver_tiny.py
## Model Overview
A **tiny**-scale implementation of the **perceiver** architecture, built for **retrieval** tasks.
## Architecture
- **Architecture**: perceiver
- **Scale**: tiny
- **Attention**: dilated
- **Fusion strategy**: cross attention
- **Task head**: retrieval
- **Activation**: gelu tanh
- **Normalization**: layernorm
- **Initialization**: kaiming
## Training
- **Optimizer**: adafactor
- **LR scheduler**: cosine
## Files
- `model_083084092_perceiver_tiny.py` — main artifact of this repository
## License
See the license field above.