https://github.com/kyegomez/simba
A simpler Pytorch + Zeta Implementation of the paper: "SiMBA: Simplified Mamba-based Architecture for Vision and Multivariate Time series"
https://github.com/kyegomez/simba
agents imagenet llms multi-modal recurrent-networks ss4 ssm transformers
Last synced: about 1 month ago
JSON representation
A simpler Pytorch + Zeta Implementation of the paper: "SiMBA: Simplified Mamba-based Architecture for Vision and Multivariate Time series"
- Host: GitHub
- URL: https://github.com/kyegomez/simba
- Owner: kyegomez
- License: mit
- Created: 2024-03-25T18:18:21.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2024-11-11T03:09:46.000Z (11 months ago)
- Last Synced: 2025-07-13T18:12:11.079Z (3 months ago)
- Topics: agents, imagenet, llms, multi-modal, recurrent-networks, ss4, ssm, transformers
- Language: Python
- Homepage: https://discord.gg/7VckQVxvKk
- Size: 2.48 MB
- Stars: 28
- Watchers: 2
- Forks: 2
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- Funding: .github/FUNDING.yml
- License: LICENSE
Awesome Lists containing this project
README
# Simba
A simpler Pytorch + Zeta Implementation of the paper: "SiMBA: Simplified Mamba-based Architecture for Vision and Multivariate Time series"
## install
`$ pip install simba-torch`
## usage
```python
import torch
from simba_torch.main import Simba
# Forward pass with images
img = torch.randn(1, 3, 224, 224)
# Create model
model = Simba(
dim = 4, # Dimension of the transformer
dropout = 0.1, # Dropout rate for regularization
d_state=64, # Dimension of the transformer state
d_conv=64, # Dimension of the convolutional layers
num_classes=64, # Number of output classes
depth=8, # Number of transformer layers
patch_size=16, # Size of the image patches
image_size=224, # Size of the input image
channels=3, # Number of input channels
# use_pos_emb=True # If you want
)
# Forward pass
out = model(img)
print(out.shape)
```
## Train
Dependencies: download and extract the datasets through wget
`wget http://images.cocodataset.org/zips/train2017.zip -O coco_train2017.zip`
`wget http://images.cocodataset.org/zips/val2017.zip -O coco_val2017.zip`
`wget http://images.cocodataset.org/annotations/annotations_trainval2017.zip -O coco_ann2017.zip`
Then run the following script:
`python3 train.py`
# License
MIT
# Todo
- [ ] Add paper link
- [ ] Add citation bibtex
- [ ] cleanup