Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Deep learning framework realized by Numpy purely, supports for both Dynamic Graph and Static Graph with GPU acceleration
| Date | Stars |
|---|---|
| 2026-07-31 | 250 |
| 2026-08-01 | 250 |
| 2026-08-06 | 250 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# XShinnosuke : Deep Learning Framework
<div align=center>
<img src="https://timgsa.baidu.com/timg?image&quality=80&size=b9999_10000&sec=1597579280045&di=409d33924532df749524161e4c11f8b3&imgtype=0&src=http%3A%2F%2Fb-ssl.duitang.com%2Fuploads%2Fitem%2F201607%2F30%2F20160730144641_4UMvr.thumb.700_0.jpeg" width="300px" height="400px">
</div>
## Descriptions
XShinnosuke(short as **XS**) is a high-level neural network framework which supports for both **Dynamic Graph** and **Static Graph**, and has almost the same API to **Keras** and **Pytorch** with *slightly differences*. **It was written by Python only**, and dedicated to realize experimentations quickly.
Here are some features of XS:
1. Based on **Cupy**(GPU version)/**Numpy** and **native** to Python.
2. **Without** any other **3rd-party** deep learning library.
3. **Keras and Pytorch style API**, easy to start up.
4. Supports commonly used layers such as: **Dense, Conv2D, MaxPooling2D, LSTM, SimpleRNN, etc**, and commonly used function: **conv2d, max_pool2d, relu, etc**.
5. **Sequential** in Pytorch and Keras, **Model** in Keras and **Module** in Pytorch, **all of them are supported** by XS.
6. Training and inference supports for both **dynamic graph** and **static graph**.
7. **Autograd** is supported .
XS is compatible with: **Python 3.x (3.7 is recommended)** [==> C++ version](https://github.com/eLeVeNnN/xshinnosuke_cpp)
[1. API docs](https://elevennn.github.io/xshinnosuke/) [2. Notebook](https://github.com/eLeVeNnN/xshinnosuke/blob/master/demos/examples/XShinnosuke-API.ipynb)
## Getting started
#### Compared with Pytorch and Keras
| ResNet18(5 Epochs, 32 Batch_size) | XS_static_graph(cpu) | XS_dynamic_graph(cpu) | Pytorch(cpu) | Keras(cpu) |
| --------------------------------- | -------------------- | --------------------- | ----------------- | -------------- |
| Speed(Ratio - seconds) | *1x* - *65.05* | *0.98x* - 66.33 | **2.67x** - 24.39 | *1.8x* - 35.97 |
| Memory(Ratio - GB) | *1x* - *0.47* | **0.47x**- 0.22 | *0.55x* - 0.26 | *0.96x* - 0.45 |
| ResNet18(5 Epochs, 32 Batch_size) | XS_static_graph(gpu) | XS_dynamic_graph(gpu) | Pytorch(gpu) | Keras(gpu) |
| --------------------------------- | -------------------- | --------------------- | ---------------- | -------------- |
| Speed(Ratio - seconds) | *1x* - *9.64* | *1.02x* - 9.45 | **3.47x** - 2.78 | *1.07x* - 9.04 |
| Memory(Ratio - GB) | **1x** - *0.48* | *1.02x* - 0.49 | *4.4x* - 2.11 | *4.21x* - 2.02 |
**XS holds the best memory usage!**
---
### 1. Static Graph
The core networks of XS is a model, which provide a way to combine layers. There are two model types: **Sequential** (a linear stack of layers) and **Functional** (build a graph for layers).
For **Sequential** model:
```python
from xs.nn.models import Sequential
model = Sequential()
```
Using `.add()` to connect layers:
```python
from xs.layers import Dense
model.add(Dense(out_features=500, activation='relu', input_shape=(784, ))) # must be specify input_shape if current layer is the first layer of model
model.add(Dense(out_features=10))
```
Once you have constructed your model, you should configure it with `.compile()` before training or inference:
```python
model.compile(loss='cross_entropy', optimizer='sgd')
```
If your labels are `one-hot` encoded vectors/matrix, you shall specify loss as *sparse_crossentropy*, otherwise use *crossentropy* instead.
Use `print(model)` to see details of model:
```python
***************************************************************************
Layer(type) Output Shape Param Connected to
###########################################################################
dense0 (Dense) (None, 500) 392500
---------------------------------------------------------------------------
deExcerpt of 9,476 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:ee0437786ab0831a, llm:Description, README: 'Deep learning framework realized by Numpy purely... supports Dynamic Graph and Static Graph... Keras and Pytorch style API, layers, autograd, cupy/numpy', topics: cupy, deeplearning, keras-style, numpy, pytorch-style
matched fp:ee0437786ab0831a, llm:Description, README: 'Deep learning framework realized by Numpy purely... supports Dynamic Graph and Static Graph... Keras and Pytorch style API, layers, autograd, cupy/numpy', topics: cupy, deeplearning, keras-style, numpy, pytorch-style