https://github.com/sdgunaa/mojonet

A deep learning framework in mojo🔥
https://github.com/sdgunaa/mojonet

deep-learning mojo neural-network tensor

Last synced: 5 months ago
JSON representation

A deep learning framework in mojo🔥

• Host: GitHub
• URL: https://github.com/sdgunaa/mojonet
• Owner: sdgunaa
• License: apache-2.0
• Created: 2023-09-23T19:43:29.000Z (almost 3 years ago)
• Default Branch: nightly
• Last Pushed: 2025-12-01T22:25:58.000Z (7 months ago)
• Last Synced: 2026-01-25T00:56:13.132Z (5 months ago)
• Topics: deep-learning, mojo, neural-network, tensor
• Language: Mojo
• Homepage:
• Size: 66.1 MB
• Stars: 5
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          


  



> > ⚠️ WARNING

> Mojonet is under development and the API is subject to change.

Mojonet is a ML/DL framework written in Mojo🔥.

### Latest Nightly

Use at your own risk and be patient!

---

## 🚧 Development Roadmap / TODO

### 🔹 Core Memory Infrastructure

- [x] **UniquePointer** — move-only smart pointer (ownership-safe)

- [x] **DataPointer** — device-aware smart pointer (deleter-managed)

- [x] **Allocator Registry & GlobalContext**

- [x] **Pluggable Allocator interface**

- [x] **Caching Allocators** (CPU / CUDA)  

- [x] **Memory Pools & Block Reuse**

- [x] **Storage & StorageImpl (ref-counted)**

### 🔹 Tensor System

- [ ] **Tensor API (basic constructors)**

- [ ] **Tensor metadata: shape, strides, dtype**

- [ ] **Layout system support (Strided first)**

- [ ] **Tensor views / slicing / reshape / expand**

- [ ] **Automatic broadcasting rules**

### 🔹 Device & Backend Abstraction

- [ ] **CPU backend**

- [ ] **CUDA backend**

- [ ] **Pluggable backend interface**

- [ ] **Optional integration with MAX (Modular Accelerated Compute)**

### 🔹 Kernels & Operations

- [ ] **Core math ops (add, mul, matmul, etc.)**

- [ ] **Reduction ops (sum, mean, …)**

- [ ] **Elementwise + universal functions**

- [ ] **BLAS integration (cuBLAS, etc.)**

### 🔹 Autograd & Training

- [ ] **Autograd engine**

- [ ] **Gradient tape & backward ops**

- [ ] **Optimizers (SGD, Adam, …)**

### 🔹 Distributed / Advanced

- [ ] **Multi-device support**

- [ ] **JIT / lazy execution**

- [ ] **Graph optimization**

---

> This project is experimental — contributions and feedback are welcome!  

> Stay tuned for updates as Mojonet evolves into a high-performance ML research stack 🌟