KO | EN

GitLite

Minimalist Open Source Repository Search Engine

gitlite — search

#typescript #ai-agents #developer-tools #claude-code #codex #cli #agent-skills #ai #dsh-plugin #mcp #windows #agent

Needle

★ 30 Open GitHub ↗

Imperative deep learning framework with customized GPU and CPU backend

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Needle

An imperative deep learning framework with **customized GPU and CPU backend**.

## TODO

- [ ] To refactor the codebase<br>
  Needle adopts a monolithic design, e.g., the NDArray backend for a device is in a single file. A modular design allows agile optimization of operators, layers, modules, etc.
- [ ] To extend the reservoir of operators<br>
  To support self-attention, the [*batched matrix multiplication*](https://pytorch.org/docs/stable/generated/torch.bmm.html) operator is necessary. `BatchNorm2d` can be more efficient if **fused**. Other **customized** operators include **fused** [differentiable volume rendering](https://yconquesty.github.io/blog/ml/nerf/nerf_rendering.html), sparse matrix multiplication for graph embeddings, I/O kernels, etc.
- [ ] To optimize the NDArray backend<br>
  [This summary](./backend.md) gathers a series of blog posts on maximizing the throughput of operators. Also refer to [*Programming Massively Parallel Processors*](https://www.elsevier.com/books/programming-massively-parallel-processors/hwu/978-0-323-91231-0) for more topics on CUDA. The goal is to **exceed the performance of official CUDA libraris** like [cuBLAS](https://developer.nvidia.com/cublas) with hand-crafted kernels in certain tasks.
- [ ] To incorporate [`tcnn`](https://github.com/NVlabs/tiny-cuda-nn) as MLP intrinsics
- [ ] To accelerate computational graph traversal with CUDA[^gunrock][^graphblast]

## Acknowledgement

This project is inspired by [10-414/714](https://dlsyscourse.org) *Deep Learning Systems* by [Carnegie Mellon University](https://www.cmu.edu). Switch to the branch [`hw`](https://github.com/YconquestY/Needle/tree/hw) for homework, [`proj`](https://github.com/YconquestY/Needle/tree/proj) for the course project, and [`lec`](https://github.com/YconquestY/Needle/tree/lec) for lectures.

[^gunrock]: [*Gunrock: a high-performance graph processing library on the GPU*](https://dl.acm.org/doi/abs/10.1145/2851141.2851145)

[^graphblast]: [*GraphBLAST: A High-Performance Linear Algebra-based Graph Framework on the GPU*](https://dl.acm.org/doi/full/10.1145/3466795)

Needle

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README