#178178th of 242 in ML frameworks and libraries
handwired
Compiles plain JavaScript into an exact, differentiable matmul-and-ReGLU network
- Stars
- 2.2k
- License
- MIT
- Last commit
- Nov 2026
- Language
- TypeScript
1,689 on 17 Oct 20262,160 on 14 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
Handwired rewrites JavaScript control flow (if-else, loops, state) as matrix multiplications plus ReGLU over floats, with hand-written weights and no training. The resulting network runs on plain JS or WebGPU and stays differentiable, so weights can be unfrozen and trained, or a frozen program attached to a generic neural net. It is an experimental project with live demos, including a JSON search over 100k documents.
Who it is for: Researchers and tinkerers exploring differentiable programs and WebGPU
Strengths
- Program output is exact, with weights written by hand rather than trained
- Only two primitives (matmul, ReGLU), so GPU fusion and sparsity optimizations apply
- Runs on JS or WebGPU; JSON search demo covers 100k documents in milliseconds
- Differentiable end to end, so frozen programs can combine with trainable nets
Weaknesses
- Experimental; README describes use cases as fun to try, not production
- Both branches of every conditional are computed, which adds work
- No releases published; the supported JavaScript subset is not documented in the README
- Setup requires Bun and cloning the repo; no package install shown
What it needs
- GPU optional
- Needs Bun
Also in ML frameworks and libraries
See all 242| Rank | Project | Score |
|---|---|---|
| 1 | transformersModel-definition library for pretrained text, vision, audio and multimodal models | 94 out of 100 |
| 2 | ultralyticsPython library and CLI for training and running YOLO vision models | 94 out of 100 |
| 3 | pytorch-image-modelsPyTorch library of image models, pretrained weights and training scripts | 94 out of 100 |
| #4 | pytorchPython tensor library with GPU support and tape-based autograd | 93 out of 100 |
| #5 | supervisionPython toolkit for loading, annotating and counting computer vision detections | 92 out of 100 |