Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 3 additions & 3 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -7,7 +7,7 @@
![Tests](https://img.shields.io/github/actions/workflow/status/Terminay/LeanPass/summary.yml?label=tests)
![CodeQL](https://img.shields.io/github/actions/workflow/status/Terminay/LeanPass/codeql.yml?label=CodeQL)
![PyPI Publish](https://img.shields.io/github/actions/workflow/status/Terminay/LeanPass/publish-pypi.yml?label=PyPI%20publish)
![Size](https://img.shields.io/badge/size-~118KB-lightgrey)
![Size](https://img.shields.io/badge/size-~8.5KB-lightgrey)
![Real Downloads](https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/Terminay/LeanPass/main/badge.json)
[![Stars](https://img.shields.io/github/stars/Terminay/LeanPass?style=social)](https://github.com/Terminay/LeanPass/stargazers)

Expand Down Expand Up @@ -155,13 +155,13 @@ LeanPass is designed to be minimal. Here is how it compares to other frameworks:

| Library | Package Size | Lines of Code | Dependencies | Install Time |
|---------|-------------|---------------|-------------|--------------|
| **LeanPass** | **~118 KB** | **~730** | **1 (NumPy)** | **~2 seconds** |
| **LeanPass** | **~8.5 KB** | **~730** | **1 (NumPy)** | **<1 second** |
| Micrograd | ~15 KB | ~150 | 0 | ~1 second |
| PyTorch | ~800+ MB | millions | many (CUDA, etc.) | ~minutes |
| TensorFlow | ~1+ GB | millions | many (CUDA, etc.) | ~minutes |
| JAX | ~200+ MB | millions | several | ~minutes |

LeanPass is **~7000x smaller** than PyTorch and **~9000x smaller** than TensorFlow, while still providing the core autodiff, neural network layers, and optimizers needed for small-scale experiments and learning.
LeanPass is **~14,000x smaller** than PyTorch and **~40,000x smaller** than TensorFlow, while still providing the core autodiff, neural network layers, and optimizers needed for small-scale experiments and learning.

## Examples

Expand Down
2 changes: 1 addition & 1 deletion src/pages/index.js
Original file line number Diff line number Diff line change
Expand Up @@ -58,7 +58,7 @@ export default function Home() {
<em>that</em> they flow, you need source code that fits in your head.
</p>
<p>
This is that library. ~500 lines of pure NumPy. Every
This is that library. ~730 lines of pure NumPy. Every
<code>backward()</code> call traces through the graph step by step.
No magic, no abstraction layers, just vector calculus you can read
line by line.
Expand Down
Loading