Metadata-Version: 2.4
Name: DepthML
Version: 0.2.0
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: numpy>=2.3.1
Requires-Dist: depthtensor>=2.0.2
Dynamic: license-file

# DepthML: A Deep Learning Framework

DepthML is a high-performance, Pythonic deep learning framework, which provides a hybrid structure (**Keras-like** lazy initialization combined with **PyTorch-like** imperative programming), while delegating all low-level tensor calculus to the [DepthTensor](http) backend.

## Features

- Hybrid Object-Oriented API: Incorporates design paradigms present in both Keras and PyTorch.

- Hardware Agnostic: Easily switches between the CPU (numpy) and the GPU (cupy) via the DepthTensor backend.

## Benchmarks

DepthML is capable of training standard architectures.

**Task**: MNIST Digit Classifcation (60k samples)

**Architecture**: 3-Layer MLP (784 -> 32 -> 10)

**Hardware**: NVIDIA GPU

| Metric         | Result |
| -------------- | ------ |
| Final Accuracy | 93.86% |
| Training Time (5 epochs) | ~45 seconds |
| Average Step Time | 7.4 ms |
| Convergence | < 2 epochs |

This framework achieves within 3x the step-latency of optimized C++ frameworks for small-scale MLPs.

## Installation

```bash
pip install depthml
```

