Skip to main content

OneMoreEpoch

PyPI Python versions CI License: MIT

A deep learning framework you can actually read. Built in Python for everyday use, with a Rust engine underneath for speed, a real automatic differentiation system, and an optional Hindi/Hinglish personality that roasts you when your code breaks.

It's not trying to replace PyTorch or TensorFlow. It's built to be understood — every layer of it — while still being genuinely usable for training real models.

pip install onemoreepoch

What this actually is

Training a neural network means: run some numbers through a model, see how wrong the answer was, and nudge every internal number a tiny bit to be less wrong next time. Do that thousands of times and the model gets good at its job.

OneMoreEpoch is the machinery that makes that loop work, split into pieces that each do one job well:

  • Tensors are the basic unit of data here — think of them as spreadsheets of numbers (a photo, a batch of text, a set of measurements) that the framework can do math on quickly.
  • Autograd (automatic differentiation) is the part that automatically figures out which direction to nudge every number in the model so it gets better, without a human having to work out that math by hand for every new model shape.
  • Layers and optimizers are the reusable building blocks (like "a fully connected layer" or "the Adam update rule") you snap together to build and train a model, instead of writing the math from scratch every time.
  • A Python front door, a Rust engine room. You write plain Python. Under the hood, the actual number-crunching can run through a compiled Rust engine instead of Python — same results, much faster, because Rust runs closer to the metal than Python does.
  • A personality system. When something goes wrong (and in machine learning, something always goes wrong), the error message can talk to you in plain professional English, Hinglish ("bhai, ye shapes ka rishta nahi ho sakta"), or a savage-but-friendly roast — your choice.

None of this needs you to understand Rust, or even much Python, to use it — those details are exactly what the framework hides from you.


Installation

pip install onemoreepoch

or with uv:

uv add onemoreepoch

OneMoreEpoch's faster Rust backend ships as a compiled extension. When a pre-built wheel exists for your platform and Python version, that's all pip needs — no Rust, no compiler, nothing else to install. If pip has to build from source instead (an uncommon platform, for example), you'll need a Rust toolchain on your machine; either way, the default NumPy backend always works with no extra setup.

Quick start

from onemoreepoch import nn
from onemoreepoch.core import Tensor
from onemoreepoch.optim import Adam

model = nn.Sequential(nn.Linear(10, 32), nn.ReLU(), nn.Linear(32, 1))
optimizer = Adam(model.parameters())

x = Tensor.randn(32, 10)
target = Tensor.randn(32, 1)

for _ in range(100):
    optimizer.zero_grad()
    loss = nn.MSELoss()(model(x), target)
    loss.backward()
    optimizer.step()

Want the personality system? One line:

from onemoreepoch import config
config.set_message_mode("hindi")  # or "roast" — "classic" is the default

How the pieces fit together

Your code (Python)
        │
        ▼
 Tensor / nn / optim      ← the public API you actually write
        │
        ▼
   Autograd engine        ← figures out how to improve the model
        │
        ▼
  Backend (pluggable)     ← does the actual math
        │
   ┌────┴────┐
   ▼         ▼
 NumPy      Rust           ← two interchangeable "engines"; pick
(default)  (faster)          whichever one is available/faster

The important idea is the backend: the framework never does math directly, it always asks "whichever backend is active" to do it. That's what let a whole Rust engine get added without changing a single line of the training code, layers, or optimizers above it — and it's the same seam GPU support will plug into next (see Roadmap below).

The personality system, and why it's not a gimmick

Error messages in most tools are dry and unhelpful. Here, every error is backed by a real, structured, machine-readable exception underneath — the personality layer only changes the words shown to you, never what actually happened or how the program behaves. So bhai, ye shapes ka rishta nahi ho sakta and ShapeError: matrix dimensions must agree are the exact same error, just spoken differently. A joke can never hide a real problem.


Project layout

OneMoreEpoch/
├── src/onemoreepoch/
│   ├── core/            Tensor, Module, Parameter, Device, and backend/
│   │   └── backend/         the pluggable math engines (NumPy, Rust, ...)
│   ├── autograd/         the "figure out how to improve" engine
│   ├── nn/               layers (Linear, Conv2D, Dropout, BatchNorm, ...)
│   ├── optim/             training rules (SGD, Adam, AdamW, RMSProp, AdaGrad)
│   ├── data/              loading and batching datasets
│   ├── metrics/           accuracy, precision, recall, and friends
│   └── messages/          classic / hindi / roast personalities + memes
├── rust/onemoreepoch-core/   the compiled speed engine (PyO3 + Rust)
├── tools/meme_updater/       local pipeline for adding new personality lines
├── tests/                    unit, integration, and personality-system tests
└── examples/                  runnable end-to-end scripts

Roadmap: what's next — GPU acceleration

The next piece of work is a GPU backend, so training can run on a graphics card instead of just the CPU (typically a 10–100x speed-up for larger models).

This is the reason the backend system was built to be pluggable in the first place: adding the Rust engine didn't require touching Tensor, nn, or optim at all — it was purely a new engine registered alongside the NumPy one. GPU support is planned to slot in the exact same way, which two things fall out of directly:

  • Where it plugs in: src/onemoreepoch/core/backend/ — a new backend implementation alongside numpy_backend.py and rust_backend.py.
  • What it needs to solve that's genuinely new: unlike swapping NumPy for Rust, a GPU also needs a scheduler — something deciding which computations run on the GPU, when data moves between CPU and GPU memory (that trip is slow, so minimizing it matters), and how multiple operations get batched together so the GPU stays busy instead of waiting around for the CPU to hand it work one step at a time.

This section is intentionally short — it's a marker for the next phase of work, not a spec written in advance of actually doing it.


Development

Contributing or just poking around the source (rather than installing the published package)?

git clone https://github.com/Akash4467/OneMoreEpoch.git
cd OneMoreEpoch
uv sync --group dev      # installs dependencies and builds the Rust engine
uv run pytest              # 200+ tests, should all pass
cargo test --no-default-features --manifest-path rust/onemoreepoch-core/Cargo.toml

This needs a Rust toolchain regardless of platform, since you're building the extension yourself rather than installing a pre-built wheel.

License

MIT — see LICENSE.

Metadata

Release files for onemoreepoch 0.1.2

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for onemoreepoch 0.1.2
File Size Uploaded Provenance Actions
onemoreepoch-0.1.2.tar.gz 50.0 kB With provenance Details for onemoreepoch-0.1.2.tar.gz

Built distributions (wheels)

Table of built distributions (wheels) for onemoreepoch 0.1.2
File Interpreter ABI Platform Actions
onemoreepoch-0.1.2-cp312-cp312-win_amd64.whl CPython 3.12 CPython 3.12 Windows x86-64 Details for onemoreepoch-0.1.2-cp312-cp312-win_amd64.whl
onemoreepoch-0.1.2-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl CPython 3.12 CPython 3.12 Linux glibc 2.17+ x86-64 Details for onemoreepoch-0.1.2-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
onemoreepoch-0.1.2-cp312-cp312-macosx_10_12_x86_64.macosx_11_0_arm64.macosx_10_12_universal2.whl CPython 3.12 CPython 3.12 macOS 10.12+ x86-64, macOS 11.0+ ARM64, macOS 10.12+ universal2 (ARM64, x86-64) Details for onemoreepoch-0.1.2-cp312-cp312-macosx_10_12_x86_64.macosx_11_0_arm64.macosx_10_12_universal2.whl

Total release size: 1.3 MB

Release files / onemoreepoch-0.1.2.tar.gz

Download URL onemoreepoch-0.1.2.tar.gz
Size 50.0 kB
Tags Source
SHA-256 checksum
How to use checksums
84edfd4b78d3eea2b7dec625a3734e14fd43a806623b411ccf1bef3d9a8c43a2
BLAKE2b-256 checksum
How to use checksums
d0a70412fb46d227077e3eff14e454d7aac5877e11bec210ef9f4eee69797bb9
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Aug 25, 2026.

Transparency log

Release files / onemoreepoch-0.1.2-cp312-cp312-win_amd64.whl

Download URL onemoreepoch-0.1.2-cp312-cp312-win_amd64.whl
Size 248.5 kB
Tags CPython 3.12 Windows x86-64
SHA-256 checksum
How to use checksums
48fa009fe4b85c1a504ac6f891af0c2f73d753707ab879033b48d806ae42c8d8
BLAKE2b-256 checksum
How to use checksums
1b350b4ae142fbb1bdc92415eaf92ef0645f40cc9b451e446c7c1091b6a472c5
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Aug 25, 2026.

Transparency log

Release files / onemoreepoch-0.1.2-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl

Download URL onemoreepoch-0.1.2-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
Size 388.2 kB
Tags CPython 3.12 Linux glibc 2.17+ x86-64
SHA-256 checksum
How to use checksums
a46d8169b9905827087eec32d25f9eac7b76ac43e9a7d4ef9587f45aa178f306
BLAKE2b-256 checksum
How to use checksums
0148bea41833fd38375d0a3a40fdfabffba1819737a0cc98f19cc0acbea3ea26
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Aug 25, 2026.

Transparency log

Release files / onemoreepoch-0.1.2-cp312-cp312-macosx_10_12_x86_64.macosx_11_0_arm64.macosx_10_12_universal2.whl

Download URL onemoreepoch-0.1.2-cp312-cp312-macosx_10_12_x86_64.macosx_11_0_arm64.macosx_10_12_universal2.whl
Size 612.9 kB
Tags CPython 3.12 macOS 10.12+ universal2 (ARM64, x86-64) macOS 10.12+ x86-64 macOS 11.0+ ARM64
SHA-256 checksum
How to use checksums
9bc15e47fb46bb65b23862540e36903c2b0db980552733a0ed4a05f1484c5fcb
BLAKE2b-256 checksum
How to use checksums
625af1cf852c41b5a69e8f0ef8b7bc36249108c4e9dd4e97aae6967f2c6d1eaa
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Aug 25, 2026.

Transparency log

Security for onemoreepoch 0.1.2

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

All 4 files in this release have been uploaded with provenance metadata.

Select from 4 files

Attestation for onemoreepoch-0.1.2.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Aug 25, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Release history Release notifications | RSS feed

This release

0.1.2 This release

4 release files Full provenance

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page