potatoplm 0.2.0

A pure Python LLM library implemented with NumPy (Llama-style architecture).

PotatoPLM

PotatoPLM is a Pure Python LLM library implemented from scratch using NumPy. It aims to be a transparent and educational implementation of the Transformer architecture.

Features

• Pure NumPy-based Transformer implementation.
• Llama-style architecture:
  • RMSNorm for normalization.
  • RoPE (Rotary Positional Embeddings).
  • SwiGLU activation function.
• Minimal dependencies (only numpy).
• Python 3.10 compatible.
• GPT-style Decoder-only architecture.

Installation & Setup

We recommend using a virtual environment:

# Create and activate virtual environment
python3 -m venv venv
source venv/bin/activate

# Install the package in editable mode
pip install -e .

Usage Example

Run the toy inference example:

python3 examples/toy_inference.py