erogol/BlaGPT

A library for mechanistic interpretability of GPT-style language models

Turning LLM architecture exploration into reusable knowledge for the community. 🌳

Stable Looped Models and their Scaling Laws

A 124.6M LLM trained from scratch on 13B tokens on a single RTX 4090 — tokenizer, pretraining, post-training, evaluation, and reproducible inference.

The fastest way to build and start training your own LLM. CLI tool that scaffolds production-ready PyTorch training projects in seconds. Like create-next-app but for language models.

An LLM playground you can run on your laptop

Code for BLT research paper

Build LLM from scratch