🧪autoresearch
AI agents running research on single-GPU nanochat training automatically
⭐ Hugely popular: 94k stars, gaining about 563 a week
View on GitHub ↗repo profile
momentum
durability
bus factor = how many people it takes to cover more than half the commits (6 months). 1 is a solo project; higher means the work is spread across a team. top-author share is the single busiest author's slice of those commits.
since we covered it
why it's a big deal
- It turns model research into an agent-native loop, where the LLM is not just helping with code but actually running bounded empirical experiments end-to-end.
- The design is unusually legible: one GPU, one editable file, one metric, and a fixed five-minute wall-clock budget make the whole system easy to reason about.
- It points to a broader shift in OSS AI tooling from static copilots toward closed-loop systems that can test, compare, and improve themselves.
under the hood
- The repo revolves around three files: prepare.py for fixed setup and utilities, train.py as the only mutable file, and program.md as the human-written instruction layer for the agent.
- Every experiment runs for exactly five minutes and is measured on val_bpb, a vocab-size-independent validation metric designed to keep comparisons fair across architectural changes.
- The codebase is deliberately self-contained and minimal, built around a simplified single-GPU nanochat training stack with no distributed setup or large config machinery.
our take from PR#30, 2026-03-25
star history
- PR#30 43k 2026-03-25
- now 94k + 51k since first covered
curve is sampled from GitHub's star history, plus our own daily readings since we covered it; the dashed stretch is before we first covered it, the solid line since. figures at coverage are the numbers we printed then (approx.), current count is live.
understory
Better known than its recent output, coasting a little on attention.
- output, commits & releases
- clout, star velocity
output = commits & releases; clout = star velocity, both 0 to 100 monthly indices; the gap where output runs above clout is the understory. The understory →
covered in
-
Autonomous overnight experimentation for single-GPU model training
similar projects
compare these →- 🖥️ Open WebUI
User-friendly AI Interface (Supports Ollama, OpenAI API,...)
149k ACTIVE - 🦙 LLaMA-Factory
Unified Efficient Fine-Tuning of 100+ LLMs & VLMs (ACL 2024)
74k ACTIVE - 💻 nanoGPT
The simplest, fastest repository for training/finetuning medium-sized GPTs.
62k ACTIVE
comments
Sign in with GitHub to add your blip on autoresearch.