OpenEBFM
by datamllab
OpenEBFM is a Python codebase for training, fine-tuning and evaluating energy-based language models, a different design from the usual next-word prediction models. It also offers pretrained models on Hugging Face and a simple browser chat demo.
New and promisingUnder two months old. Expect rough edges and fast changes.
- Stars
- 214 +5 a day, est.
- Licence
- MIT
- Runs on
- A gpu machine, with a GPU
- Setup
- Technical
- Language
- Python
- Age
- 41 days
What you can do with it
- Pretrain a language model from scratch on datasets such as DCLM, FineWeb or ClimbMix
- Fine-tune a published model for chat or another task
- Run a pretrained 1B model in a browser chat or evaluate it on benchmarks
Ways a business could use it
- Fine-tune a small base model on your own text to prototype a custom chat assistant without paying for a hosted model
- Use it for in-house research into alternative language model designs before committing to a larger training budget
Can you use it commercially?
Free to use, change and sell, including in paid products. Keep the copyright notice.
Not legal advice. Read the project's licence file before building a product on it.
Getting it running
Install dependencies with uv sync, prepare the data with the supplied shell scripts, then run training or chat with the launcher scripts.
Follow the official instructions in the project README.
Things to check
- Research code, so likely rough around the edges for production use
- Needs GPU hardware and large dataset downloads
- Released models are small (up to 1B) and will be weaker than mainstream chat models
- Training uses Weights & Biases logging, which may need an account
Who's talking about it
- news-radar (automated daily digest): AI Open Source Trends 2026-10-31 (Issue #937) Featured in a roundup
Checked 01 Nov 2026. Summary drafted by AI from the project's README.
