A field guide to LLMs your lawyer won't hate

What it does This repository is a single, meticulously maintained markdown table that catalogs open LLMs—from T5 and BLOOM to LLaMA 2 and Falcon—explicitly filtered for licenses that permit commercial use. Each entry links to checkpoints, papers, parameter counts, context windows, and the specific license text. It is essentially a reference sheet for teams who need to pick a model without triggering a legal review spiral.

The interesting bit The value is entirely in the curation. While model hubs list everything, this list applies a hard filter: Apache 2.0, MIT, OpenRAIL-M, and a few carefully noted custom licenses. The author even flags edge cases, such as LLaMA 2’s user-count cap and ChatGLM’s registration requirements, turning a compliance chore into a quick scan.

Key highlights

• Covers models from 60M to 176B parameters, spanning encoder-decoder, RNN, and transformer architectures.
• Explicitly tags license types and links directly to checkpoint files on HuggingFace or official repos.
• Notes context lengths and unusual traits, such as RWKV’s theoretically infinite context and MPT-7B’s 84k ALiBi window.
• Includes “Try it” inference links for a handful of models where quick experimentation is available.
• 12.8k stars and open contributions suggest the community treats it as a de facto canonical reference.

Caveats

• Some entries carry custom licenses with usage restrictions (e.g., LLaMA 2’s 700M-user cap, ChatGLM’s registration requirement), so “commercial use” still demands reading the fine print.
• The repository is a curated markdown table, not a framework or benchmarking suite; you will need to fetch weights and run inference elsewhere.

Verdict Worth bookmarking if you are building a product and need to shortlist models that legal will actually let you deploy. Skip it if you are looking for training code, inference wrappers, or performance benchmarks.