Unbounded Labs Open-Sources Bart, an LLM Trained Only on Pre-1931 Text

Unbounded Labs has open-sourced Bart, a 2.82B-parameter retro LLM trained entirely on 201B tokens of pre-1931 English text for about $800, exploring whether models can reach the same conclusions as past scientists.

2026-08-25 ~ 2026-08-25 · 2 related posts

• Unbounded Labs releases Bart, a vintage LLM trained on pre-1931 text — soggydoggy8 · 2026-08-25

1 near-duplicate retellings: soggydoggy8