Inside BLOOM, the Year-Long Experiment in Open, Multilingual AI

BLOOM, released by the BigScience workshop in July 2022, was the product of a roughly year-long, genuinely international collaboration involving researchers across dozens of countries -- an unusual model of open science applied to large-scale AI at a time when frontier model training was already becoming concentrated inside a handful of well-funded companies.

Why it mattered

At 176 billion parameters, BLOOM was, at the time, one of the largest genuinely open multilingual language models ever released, trained on 46 natural languages and 13 programming languages. It demonstrated that large-scale model training didn't have to be the exclusive domain of a single corporation -- though its RAIL license also introduced usage restrictions that sparked ongoing debate about what "open" should really mean for an AI release, a debate covered in more detail in Open Weights vs. Open Source AI.

On the record

• Issuer: BigScience
• License: RAIL license
• Claimed cutoff: 2022-07-11
• Status: hashed, queued for on-chain attestation

See the full hash breakdown on BLOOM-176b's Explorer page.