News
ReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior ReleaseReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior Release
EveryLLM
BLOOM

BLOOM

A 176-billion-parameter multilingual text model built openly by a volunteer research workshop.

BLOOM is a large language model released in July 2022 as the main result of the BigScience workshop, a year-long volunteer effort coordinated by Hugging Face with French government funding. Its 176 billion parameters generate text across 46 natural and 13 programming languages, and the weights ship under a licence that allows reuse with conditions.

Share on WhatsApp

  • 176 billion parameters, released July 2022
  • Text generation across 46 natural and 13 programming languages
  • Trained March to July 2022 on the Jean Zay supercomputer in France
  • ROOTS training corpus, about 366 billion tokens (1.6TB)
  • Weights under the BigScience Responsible AI License (RAIL) v1.0

What it is

BLOOM is a large language model that writes text, released to the public in July 2022. It holds 176 billion parameters, the adjustable numbers a model learns during training. The BigScience workshop produced it, with Hugging Face coordinating, and the aim was a transparently built rival to systems kept private by companies. It works across 46 human languages plus 13 programming languages, so it can continue code as well as prose. The weights are downloadable, under a licence that permits reuse but sets limits on certain uses.

How it came about

BigScience ran as a year-long research workshop bringing together several hundred volunteers from universities and industry. Hugging Face coordinated the work, with money from the French government. Training took place from March to July 2022 on Jean Zay, a public supercomputer in France run by GENCI and IDRIS, part of CNRS. The team aimed at multilingual coverage from the start, unlike GPT-3. After the base model, the group published xP3, a multilingual instruction dataset, and BLOOMZ, a version tuned on xP3 to follow directions.

Inside the model

The system is a transformer, the common architecture for text generation, and it is autoregressive: each new token is predicted from the ones before it. Training ran on roughly 366 billion tokens, about 1.6TB of text, drawn from a collection called ROOTS. ROOTS mixes material scraped from the OSCAR web corpus, making up 38% of it, with data gathered from a documented, hand-picked set of language sources. Tooling came from the open-source DeepSpeed Megatron libraries. Source code carries an Apache 2.0 licence, while the parameters fall under the project's Responsible AI License, which grants access and reuse but attaches restrictions.

What it can do

BLOOM produces text in 46 natural languages and 13 programming languages, a breadth that made it useful for multilingual chatbots such as BLOOMChat and HuggingChat. Its multilingual reach was a deliberate design choice rather than an afterthought. The BLOOMZ variant adds instruction-following, meaning it responds to directions instead of simply continuing a prompt.

How to use it

Anyone can download the parameters from the project's repository on Hugging Face. The code is under Apache 2.0. The weights use the BigScience Responsible AI License (RAIL) v1.0, which allows open access and reuse subject to usage limits. The project's website is hosted by Hugging Face.

Safety and testing

The licence itself carries the safety stance: it is called a Responsible AI License and it attaches conditions on how the model may be used, going beyond a plain open release.

Reception and debate

BLOOM was picked up by chatbot projects, including BLOOMChat and HuggingChat, chosen for its language coverage. It stands as an example of a large model built in the open rather than inside a single company.

Why it matters

The project showed that a 176-billion-parameter model could be trained by a distributed volunteer effort on public infrastructure, with its data sources documented. It also fed further work: the xP3 instruction dataset and the BLOOMZ tuning both came out of the same initiative, and the model went on to power multilingual assistants.

Questions people ask about BLOOM

Who made BLOOM?

It came out of BigScience, a year-long research workshop run by several hundred volunteers from academia and private companies. Hugging Face coordinated the effort, and the French government provided funding.

How big is it and what can it do?

The model has 176 billion parameters and generates text in 46 natural languages and 13 programming languages. That language range is why chatbot projects such as BLOOMChat and HuggingChat used it.

Can I download and use it?

Yes. The parameters are published on Hugging Face under the BigScience Responsible AI License (RAIL) v1.0, which allows access and reuse but places some limits on use. The source code is under Apache 2.0.

What was it trained on?

Training used ROOTS, a corpus of about 366 billion tokens (roughly 1.6TB). Around 38% came from the OSCAR web corpus, with the rest gathered from a documented, manually chosen list of language data sources.

Is there a version that follows instructions?

Yes. BLOOMZ is a variant fine-tuned on xP3, a multilingual dataset for supervised learning that BigScience released alongside the base model.

Good reading and sources

Where this page comes from

The facts on this page come from the Wikipedia article "BLOOM (language model)", written by its contributors and used under the CC BY-SA 4.0 licence, and from its Wikidata record (CC0). EveryLLM rewrote the text in its own words, with AI assistance, and compares every page with the article to make sure no passage is copied. It may be reused under the same CC BY-SA 4.0 licence with a link back.

It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

The weekly digest

Every Monday: the model of the day, new pages on EveryLLM, and the conferences and anniversaries of the week ahead.