In developing BLOOM, the BigScience team set out to create an LLM that’s more accessible than any other major model to date. Unlike other LLMs, BLOOM is available to pretty much any individual or institution — previously, Meta’s OPT-175B was one of the most easily accessible LLMs to date, but it is only available to researchers for non-commercial use.
“This is only the beginning. BLOOM’s capabilities will continue to improve as the workshop continues to experiment and tinker with the model,” the developers wrote in a blog post announcing BLOOM’s launch.
Hugging Face launched the BigScience initiative in 2021, in an effort to “democratize” LLMs. The team of developers consisted of more than 1,000 researchers based in over 60 countries. In addition to being accessible to a much larger group of users, BLOOM is also capable of producing text in 46 different languages and dialects, as well as 13 programming languages. This contrasts significantly with other mainstream models which typically focus on one or two languages at a time, most commonly English and Chinese.
According to a recent report from NBC News, the developers behind BLOOM set out to address some of the pain points that LLMs often struggle with. By making the model open-source, the researchers believe that it will be more accessible to users interested in learning about and developing LLMs.
“Large ML models have changed the world of AI research over the last two years but the huge compute cost necessary to train them resulted in very few teams actually having the ability to train and research them,” said Thomas Wolf, co-founder of Hugging Face, the startup the spearheaded the BigScience initiative, in an interview the The Next Web.
Additionally, with 176 billion parameters, it’s larger than the mainstream options like OPT-175B and GPT-3.
“BLOOM is a demonstration that the most powerful AI models can be trained and released by the broader research community with accountability and in an actual open way, in contrast to the typical secrecy of industrial AI research labs,” said Teven Le Scao, one of the co-leads for BLOOM’s development.