Artificial intelligence

BLOOM: the open revolution in artificial intelligence

Today's large AI models need enormous computational resources to train.

A neural network made up of billions or even billions of parameters requires resources in the tens of millions of Euros.
With such huge costs, the only actors able to build and train a great model of artificial intelligence are the multinationals.

Cost of AI research as a barrier to entry

With these characteristics, the cost of AI research constitutes a barrier to entry.
Over the last few years we have witnessed the proliferation of smaller and less complex models, however far from the large LLM linguistic models.

BLOOM and the open revolution

In recent years we have seen a turnaround. For example, Meta's release of OPT-175B (Open Pretrained Transformer), a language model trained with public datasets and made available to researchers in “quasi” open mode.
But the news of the moment is the release of BLOOM LM by BigScience.

BLOOM is an open access multilingual language model that contains 176 billion parameters and has been trained for 3,5 months on 384 A100–80 GB GPUs.
A BLOOM checkpoint takes up 330GB of disk space, so it seems impossible to run this model on a desktop computer.
However, you just need enough disk space, and at least 16GB of RAM to run this model on your computer.

Innovation newsletter
Don't miss the most important news on innovation. Sign up to receive them by email.

BLOOM is a collaborative effort of over 1.000 scientists.
It is important that such a broad multilingual model is openly available to all.

BLOOM architecture

BLOOM is a causal model language, which means it has been trained as a predictor of the next token.
This seemingly simple strategy of predicting the next token in a sentence, based on a set of previous tokens, has been shown to capture a certain degree of reasoning power for large language models.
This allows BLOOM and similar models to connect multiple concepts in a sentence and to be able to solve non-trivial problems such as arithmetic, translation and programming with fair accuracy.
BLOOM uses a Transformer architecture consisting of an input embedding layer, 70 Transformer blocks, and an output language modeling layer, as shown in the following figure.

Article extracted from the Post of Luca Sambucci, if you want to read theNews.AI

Innovation newsletter
Don't miss the most important news on innovation. Sign up to receive them by email.

Latest Articles

Innovative intervention in Augmented Reality, with an Apple viewer at the Catania Polyclinic

An ophthalmoplasty operation using the Apple Vision Pro commercial viewer was performed at the Catania Polyclinic…

May 3, 2024

The Benefits of Coloring Pages for Children - a world of magic for all ages

Developing fine motor skills through coloring prepares children for more complex skills like writing. To color…

May 2, 2024

The Future is Here: How the Shipping Industry is Revolutionizing the Global Economy

The naval sector is a true global economic power, which has navigated towards a 150 billion market...

May 1, 2024

Publishers and OpenAI sign agreements to regulate the flow of information processed by Artificial Intelligence

Last Monday, the Financial Times announced a deal with OpenAI. FT licenses its world-class journalism…

April 30 2024