AI - Bloom
Bloom uses Hugging Face to provide access to BigScience's multilingual language models, supporting inference and training in 46 languages and 13 programming languages.
- Last Audited At
About Bloom
Bloom is a large-scale language model developed by BigScience, an open science initiative inspired by other collaborative research projects. The architecture of Bloom is similar to GPT3 but has been trained on 46 different languages and 13 programming languages. Bloom is available in several versions, including bloom-560m, bloom-1b1, bloom-1b7, bloom-3b, bloom-7b1, and bloom, each with varying numbers of parameters (ranging from 560 million to 7 billion).
Bloom's models are accessible via Hugging Face's model hub. The configuration classes for these models (e.g., BloomConfig) are used to instantiate a Bloom model according to the specified arguments, defining its architecture. Instantiating a Bloom configuration with default settings will yield a similar configuration to bigscience/bloom.
Bloom's tokenizer, BloomTokenizerFast, is utilized for converting inputs into their corresponding input IDs and vice versa. The class supports various options such as handling errors when decoding bytes to UTF-8.
Bloom's capabilities include inference and training. Inference involves using the model for tasks like language modeling, text classification, token classification, and question answering. Training entails fine-tuning the model on specific datasets or tasks to improve its performance. There are several resources available to help users get started with Bloom, including blogs on optimization, training technology, and inference techniques.
