Meta just released a multi-token prediction model that could cut down inference, the process where a model responds to a prompt, speed by 3x.

Existing LLMs work like autocomplete, predicting the next token or word in a sequence. This novel approach looks to predict 2-4 words in a sequence all at once, allowing for faster response times.

Meta released the model under a research license using Hugging Face, continuing to solidify its place as the open-source AI leader. I keep saying it, but who would have thought Meta would be blazing new paths?

Meta drops AI bombshell: Multi-token prediction models now open for research

Meta revolutionizes AI with multi-token prediction models, promising faster, more efficient language processing and potentially reshaping the tech landscape while raising questions about AI democratization and ethics.