Aya Expanse Combining Research Breakthroughs For A New Multilingual Frontier 2024 12 06
Source
cohere.com
Author
Cohere
Date
Context
Cohere's Aya Expanse research asks whether relatively compact models can deliver strong performance across languages. The family contains models with eight billion and 32 billion parameters, the learned values that determine their behavior. Its research abstract reports evaluations on a demanding prompt set translated into 23 languages.
In that evalA repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.Full definition →, the authors report a 54% win rate for Aya Expanse 32B against Meta's Llama 3.1 70B. The finding shows that the larger model did not automatically prevail on these multilingual prompts. Releasing the model weights and multilingual evaluation dataset also gives researchers material for further comparisons.
Terms in this piece · Glossary
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
open weights — A model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.
postCohere releases open-weights translation model North Small Translate
articleThe Data Provenance Initiative A Large Scale Audit Of Dataset Licensing And Attribution In Ai 2023 10 25
