Image credits: Bryce Durbin / TechCrunch
Meta has unveiled the newest additions to its Llama series of open generative AI models, introducing two models within the Llama 3 family. Additional models are expected to be released at a future date yet to be announced.
Meta's latest Llama 3 models are trained on a new dataset from public sources, with four times more code than the Llama 2 dataset. This large dataset covers about 30 languages, which makes up 5% of the total data, aiming to improve how the model works with different languages. Also, Meta added synthetic (AI-generated) data to train the models on longer documents. This approach has sparked discussions about its effect on how well the models perform.
“Our goal in the near future is to make Llama 3 multilingual and multimodal, have longer context and continue to improve overall performance across core [large language model] capabilities such as reasoning and coding,” Meta writes in a blog post. “There’s a lot more to come.”
Source: TechCrunch
Comentários