Meta highlighted that Llama 3 had improved on previous models’ limitations [File]
| Photo Credit: REUTERS
Meta on Thursday introduced its latest large language model, Llama 3, which it claims is the “most capable openly available LLM to date.”
The social media firm also said more models are slated to be released soon.
Meta highlighted that Llama 3 had improved on previous models’ limitations and that features such as multilingual/multimodal capabilities were a goal in the “near future,” along with better reasoning and coding performance.
“Thanks to improvements in pretraining and post-training, our pretrained and instruction-fine-tuned models are the best models existing today at the 8B and 70B parameter scale,” said Meta in a blog post on Thursday.
“With this new model, we believe Meta AI is now the most intelligent AI assistant that you can freely use,” said Mark Zuckerberg in an Instagram post on Thursday, adding that Meta AI would also be able to animate photos and speed up image generation.
(For top technology news of the day, subscribe to our tech newsletter Today’s Cache)
However, Meta AI stirred up controversy this week after the chatbot pretended to be the parent of a child while it was responding to users in a Facebook group.
Meta stressed that it used high quality data in order to train its latest model, as it faces legal action over allegations of scraping copyrighted works for AI training.
“Llama 3 is pretrained on over 15T tokens that were all collected from publicly available sources. Our training dataset is seven times larger than that used for Llama 2, and it includes four times more code,” said the company, noting that over 5% of the Llama 3 pretraining data set consisted of high-quality non-English data from over 30 languages.
However, Meta acknowledged there was still work to be done and invited feedback from users.
Llama will be made available on AWS, Databricks, Google Cloud, Hugging Face, Kaggle, IBM WatsonX, Microsoft Azure, NVIDIA NIM, and Snowflake.
Meta AI was not yet available in India.
