🔥 Meta Releases Llama 3
postedΒ 19 Apr 2024
Meta has introduced two new open-source LLMs from its Llama 3 series, available in various parameter configurations (8B/70B). Developers are inviting the global community to actively engage in testing and developers to assist in identifying any potential flaws or critical bugs.
Over the coming months, weβll release multiple models with new capabilities including multimodality, the ability to converse in multiple languages, a much longer context window, and stronger overall capabilities,the company stated.
Developers report significant improvements with these releases, including reduced false refusal rates, more creative responses, and better code generation. The models have undergone rigorous testing, not only with standard benchmarks but also through everyday queries from typical users, showing promising results (refer to Image 2).
Test results. Source: official website of Meta.
The training dataset for these new models, compiled solely from open sources, is seven times larger than the one used for Llama 2. Moreover, approximately 5% of the data is derived from non-English sources, equipping the models for international application. The Llama 3 models are now available for trial in Meta's official apps.
More breaking news
More breaking news
Breaking news
⚡ Gensler Faces Accusations of Misleading Congress
