Llama 3.1: Meta's Largest Open Model has Been Released

Llama 3.1 405B has officially been released and Meta terms the model as the world’s largest open model.

Meta hinted at Llama 3.1’s arrival in April and stated that the model would match the performance of leading private models such as OpenAI. The newly released model can be downloaded on llama.meta.com, Google Cloud, and other cloud websites. Users can also try out the new model through WhatsApp’s chatbot meta.ai.

Llama 3.1: Feature of Meta’s new model

The 405B executes various functions and tasks such as solving a math equation, and coding. The model can even summarise documents in English, Hindi, Portuguese, Thai, French, German, Spanish and Italian.

The 405 model comes with an impressive 128K context length and improved reasoning capabilities. Meta has evaluated around 150 benchmark datasets across a variety of languages. The 3.1 model has also undergone human evaluations comparing the model to different models. Meta has concluded that the newly released model competes with the likes of GPT -4o, Claude 3.5 Sonnet and GPT -4.

405B is the first Llama model trained to more than 16 thousand H100 GPUs. Additionally, the model also utilises synthetic data. To improve the model’s efficiency in languages besides English, Llama 3.1 has been trained on data not written in English. 405B’s ecosystem is partnered with over 25 partners such as Google Cloud, AWS, Azure, NVIDIA, Dell, Databricks and Groq.

Meta’s new model plans to increase components compatible with the model which includes a reference system. In addition, new safety and security tools such as the Prompt Guard and the Llama Gaurd 3 come equipped with the 405B.

Meta has also released two other small models such as the Llama 3.1 70B and the Llama 3.1 8B meant for general tasks.

OpenAI’s CriticGPT is the Solution to Improving AI-Generated Code.