Documentation

Llama 3 is a collection of pre-trained and fine-tuned generative text models ranging in scale starting at 8 billion parameters, including a 70 billion parameter model published in April 2024 by Meta. It was released under a special Llama 3 license.

Llama 2 was pre-trained on 15 trillion tokens of data from publicly available sources.

Supported variants

Airtrain supports two variants of the Llama 3 model.

llama-3-8b-chat

Llama-3-8b-chat is an 8B parameter model optimized for dialogue use cases. It outperforms many similarly sized open-source chat models on most benchmarks.

llama-3-70b-chat

Llama-3-70b-chat is a 70B parameter model optimized for dialogue use cases. It outperforms many similarly sized open-source chat models on most benchmarks, and even several popular proprietary models (Gemini Pro, Claude 3 Sonnet).

Meta-Llama-3.1-8b-Instruct-Turbo

An updated version of llama-3-8b-chat. It's multilingual and supports tool use.

Meta-Llama-3.1-70b-Instruct-Turbo

An updated version of llama-3-70b-chat. It's multilingual and supports tool use.

Meta-Llama-3.1-405b-Instruct-Turbo

The flagship Llama 3 model, it is considered to be comparable to gpt-4o.

Meta-Llama-3.2-3b-Instruct-Turbo

A small multilingual model in the Llama 3 series, designed for edge use cases.

Meta-Llama-3.2-11b-Vision-Instruct-Turbo

A multilingual model that supports image understanding and generation (though the Airtrain playground only supports text usage).

Meta-Llama-3.2-90b-Vision-Instruct-Turbo

A multilingual model that supports image understanding and generation (though the Airtrain playground only supports text usage).