Llama 3 8B Instruct

Llama 3 8B Instruct is Meta's instruction-tuned language model released on 18 April 2024. It belongs to the first Llama 3 generation and is an auto-regressive model built on an optimized Transformer architecture. The Instruct variant was tuned for dialogue using supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF).

The model has 8 billion parameters, supports an 8,192-token context window and uses Grouped Query Attention (GQA) together with a tokenizer that has a vocabulary of 128,000 tokens. It was pretrained on over 15 trillion tokens from publicly available sources, with a knowledge cutoff of March 2023.

The weights are distributed under the Meta Llama 3 Community License, allowing local execution, fine-tuning and commercial deployment within the license terms.