Llama-3.1-8B-Instruct

Llama 3.1 8B Instruct is Meta's instruction-tuned, text-only large language model released in July 2024 as part of the Llama 3.1 family. Built on an optimized Transformer architecture with Grouped-Query Attention (GQA), it was pretrained on over 15 trillion tokens and fine-tuned via supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF) for helpful, safe dialogue. It supports eight languages (English, German, French, Italian, Portuguese, Hindi, Spanish, and Thai) and features a 128K-token context window, native tool-use formatting, and improved multilingual capabilities over Llama 3.

On standard benchmarks, Llama 3.1 8B Instruct delivers strong performance for its size class: MMLU 69.4, MMLU-Pro 48.3, GSM8K 82.4, MATH 51.9, HumanEval 72.6, MBPP 69.4, IFEval 80.4, and ARC 83.4. It is well suited for multilingual chat assistants, lightweight coding helpers, and on-device or cost-efficient server deployments where larger models are impractical.

Released under the Llama 3.1 Community License, weights are openly available on Hugging Face. The model requires roughly 16GB of VRAM for BF16 inference and integrates with Transformers, vLLM, and the official Llama toolchain. Knowledge cutoff is December 2023.