meta-llama/Meta-Llama-3-8B-Instruct
🧠 AI Modelmeta-llama
Meta's 8B instruction-tuned LLM powering chat, reasoning, and code generation at scale.
Meta-Llama-3-8B-Instruct is the fine-tuned, instruction-following variant of Meta's Llama 3 8B base model, designed for dialogue and assistant-style applications. It leverages a transformer architecture trained on a significantly larger and higher-quality dataset than Llama 2, with an expanded 128K-token vocabulary and a 8K context window. The instruction-tuning process uses supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF) to align outputs with user intent. It supports multilingual dialogue (primarily English) and excels at code generation, reasoning, and following complex prompts. The model is distributed as a gated open-source release under the Llama 3 Community License, requiring users to accept Meta's license terms before access. It is available in safetensors format for efficient and safe inference, and is compatible with popular inference stacks including Hugging Face Transformers, vLLM, and llama.cpp.
💡Highlights
- ├─8B params, 128K vocab, 8K context window
- ├─SFT + RLHF instruction tuning pipeline
- └─1M+ downloads on Hugging Face
🎯For
- ├─AI researchers
- ├─LLM application developers
- └─Open-source contributors