meta-llama/Llama-3.1-8B-Instruct

Llama 3.1 8B Instruct: Multilingual Dialogue and Enhanced Capabilities

Meta Llama 3.1 8B Instruct is an 8 billion parameter instruction-tuned large language model, part of the Llama 3.1 collection. Developed by Meta, this model is specifically optimized for multilingual dialogue use cases and general natural language generation. It leverages an optimized transformer architecture with Grouped-Query Attention (GQA) for efficient inference and boasts a substantial 128K token context length. The model was trained on over 15 trillion tokens of publicly available online data, with a knowledge cutoff of December 2023, and fine-tuned using supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to align with human preferences for helpfulness and safety.

Key Capabilities

• Multilingual Dialogue: Optimized for conversations in supported languages including English, German, French, Italian, Portuguese, Hindi, Spanish, and Thai.
• Extended Context Window: Supports a 128K token context length, enabling processing of longer inputs and generating more extensive responses.
• Enhanced Performance: Demonstrates improved scores over Llama 3 8B Instruct across various benchmarks, including MMLU (69.4%), HumanEval (72.6% pass@1), GSM-8K (84.5% em_maj1@1), and API-Bank (82.6% accuracy) for tool use.
• Tool Use Support: Features robust support for multiple tool use formats, facilitating integration into agentic systems.
• Code Generation: Shows strong performance in code-related tasks, with a HumanEval pass@1 score of 72.6%.

Good for

• Developing assistant-like chat applications requiring high performance in multiple languages.
• Natural language generation tasks where a large context window is beneficial.
• Applications requiring robust tool-use and function calling capabilities.
• Research and commercial use in multilingual environments, with a focus on safety and helpfulness.
• Improving other models through synthetic data generation and distillation.