Overview
NousResearch/Llama-2-13b-chat-hf is a 13 billion parameter model from Meta's Llama 2 family, specifically fine-tuned for dialogue. It is an auto-regressive language model based on an optimized transformer architecture, enhanced with supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to align with human preferences for helpfulness and safety. The model was trained on a new mix of publicly available online data, totaling 2.0 trillion tokens, with a context length of 4096 tokens.
Key Capabilities
- Dialogue Optimization: Specifically fine-tuned for assistant-like chat applications.
- Performance: Outperforms many open-source chat models on various benchmarks and achieves parity with some popular closed-source models in human evaluations for helpfulness and safety.
- Safety Alignment: Incorporates RLHF to improve safety, demonstrating 0.00% toxic generations on the ToxiGen benchmark for chat versions.
Intended Use Cases
- Assistant-like Chat: Ideal for conversational AI systems and chatbots.
- Commercial and Research: Suitable for both commercial products and academic research in English.
Limitations
- English Only: Primarily intended for use in English; performance in other languages is not guaranteed.
- Potential for Objectionable Outputs: Like all LLMs, it may produce inaccurate, biased, or objectionable responses, requiring developers to perform safety testing for specific applications.