Llama 2: An Overview

Llama 2 is a family of large language models developed by Meta, available in 7B, 13B, and 70B parameter sizes. This particular model is the 7B fine-tuned version, specifically optimized for dialogue applications. The models are built on an optimized transformer architecture and leverage supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to align with human preferences for helpfulness and safety.

Key Capabilities

• Dialogue Optimization: Llama-2-Chat models are fine-tuned for assistant-like conversational use cases.
• Performance: Outperforms many open-source chat models on tested benchmarks and achieves parity with some popular closed-source models in human evaluations for helpfulness and safety.
• Context Length: All Llama 2 models support a 4k token context length.
• Training Data: Pretrained on 2 trillion tokens of publicly available online data, with fine-tuning data including over one million human-annotated examples.
• Safety Alignment: Tuned versions incorporate SFT and RLHF to enhance safety and reduce objectionable responses.

Good For

• Commercial and Research Use: Intended for both commercial and research applications, primarily in English.
• Assistant-like Chatbots: The fine-tuned Llama-2-Chat models are ideal for building conversational AI assistants.
• Natural Language Generation: Pretrained models can be adapted for various natural language generation tasks.