Model Overview
This is the Hugging Face Transformers format of Meta's Llama-2-7b-chat model, a 7 billion parameter large language model. It is part of the Llama 2 family, which includes models ranging from 7B to 70B parameters, all pretrained and fine-tuned for generative text tasks. The Llama-2-Chat variants, including this 7B model, are specifically optimized for dialogue and assistant-like chat applications.
Key Capabilities & Features
- Optimized for Dialogue: Fine-tuned using supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to align with human preferences for helpfulness and safety in chat scenarios.
- Performance: Outperforms many open-source chat models on various benchmarks and achieves competitive results with some popular closed-source models like ChatGPT and PaLM in human evaluations for helpfulness and safety.
- Architecture: Employs an auto-regressive language model with an optimized transformer architecture.
- Training Data: Pretrained on 2 trillion tokens from publicly available online data, with fine-tuning incorporating over one million human-annotated examples. Data cutoff for pretraining is September 2022, with some tuning data up to July 2023.
- Context Length: Supports a context length of 4096 tokens.
Intended Use Cases
- Assistant-like Chat: Primarily intended for commercial and research use in English for conversational AI.
- Natural Language Generation: While the chat version is tuned for dialogue, the base Llama 2 models can be adapted for various natural language generation tasks.
Limitations
- English Only: Testing has been conducted in English, and use in other languages is considered out-of-scope.
- Potential for Objectionable Content: As with all LLMs, it may produce inaccurate, biased, or objectionable responses, requiring developers to perform safety testing for specific applications.