What is Reflection 70B?

Reflection 70B is a state-of-the-art open-source large language model built on Meta's Llama 3.1 70B Instruct. It features a unique 'Reflection-Tuning' methodology that enables self-correction during generation, improving accuracy and reliability. The model excels in various tasks including coding, complex reasoning, and creative writing, and is available for free online use.

How to use Reflection 70B?

To use Reflection 70B, navigate to the website and access the chat interface. No sign-up is required. Type your question or task into the chat box and press enter. The model will process your input and provide a response; you can continue the conversation by replying or asking new questions.

Core features of Reflection 70B

Innovative Reflection-Tuning training technique for self-correction
Real-time error self-correction using special tokens
Seamless compatibility with existing tools and pipelines (based on Llama 3.1 70B Instruct)
Benchmark-topping performance (MMLU, HumanEval, MATH, IFEval, GSM8K)
High-precision task mastery with step-by-step reasoning
Advanced coding capabilities outperforming other open-source models