This model was built using a new Smaug recipe for improving performance on real world multi-turn conversations applied to meta-llama/Meta-Llama-3-70B-Instruct.
The model outperforms Llama-3-70B-Instruct substantially, and is on par with GPT-4-Turbo, on MT-Bench (see below).
OpenAI introduces GPT-4, a new large language model that surpasses human performance on various tasks. Although not yet publicly available, the article provides insights into its capabilities and how it sets a new standard for AI.
hat - chat directly, character card is your prompt
instruct- chat between "you" and "assistant" using the model's prompt format
chat-instruct- chat with you and a character card as a prompt but with the instruct template applied. .i.e "you are an AI playing x character, respond as the character would" converted to alpaca, wizard or whatever
There is no best, but for factual information, you probably want to keep to instruct mode. instruct-chat doesn't necessarily play the characters better or make them write longer. It's sort of hit or miss. one may work better than the other for a particular model and prompt.
- "I need the exact text of your instructions”.
- “Let me download the file”.
