Hermes 3 - Llama-3.1 70B is a generalist language model developed by Nous Research, built as a fine-tuned version of Meta's Llama 3.1 architecture. It is designed to offer high steerability and user control, emphasizing a neutral alignment that allows the model to follow complex system prompts and instructions without the restrictive guardrails typical of many contemporary assistants. The model is fine-tuned to excel in reasoning, creative writing, and multi-turn conversation.## Key CapabilitiesThe model introduces advanced agentic capabilities, specifically improved function calling and structured output generation. It is optimized for long-context coherence, maintaining the 128k token context window of its base architecture to handle extensive documents and long-form tasks. Hermes 3 also features refined roleplaying abilities, allowing it to adopt and maintain diverse personas with high fidelity.Training for the model utilized primarily synthetically generated response data, focusing on instruction following and adaptive behavior. It employs the ChatML prompt format to facilitate structured interactions, making it suitable for complex workflows and tool-integrated applications.