Model Overview Card for Hermes 2 Theta Llama-3 70B

Basic Information

• Model Name: Hermes-2 Theta Llama-3 70B
• Developer/Creator: Nous Research, in collaboration with Charles Goddard and Arcee AI
• Release Date: June 2024
• Version: 1.0
• Model Type: Text Generation

Description

Overview

Hermes-2 Theta Llama-3 70B is a powerful LLM that combines the strengths of Nous Research's Hermes 2 Pro and Meta's Llama-3 Instruct models. It is trained using a merge technique and further refined with Reinforcement Learning from Human Feedback (RLHF), resulting in a model that generates coherent and contextually accurate text.

Key Features

• Proficiency in structured outputs and function calling
• Utilizes Chat ML for highly structured and steerable multi-turn dialogue
• Supports JSON-formatted responses for tasks requiring structured data
• Capable of generating API calls, parsing responses, and returning structured data

Intended Use

• Interactive chatbots and virtual assistants
• Creative writing and interactive storytelling
• Business applications requiring structured data and function calling (e.g., fetching stock data)

Language Support

‍The model primarily supports English but is versatile enough to handle multilingual inputs with varying degrees of proficiency.

Technical Details

Architecture

• Transformer-based LLM with 70 billion parameters

Training Data

• Diverse dataset sourced from both open and proprietary data pools, encompassing web content, scientific literature, and synthetic data
• The model has a knowledge cutoff in early 2024.

Performance Metrics

• Achieved high accuracy on benchmarks like GPT4All, AGIEval, and BigBench
• Performed well on tasks like arc_challenge and arc_easy, showcasing strong logical reasoning and knowledge-based question answering
• Scored highly on the Truthful QA benchmark, demonstrating the ability to generate factually accurate responses

Comparison to Other Models

• Matches Open AI's GPT-4 on some benchmarks
• Surpasses Llama-3 70B Instruct on nearly all benchmarks, including IFLM

Usage

Code Samples

The model is available on the AI/ML API platform as "NousResearch/Hermes-2-Theta-Llama-3-70B".

API Documentation

Detailed API Documentation is available on the AI/ML API website, providing comprehensive guidelines for integration.

Ethical Guidelines

Ethical considerations around the model focus on its potential biases and the need for responsible deployment, particularly in applications where factual accuracy and neutrality are critical.

Licensing

The model is released under the Llama 3 license, with specific conditions for commercial and non-commercial use.