Nous Hermes Llama2 13B

Nous Hermes Llama2 13B is a 13 billion parameter open-weight model fine-tuned on over 300,000 instructions generated almost entirely from synthetic GPT-4 outputs, supplemented with diverse datasets including GPTeacher, WizardLM, and CodeAlpaca. It was trained at 4,096 sequence length on an 8x A100 80GB machine, producing a model known for long responses, a lower hallucination rate, and the absence of OpenAI's censorship mechanisms. The model uses the Alpaca prompt format and, on Nous Research's reporting, ranked first on ARC-c, ARC-e, Hellaswag, and OpenBookQA, with a GPT4All benchmark average of 70.0. It ships under an MIT license that allows commercial use. Nous Hermes Llama2 13B serves developers prioritizing open-weight models for research, fine-tuning, or deployment where transparency and customizability matter. Its size makes it usable with quantization for resource-constrained settings, though newer Hermes 3 and Hermes 4 variants offer stronger reasoning and agentic capabilities.