Guanaco
#56 in Open-Source LLMsUnknown · 2× · last seen Sep 24, 2026
20
Momentum
Guanaco is a family of Open Source language models (7B, 13B, 33B, and 65B parameters) Fine-Tuned by researchers at the University of Washington using the QLoRA method based on Meta's LLaMA models. QLoRA enables Fine-Tuning of a 65B-parameter model on a single 48-GB GPU through 4-bit quantization while maintaining the performance of 16-bit Fine-Tuning. According to the authors, the best Guanaco model achieved 99.3% of ChatGPT's performance on the Vicuna Benchmark, surpassing all previously published Open Source models. Guanaco is designed as a research project; code and model weights have been published via GitHub and Hugging Face.
Momentum trend
29.06.27.09.
Features
| Key Benchmark (%) | 99.3% of ChatGPT's performance level on the Vicuna benchmark (GPT-4 evaluation) |
| License | QLoRA code: MIT license; Guanaco adapter weights: Apache 2.0; use requires access to LLaMA base model under LLaMA license |
| Multimodality | No – text-only language model (chatbot), no multimodal capabilities documented |
| Platform | Available on GitHub (artidoro/qlora) and Hugging Face; demo via Gradio/Colab |
| Release Date | May 23, 2023 (arXiv paper), GitHub repo updated through July 18, 2023 |