Falcon vs. Guanaco

Overview

Falcon LLM is the flagship LLM of the Technology Innovation Institute in Abu Dhabi.

Falcon LLM is a powerful LLM developed by the Technology Innovation Institute (https://www.tii.ae). Unlike other popular LLMs, Falcon was not built off of LLaMA, but instead using a custom data pipeline and distributed training system. The dataset is the RefinedWeb dataset (available on Hugging Face), and the initial models are available in 7B and 40B forms. Instruct and chat-finetuned models are available as well. Perhaps most excitingly, Falcon outperforms LLaMA, the previous de facto pretrained LLM.

Initial release: 2023-05-23

Reference

https://falconllm.tii.ae/

Further Reading

• https://huggingface.co/tiiuae
• https://twitter.com/Thom_Wolf/status/1662085371826405377
• https://www.businesswire.com/news/home/20230531005608/en/UAE%E2%80%99s-Falcon-40B-World%E2%80%99s-Top-Ranked-AI-Model-from-Technology-Innovation-Institute-is-Now-Royalty-Free

Overview

Guanaco is an LLM based off the QLoRA 4-bit finetuning method developed by Tim Dettmers et. al. in the UW NLP group. Guanaco achieves 99% ChatGPT performance on the Vicuna benchmark.

Guanaco is an LLM that uses a finetuning method called LoRA that was developed by Tim Dettmers et. al. in the UW NLP group. With QLoRA, it becomes possible to finetune up to a 65B parameter model on a 48GB GPU without loss of performance relative to a 16-bit model. The Guanaco model family outperforms all previously released models on the Vicuna benchmark. However, given the models are based off of the LLaMA model family, commercial use is not permitted.

Initial release: 2023-05-23

Reference

https://github.com/artidoro/qlora

Further Reading

• https://arxiv.org/abs/2305.14314
• https://twitter.com/Tim_Dettmers/status/1661379354507476994