Hugging Face, aims to democratise artificial intelligence through open source and open science.
Falcon 40B, the first 40 billion parameter, large-scale open source AI model developed by Abu Dhabi’s Technology Innovation Institute (TII), shot to the top of Hugging Face’s most recent Open Large Language Model (LLM) Leaderboard last week.
In order to get the desired ranking, Falcon 40B outperformed well-known models including LLaMA from Meta (including its 65B model), StableLM from Stability AI, and RedPajama from Together. The index makes use of four important benchmarks from the Eleuther AI Language Model Evaluation Harness, a comprehensive framework that evaluates generative language models on: the AI2 Reasoning Challenge (25-shot), a set of elementary school science questions; HellaSwag (10-shot), a test of common sense inference that is simple for humans but difficult for SOTA models; MMLU (5-shot), a test to assess a text model’s multitask accuracy; and TruthfulQA (0-shot),
LLMs and chatbots are tracked, ranked, and evaluated as they are deployed using Hugging Face’s Open LLM Leaderboard, an impartial review tool available to the AI community. Falcon 40B, which was trained on one trillion tokens, represents a pivotal moment in the UAE’s progress towards AI leadership by making it possible for everyone to access the model’s weights for both research and commercial use. The model’s ability to improve AI for the benefit of all humanity by making it more transparent, inclusive, and accessible is supported by the new rating.
With this most recent breakthrough, TII was able to add the UAE to the exclusive group of nations aiming to promote AI innovation and cooperation. The UAE will now have a place at the table when it comes to generative AI models. The 180B AI model, which will be the next iteration of Falcon, is already under development by TII. Visit FalconLLM.TII.ae to find out more about the most recent open-sourced Falcon 40B AI model. You may access the initial Falcon 40B announcement here: Launch of the open-source “Falcon 40B” large language model for research and commercial use by the UAE’s Technology Innovation Institute.
