Equall

We’re honored to share that our groundbreaking research on legal AI has been accepted to the NeurIPS 2024 conference, the preeminent conference for machine learning and artificial intelligence. SaulLM is a family of legal AI models purpose-built for legal reasoning, and we couldn’t be more excited to share our work with the global AI community. Check out our latest blog for all the details: https://lnkd.in/erQ-qfuB #NeurIPS2024 #LegalTech #AI

🚀 Building the Future of Multilingual AI with EuroLLM. We are incredibly proud to announce the launch of EuroLLM, a new multilingual large language model supporting all 24 official EU languages! Developed from scratch in collaboration with Instituto Superior Técnico, The University of Edinburgh and Université Paris-Saclay, EuroLLM was trained on MareNostrum, one of Europe’s most advanced supercomputers. This marks an exciting milestone for multilingual AI— with the goal of delivering unparalleled accuracy, efficiency, and speed for task-specific applications. EuroLLM is yet another step towards cementing its AI sovereignty as the project aims to create a flywheel for innovation by sparking new opportunities for collaboration and growth whilst levelling the playing field by closing language gaps across Europe and the world. Thank you to our partners EuroHPC Joint Undertaking (EuroHPC JU) for making this possible! 🌍✨Go try it on Hugging Face today and available for users of Widn.AI https://hubs.li/Q02ZZB3Z0

At AMD, we're thrilled to see our technology pushing the boundaries of domain-specific AI. Equall, a trailblazer in legal AI, has harnessed the power of AMD Instinct MI250X accelerators to train SaulLM—the first family of open language models designed specifically for law. Pierre Colombo, Chief Scientist at Equall, shares: "The MI250X cards enabled large-scale model training, accelerating our experimentation and paving the way to scalable, flexible, and cost-effective solutions. With ROCm support, integrated with PyTorch, DeepSpeed, and other deep-learning libraries, we utilized advanced features like flash attention and are excited about incorporating the MI300X in future work." Learn more about how Equall is transforming the legal industry with AMD technology: https://lnkd.in/g6e36iRp

At AMD, we're thrilled to see our technology pushing the boundaries of domain-specific AI. Equall, a trailblazer in legal AI, has harnessed the power of AMD Instinct MI250X accelerators to train SaulLM—the first family of open language models designed specifically for law. Pierre Colombo, Chief Scientist at Equall, shares: "The MI250X cards enabled large-scale model training, accelerating our experimentation and paving the way to scalable, flexible, and cost-effective solutions. With ROCm support, integrated with PyTorch, DeepSpeed, and other deep-learning libraries, we utilized advanced features like flash attention and are excited about incorporating the MI300X in future work." Learn more about how Equall is transforming the legal industry with AMD technology: https://lnkd.in/g6e36iRp

The legal-specific LLM movement takes another step forward 🚀 - this time in India 🇮🇳 . AL explores and looks at the way ahead, i.e. via combining L-LLMs, or perhaps small, very focused legal models, with general models to get the best outcome. Or, as Zach Abramowitz suggests, we can just wait for GPT-5 / 6 / 7 to come along and make specific training redundant, which is a valid point if things work out that way. Personally, as discussed with Jake Jones last week, I favour a combination approach right now. So, small, large, general, and specific - let's see what works best in unison. After all, the goal here is not to prove if X or Y is better, but to achieve better legal efficiency ASAP. Why wait? The move also comes after Equall + Thomson Reuters also made L-LLM moves this month. HT Joel Hron + Lexlegis.ai The debate continues.... LINK: https://lnkd.in/eRqqG9pP #legaltech #genAI #LLMs

Things just got very serious for Legal-specific LLMs....! AL talks to Joel Hron, CTO at Thomson Reuters about the Safe Sign deal and why, despite assumptions, L-LLMs with the right training data and support, may well be the way ahead. Cue debate on general vs specific LLMs. TBH, I have to agree with Hron here, if you have massive amounts of high quality legal-specific data it may well give you an advantage over a general LLM that has been system-prompted and refined on legal. Will this give TR the edge? AL explores. HT Equall LexisNexis vLex Safe Sign Technologies LINK: https://lnkd.in/encaBz_b

I'd like to give an absolutely massive shout-out to Pierre Colombo and the rest of his team for dropping SaulLM-141B this month, the world's largest open legal LLM, with a whopping 141 billion parameters, trained, in small part, on the Open Australian Legal Corpus! Congrats guys, I look forward to playing with it and seeing what comes next. You can find their model here https://lnkd.in/gjx6xnhJ and their earlier 7 billion 'baby' version here https://lnkd.in/gknhkz5Y. Their paper can be found here https://lnkd.in/gKVECsgy.