Arcee.ai advances model merging innovations with launch of MergeKit hackathon

MIAMI, April 19, 2024 /PRNewswire/ -- Arcee.ai, the industry leader in specialized language models for enterprise generative AI, is hosting a Model Merging Hackathon?offering cash prizes to developers who create the most innovative projects with Arcee.ai's open source tool MergeKit.

Model Merging is a state-of-the-art technique for combining various large language models (LLMs), each fine-tuned on distinct tasks, into one single model. It's extremely cost-effective, enabling organizations to build more performant LLMs while dramatically reducing the need for GPUs.

Arcee.ai is proud to have joined forces recently with one of the world's leading model merging researchers: the creator of MergeKit, former NASA engineer Charles Goddard. He has joined the Arcee.ai staff to lead a team of researchers who are actively building out the MergeKit repository.

Goddard says that "the potential of model merging has only just begun to be tapped," and he's thrilled that Arcee.ai shares his personal commitment to the open source LLM community.

Arcee.ai CEO Mark McQuade says the efficiency of model merging paired with continual pre-training is at the heart of the language model ecosystem Arcee.ai provides to customers, which include Thomson Reuters and Guild: "Model merging is the next leap forward for transfer learning. Our MergeKit-powered platform has shown significant improvements in our evaluations for domain-adaptation?with particularly strong results in the medical and legal verticals."

The open source LLM community has proven to be very enthusiastic about model merging, with MergeKit quickly surpassing 3-thousand Github stars. Hugging Face's Head of Platform Omar Sanseviero says model merging is pushing forward the capabilities of LLMs, and that "many of the top models on the Open LLM Leaderboard are thanks to the impressive work from the model merging community, which has explored the different merging tools and techniques to achieve significant gains across benchmarks."

The hackathon starts April 19 and runs through May 6, with winners announced the following week. Arcee.ai will award a total of $9,000 in cash prizes across these categories:

• Best new merge
• Best integration with other ecosystems
• Merge that's most upsetting to the natural order.

Go to the Arcee.ai hackathon blog post for full details. Projects will be judged by Charles Goddard and Mark McQuade who say they're looking forward to celebrating the creativity they've been seeing in the emerging Model Merging community.

Contact: Mary MacCarthy | [email protected]

SOURCE Arcee.ai