Mark McQuade is co-founder and chief executive of Arcee AI, an open-weight model lab. His work includes the company’s move from adapting existing models to pretraining its own, and coauthorship of the MergeKit model-merging paper.[1][6][2]
Start with MergeKit →
Links
Arcee AI Hugging Face
GitHub Co-founder and chief executive officer · Arcee AI
Career
McQuade describes an early career at Hugging Face as part of the motivation for Arcee’s open-model work. In an interview with investor Emergence, he explains the company’s decision to pretrain its own models rather than depend on other labs continuing to release competitive weights. Arcee’s later Series B announcement describes funding further model development and tools for deploying and operating those models.[1][4]
Contributions
McQuade coauthored the MergeKit paper, which presents an open-source toolkit for combining the parameters of existing model checkpoints. The method works with trained models rather than requiring a fresh training run for every combination. MergeKit provides multiple merging strategies and supports experiments with task-specialized checkpoints. A merged model still needs evaluation: combining weights does not guarantee that every source model’s strengths will be retained.[6][7]
Ideas
McQuade argues that open weights let organizations adapt and operate AI as an asset. That position underlies Arcee’s emphasis on models that can run on infrastructure chosen by the customer. It is a business and engineering argument about control and dependence, rather than evidence that every open model is cheaper or better for every workload.[1]
Tools, writing & talks
Tools
Start here
MergeKitgithub.comAn open-source toolkit for combining trained language-model checkpoints through multiple merging methods.
Start with the explanation of merging methods and model compatibility. You need Python and familiarity with model checkpoints; evaluate the result on your tasks rather than assuming combined weights preserve every capability.