CrowdTensor is a decentralized framework designed for volunteer LoRA training that allows contributors to submit model updates without remaining online for the entire process. The system utilizes a "Campaign" structure where "Cells" perform bounded work units and submit deltas to a Coordinator for validation and aggregation. This architecture enables shared model checkpoint progression across intermittent GPU resources.