CrowdTensor: Volunteer LoRA Training for Intermittent GPUs
CrowdTensor enables volunteer LoRA training that progresses on intermittent GPUs. Beta enrollment is now open.
CrowdTensor is designed to enable ordinary machines to advance a shared model checkpoint without requiring every contributor to remain online. The project focuses on local LoRA updates, allowing users to participate without continuous connectivity. A public campaign has launched on SmolLM2-135M/WikiText-2, with beta enrollment open for limited contributions.
This synthesis was produced from its source by AI; there is no human editor or manual review step. How we work