CrowdTensor is an Apache-2.0 open-source protocol for checkpointed volunteer model-training campaigns that pins immutable model and data revisions to give admitted CPU, GPU, or TPU cells bounded work. The project has released a Beta enrollment process alongside a Draft RFC for a Qwen2.5-7B GSM8K campaign.
A completed feasibility run using Qwen/Qwen2.5-7B-Instruct and openai/gsm8k demonstrated the protocol's capabilities with 256 real LoRA/SFT optimizer steps and 262,144 non-padding tokens. The test included a complete T4x2 worker replacement at step 128, resulting in a normalized exact match improvement from 71.875% to 74.219%. A smaller Founding SmolLM2/WikiText-2 Campaign is now serving live aggregate status, seeded through public HTTPS contribution paths.
The project seeks review of the Draft 7B RFC, specifically regarding the fresh-holdout design, step extension rules, and delta validation policies. Enrollment remains controlled, and the authors note that independent physical multi-host evidence and permissionless adversarial safety are still lacking.