Room 0 · test key· reading the room…
The private model
One small model, trained inside the room on everyone's text negatives: a LoRA adapter (rank 8) on Qwen2.5 0.5B Instruct, retrained each week on CPU. Its weights never leave the room. Every call pays 100% to the people whose text trained it.
BaseQwen2.5 0.5B Instruct
AdapterLoRA, rank 8, weekly
Price0.002 USDC a call
To owners100%
Training data0 / 25 text negatives
First training as soon as 25 text negatives are uploaded. Trap rows and deleted negatives are never trained on.
Call it once it trainsx402
POST /api/model
{"prompt":"Write a dream about a lighthouse"}
→ 402 · 0.002 USDC