Model Lifecycles
How are models uploaded, registered, and stored in DeepNode?
Models are uploaded as drafts through the portal or API by Model Creators. Drafts can be updated until the Miner is ready to register the model. Upon registration, the model is minted as an ERC-1155 NFT, which becomes its permanent identifier in the system. Metadata (description, category, costs, and other attributes) is stored on IPFS, while references and lifecycle events are anchored on-chain. This provides decentralized storage, immutability, and full discoverability through the Models Marketplace.
How are models validated before being listed?
Before a model becomes available to Consumers, it must pass a multi-stage validation process:
Metadata validation — Validators confirm completeness and accuracy of metadata.
Test deployment — The model is deployed on a Node in a test environment.
Functional validation — Validators run test cases to confirm correctness, performance, and security.
If validation passes, the model is promoted to production and listed in the marketplace. If it fails, the Miner must update and redeploy until success.
How are updates, versioning, and retirement handled?
Model Creators can update models by submitting new metadata or deploying an improved version. Each update goes through validation before replacing the previous version. Versioning is tracked through the NFT identifier and associated on-chain events. If a Miner decides to discontinue a model, they can trigger the Remove Model flow, which burns the NFT and halts further usage. Consumers can still view historical usage data, but the model becomes unavailable for new requests.
How does DeepNode handle malicious or low-quality models?
DeepNode uses Validators and dynamic trust weights to identify malicious or low-quality models early in the validation phase. In addition:
Models failing test cases are rejected before listing.
On-chain rating mechanisms (driven by user feedback and request outcomes) continuously adjust model reputation.
Validators and Miners can report errors that reduce a model’s score.
Models flagged as harmful, non-functional, or violating policies may be retired through governance or forcibly removed.
These safeguards ensure that only reliable, safe, and high-quality models remain accessible to Consumers.
Last updated