ritual-dapp-onnx

On-chain machine learning inference is executed by calling the ONNX precompile directly from smart contracts at address 0x0000000000000000000000000000000000000800. This provides synchronous, deterministic results during block execution without relying on off-chain services.

The ONNX precompile is a network-level endpoint at address 0x0000000000000000000000000000000000000800 that enables deterministic ML inference directly on-chain. It processes RitualTensor-encoded inputs and returns results synchronously within the same transaction.

FLOAT32 tensors must be encoded using the RitualTensor format via the provided encoding routines in TypeScript and Solidity. This encoding standardizes input and output parameters for the ONNX precompile to process classification, regression, or embedding tasks.

Yes, dApps can perform synchronous ML inference without off-chain services by calling the ONNX precompile. Smart contracts receive inference results in the same transaction, enabling real-time on-chain decisions for scoring and classification tasks.

The ONNX precompile requires inputs encoded in RitualTensor format and direct ABI calls to a fixed precompile address. It is limited to models compatible with ONNX and FLOAT32 data encoding, requiring parameter handling via provided Solidity and TypeScript routines.