mirror of
https://github.com/hanzoai/chains.git
synced 2026-10-30 18:46:11 +00:00
zeekayandGitHub
96da029e9a
* hanzo-evm: aiquorum + aimarket precompiles, evmllm (geth AI precompile + e2e), operator/canonical (commit/reveal, int8-quant), aiffi bench. Thinking-chain substrate. * hanzo-evm/evm: wire AI precompiles (inference 0x0300..03 + modelregistry) Registers deterministic int8 inference (Tier-1 in-consensus AI) + ModelSpec registry into the Hanzo EVM. threshold dep aligned to v1.9.4 (go.mod match). Builds green (55M binary). aivmbridge follows luxfi/precompile release.
48 lines
2.2 KiB
Go
48 lines
2.2 KiB
Go
// Copyright (C) 2026, Hanzo AI Inc. All rights reserved.
|
|
// See the file LICENSE for licensing terms.
|
|
|
|
package canonical
|
|
|
|
import (
|
|
"github.com/luxfi/crypto"
|
|
"github.com/luxfi/geth/common"
|
|
)
|
|
|
|
// OutputHashRaw is the consensus hash for RAW (free-text) jobs:
|
|
//
|
|
// output_hash = keccak256( deterministic UTF-8 model output bytes )
|
|
//
|
|
// The bytes are exactly what hanzo_ffi_infer returns. Greedy decoding on a fixed
|
|
// engine build is bit-identical for the same (model, prompt) on the same host
|
|
// (proven), so two honest operators produce the same bytes and therefore the
|
|
// same hash. Any divergence in the output bytes — a single different token —
|
|
// yields a different hash and excludes that operator from the quorum group,
|
|
// which is precisely the behavior we want: the chain counts byte-agreement, not
|
|
// semantic agreement.
|
|
func OutputHashRaw(output []byte) common.Hash {
|
|
return common.BytesToHash(crypto.Keccak256(output))
|
|
}
|
|
|
|
// OutputHashGovernance is the consensus hash for GOVERNANCE jobs. It is computed
|
|
// over the CANONICAL STRUCTURED DECISION only — never the free-form rationale:
|
|
//
|
|
// canonical_output_bytes = model_spec_hash(32) || vote_byte(1) || u16be(bucket_bps)(2)
|
|
// output_hash = keccak256( canonical_output_bytes )
|
|
//
|
|
// See governance.go for the full rationale-exclusion justification and the
|
|
// Decision type that produces (vote_byte, bucket_bps). Binding model_spec_hash
|
|
// into the preimage means a decision is only ever equal to another decision made
|
|
// under the IDENTICAL spec — an operator cannot have its "yes under spec A"
|
|
// counted toward a quorum running spec B.
|
|
func OutputHashGovernance(d Decision) common.Hash {
|
|
return common.BytesToHash(crypto.Keccak256(d.consensusPreimage()))
|
|
}
|
|
|
|
// EmptyEmbeddingHash is the embedding_hash value a job carries when it has no
|
|
// embedding component. The precompile treats embedding_hash as an opaque 32-byte
|
|
// field bound into the commit, so a job without embeddings binds the zero hash;
|
|
// all operators in such a job use this same value and it drops out of any
|
|
// disagreement. (It is the zero hash, stated explicitly so call sites read
|
|
// clearly instead of constructing a bare common.Hash{}.)
|
|
var EmptyEmbeddingHash = common.Hash{}
|