1 vin.qh_bundle

vin.qh_bundle

Persistence-neutral identity for one immutable Q_H inference bundle.

1.1 Classes

Name Description
QhInferenceBundleRef Content-bound location of one immutable Q_H inference bundle.
QhInferenceRuntime Verified scorer plus the immutable identities required by inference callers.

1.1.1 QhInferenceBundleRef

vin.qh_bundle.QhInferenceBundleRef(bundle_path, schema_version, manifest_sha256)

Content-bound location of one immutable Q_H inference bundle.

1.1.1.1 Attributes

Name Description
bundle_path Absolute directory containing the verified manifest and payloads.
schema_version Closed bundle schema expected by the loader.
manifest_sha256 SHA-256 of the canonical manifest payload excluding its own digest field.

1.1.2 QhInferenceRuntime

vin.qh_bundle.QhInferenceRuntime(
    scorer,
    bundle_manifest_sha256,
    scorer_state_sha256,
    scorer_config_hash,
    implementation_sha256,
    actor_state_contract_hash,
    learning_contract_hash,
    target_protocol,
    candidate_config_hashes,
    action_mask_semantics,
    representation_semantics,
    trained_horizons,
)

Verified scorer plus the immutable identities required by inference callers.

1.1.2.1 Attributes

Name Description
scorer Evaluation-mode scorer loaded from the verified scorer-state artifact.
bundle_manifest_sha256 Digest of the verified bundle manifest.
scorer_state_sha256 Digest of the verified scorer-state artifact.
scorer_config_hash Digest binding the scorer configuration to its serialized state.
implementation_sha256 Digest of the complete importable aria_nbv package tree.
actor_state_contract_hash Exact fit-admission digest of actor representation and complete TRAIN provenance.
learning_contract_hash Digest of replay and fitted-Q learning semantics.
target_protocol Target-input protocol admitted by the learning contract.
candidate_config_hashes Candidate-generation identities admitted by the learning contract.
action_mask_semantics Meaning of actor action masks for this runtime.
representation_semantics Meaning of the actor representation consumed by the scorer.
trained_horizons Sorted scalar horizons with positive manifest-bound training support.
Back to top