Skip to main content

Module inference

Module inference 

Source
Expand description

Inference invoke seam — exposes semantic skills, grounding, and post-turn verification.

Future crate: qualia-inference.

Functions§

conditioning_activate
Conditioning.activate — move the active registry pointer after approval gates.
conditioning_compile
Conditioning.compile — compile against an explicit backend capability and authority snapshot.
conditioning_evaluate
Conditioning.evaluate — start a bounded evaluation campaign.
conditioning_inspect
Conditioning.inspect — return a redacted requirement/token/evidence trace.
conditioning_rollback
Conditioning.rollback — restore a prior approved version.
conditioning_validate
Conditioning.validate — validate one profile and return structured violations.
constrained_decode
Inference.constrained_decode — run DOMINO constrained decoding on a token sequence.
detect_ungrounded
Inference.detect_ungrounded — check if a generation output is ungrounded. Returns a boolean and optional reason.
embed
Inference.embed — embed text into a vector using the default TextEmbedder.
gelu
Inference.gelu — { x: [f64] } → { out: [f64] }.
grounding
Inference.grounding — check whether a generation is grounded against the quant graph fact store. Returns a record with text, repaired, and optional reason.
load_model
Inference.load_model — load a GGUF/P64 model into resident memory.
relu
Inference.relu — { x: [f64] } → { out: [f64] }.
rms_norm
Inference.rms_norm — { x, weight: [f64], eps? } → { out: [f64] }.
run_classifier
Inference.run_classifier — run a statistical classifier (KNN, NB, or SVM).
run_reranker
Inference.run_reranker — rerank candidate documents by relevance to a query.
run_transformer
Inference.run_transformer — run a transformer forward pass.
sigmoid
Inference.sigmoid — { x: [f64] } → { out: [f64] }.
softmax
Inference.softmax — { x: [f64] } → { out: [f64] } (sums to 1).
unload_model
Inference.unload_model — drop the resident model mmap. Native only — returns E300 on WASM.
vector_search
Inference.vector_search — search the vector store for nearest neighbours.
verify_turn
Inference.verify_turn — verify and heal a completed generation turn. Returns the final text, repair status, and individual checks.