Module cm_model 

ContextModel – flexible context model using ContextMap + StateMap.

Phase 4+: Three CM model variants, each now producing 3 predictions:

1. StateMap prediction (p1): existing state->probability mapping
2. Run-count prediction (p2): consecutive same-bit count in this context
3. Byte-history prediction (p3): last byte seen in this context predicts current bit

The run-count uses the same bit-level hash as the StateMap. The byte-history uses a separate byte-level hash (independent of partial byte c0) so that the stored byte is accessible at any bpos within the same byte context.

Structs

AssociativeContextModel

A context model using 2-way set-associative ContextMap. Best for order-5+ where collision rates are highest.

ChecksumContextModel

A context model using ChecksumContextMap for reduced collision damage.

ContextModel

A context model backed by a ContextMap (hash->state) + StateMap (state->prob). Now also produces run-count and byte-history predictions.

Type Aliases

DualPrediction

Dual prediction from a context model. (state_p, run_p) – both 12-bit probabilities in [1, 4095].