Module file_loader 

Model File Loader

Loads a .tern.bin ModelCoherence file produced by transmute_llama.py and maps its ternarized layers into a 13-expert ExpertBank for EPIS inference.

Layer → Expert mapping

Round-robin: layer i → expert (i % 13). When multiple layers land on the same expert they are fused by majority vote: agreement → keep the value, disagreement → 0 (hold).

Dimension handling

Large transformer layers (e.g. 2048×2048) are truncated to EXPERT_DIMS. Layers smaller than EXPERT_DIMS are zero-padded.

Structs

ModelFileInfo

Metadata returned alongside a loaded ExpertBank.

Constants

EXPERT_INPUT_DIM

EXPERT_OUTPUT_DIM

Functions

load_expert_bank

Load a .tern.bin file and construct a real ExpertBank13 from its layers.