mac-test/modular-arithmetic-residue-net

Modular Arithmetic DLP Table Baseline

This repository contains an original compact DLP-table baseline for the SAIR Modular Arithmetic Challenge.

The model is designed to improve the earlier residue-table baseline while keeping the artifact relatively small. Each operand is independently encoded as an integer representation. For each supported prime, weights.bin stores learned log and exponent tables that let the model emit a single base-p digit.

This version supports every prime up to 16381. It should preserve Tier 1 and Tier 2 behavior and improve a substantial part of Tier 3. For larger unsupported primes, the model intentionally falls back to [0].

This is still a compact baseline rather than a high-tier neural solution. It is intended to provide stronger small-prime behavior while keeping the submitted implementation original and avoiding copied official example code.

Files:

• —manifest.json: submission manifest
• —model.py: DlpTableBaseline implementation
• —weights.json: table metadata
• —weights.bin: learned log and exponent tables