MultimodalMathBenchmarks

This file describes the canonical dataset release to publish from this repository to cjerzak/MultimodalMathBenchmarks. It covers the public benchmark datasets and their modality assets.

Canonical Upload Manifest

Upload the following local sources from SavedData/ to the dataset repo root on Hugging Face:

HF path	Local source	Count	Purpose
SharedMultimodalGrid.csv	SavedData/SharedMultimodalGrid.csv	10,000 rows	Canonical metadata table for the paired multimodal multiplication benchmark
TextFiles/	SavedData/TextFiles/	10,000 .txt files	Text renderings for mm_* benchmark items
Images/	SavedData/Images/	10,000 .png files	Image renderings for mm_* benchmark items
AudioFiles/	SavedData/AudioFiles/	10,000 .mp3 files	Audio renderings for mm_* benchmark items
HDSv2.csv	SavedData/HDSv2.csv	1,000 rows	Canonical heuristic-disagreement probe set
HDSv2Images/	SavedData/HDSv2Images/	144 .png files	Image renderings for the HDSv2 test split only
Trapsv2.csv	SavedData/Trapsv2.csv	30 rows	Canonical adversarial trap set
Trapsv2Images/	SavedData/Trapsv2Images/	30 .png files	Image renderings for every trap item

Dataset Summary

The release contains three benchmark families:

1. SharedMultimodalGrid.csv

   • 10,000 shared multiplication problems paired across text, image, and audio.
   • Split counts: train=7026, val=1416, test=1558.
   • Each row uses an mm_XXXXX ID that maps directly to:
     • TextFiles/mm_XXXXX.txt
     • Images/mm_XXXXX.png
     • AudioFiles/mm_XXXXX.mp3

2. HDSv2.csv

   • 1,000 heuristic-disagreement problems for fingerprinting and probe-style evaluation.
   • Split counts: train=701, val=155, test=144.
   • HDSv2Images/ contains exactly the 144 test-split item IDs from HDSv2.csv.

3. Trapsv2.csv

   • 30 adversarial trap problems designed to target heuristic-specific failures.
   • No split column; all rows are held-out trap items.
   • Trapsv2Images/ contains one PNG for every row in Trapsv2.csv.

Dataset Structure

Shared Multimodal Grid

Canonical table: SharedMultimodalGrid.csv

Columns:

id, a, b, a_times_b, template_a, template_b, digits_a, digits_b, nonzero_a, nonzero_b, digit_total, nonzero_total, complexity_c, stratum_id, split

Notes:

• id is the stable benchmark ID, e.g. mm_00001.
• a, b, and a_times_b are the multiplication operands and exact product.
• template_a and template_b record the rendering template choices used for the paired benchmark export.
• digit_total, nonzero_total, and complexity_c support difficulty analyses.
• split is the deterministic train/val/test assignment used by the pipeline.

HDSv2

Canonical table: HDSv2.csv

Columns:

id, a, b, product, design_family, canonical_target_heuristic, canonical_target_margin, ot_cost, dd_cost, rc_cost, heuristic_definition_version, target_heuristic, ot_score, dd_score, rc_score, category, notes, digit_total, nonzero_total, complexity_c, split

Notes:

• design_family is the construction family used to generate the item.
• canonical_target_heuristic is the v2 cost-model winner.
• target_heuristic, ot_score, dd_score, and rc_score are retained as legacy compatibility fields.
• heuristic_definition_version identifies the versioned heuristic labeling scheme.

Trapsv2

Canonical table: Trapsv2.csv

Columns:

id, a, b, product, trap_type, design_family, canonical_target_heuristic, heuristic_definition_version, target_heuristic, expected_error_type, notes, digit_total, nonzero_total, complexity_c

Notes:

• trap_type names the adversarial construction family.
• expected_error_type records the characteristic failure pattern the trap is intended to expose.
• In the checked-in v2 export, design_family, canonical_target_heuristic, and target_heuristic align by construction.

Modality and Evaluation Notes

• The paired multimodal benchmark includes text, image, and audio assets.
• The current Qwen experiments in this repository evaluate text and image, not audio.
• HDSv2Images/ is intentionally a test-only image release for probe evaluation.
• Trapsv2Images/ covers the full trap set.

Release Notes

• The Hugging Face dataset repo is currently configured with the gpl-2.0 license.