microsoft/benchpress-score-matrix

BenchPress Score Matrix

This dataset contains the public model-by-benchmark score matrix used by BenchPress. The release is a tabular artifact: model metadata, benchmark metadata, one row per observed score, and the paper-canonical dense subset used in the BenchPress experiments.

The source repository is microsoft/benchpress. This export was generated from commit 5be3b4eddf0188721ff25f00713b589b2cbed8e0.

Files

File	Contents
data/scores_all.csv / .parquet	All numeric score rows in the audit pool, with source and audit metadata.
data/scores_paper.csv / .parquet	Long-form rows for the paper-canonical matrix.
data/models.csv / .parquet	Model metadata and canonical evaluation settings.
data/benchmarks.csv / .parquet	Benchmark metadata and canonical benchmark settings.
data/score_matrix_paper_wide.csv	Wide model × benchmark matrix for the paper-canonical subset.
metadata.json	Export counts, source commit, and matrix construction metadata.

Quick start

from datasets import load_dataset

scores = load_dataset("microsoft/benchpress-score-matrix", "scores_paper")["train"].to_pandas()
models = load_dataset("microsoft/benchpress-score-matrix", "models")["train"].to_pandas()
benchmarks = load_dataset("microsoft/benchpress-score-matrix", "benchmarks")["train"].to_pandas()

For a complete audit-pool table:

scores_all = load_dataset("microsoft/benchpress-score-matrix", "scores_all")["train"].to_pandas()

Schema

scores_all and scores_paper include:

• model_id, model_name, provider
• benchmark_id, benchmark_name, category, metric
• score
• reference_url, source_type, audit_status, matches_canonical
• reported_setting_json, notes

models and benchmarks include an in_paper_matrix flag that identifies rows retained by the paper-canonical threshold filter.

Matrix construction

The paper-canonical matrix applies the BenchPress construction pipeline: audit-status filtering, canonical representative selection, and the iterative threshold filter. Current export counts:

• audit pool: 189 models, 316 benchmarks, 4903 numeric score rows
• paper matrix: 84 models × 133 benchmarks, 2604 observed cells (23.3% fill)

Caveats

Scores come from heterogeneous public sources: model cards, official blogs, technical reports, benchmark leaderboards, and third-party aggregators. Each row keeps the source URL, source type, audit status, and canonical-setting match flag so downstream users can choose their own filtering policy.