CoolFace
Datasetpublic

guhhhgu/harmonicbench-planir-main-full

HARMONICBench PlanIR Main Full Export This dataset is a normalized Hugging Face export of the local outputs/fixed/main/domain1-domain5 PlanIR artifacts. Coverage Total domains: 5 Total samples: 500 Total condition-plan rows: 7254 Total reference-plan rows: 500 Total domain5 image-description rows: 100 Per-domain coverage: domain1: 100 samples, 1441 condition-plan rows domain2: 100 samples, 1557 condition-plan rows domain3: 100 samples, 1658 condition-plan rows… See the full description on the dataset page: https://huggingface.co/datasets/guhhhgu/harmonicbench-planir-main-full.

sourceHugging Faceotherupdated 5mo agoView on Hugging Face
0likes9downloads
Dataset Card

HARMONICBench PlanIR Main Full Export

This dataset is a normalized Hugging Face export of the local outputs/fixed/main/domain1-domain5 PlanIR artifacts.

Coverage

  • —Total domains: 5
  • —Total samples: 500
  • —Total condition-plan rows: 7254
  • —Total reference-plan rows: 500
  • —Total domain5 image-description rows: 100

Per-domain coverage:

  • —domain1: 100 samples, 1441 condition-plan rows
  • —domain2: 100 samples, 1557 condition-plan rows
  • —domain3: 100 samples, 1658 condition-plan rows
  • —domain4: 100 samples, 1298 condition-plan rows
  • —domain5: 100 samples, 1300 condition-plan rows

Tables

  • —tables/condition_plans.parquet: one row per (domain_id, sample_id, condition_id) with plan JSON, rendered plan text, admissibility/roundtrip metadata, and scalar condition fields.
  • —tables/samples.parquet: one row per sample/run with canonical instance content, manifests, stage summaries, prompt/context text, and source metadata.
  • —tables/reference_plans.parquet: one row per sample with reference planning artifacts such as good_condition_plan, p_dagger.merged, validation, and raw calls.
  • —tables/image_descriptions.parquet: text-only domain5 image descriptions. Raw image files are not included.
  • —tables/domain_stats.parquet: compact per-domain counts and condition-id coverage.
  • —tables/dataset_summary.json: top-level export summary.

Viewer configs

  • —default: preview-friendly condition-plan rows across all 5 domains.
  • —samples_preview: sample-level metadata with shortened prompt/context fields.
  • —reference_plans_preview: sample-level reference-plan summaries.
  • —image_descriptions: flattened text-only domain5 image descriptions.
  • —domain_stats: per-domain coverage counts.

Notes

  • —Domain5 raw image inputs are intentionally excluded. Local image attachment paths are sanitized to compact identifiers, and text image descriptions are stored separately.
  • —Hugging Face preview uses the lighter viewer/* parquet files. The full-fidelity parquet exports remain under tables/* for download and offline analysis.
  • —Upstream source material has mixed provenance and licensing. Inspect fields such as source_name, source_split, source_instance_id, and any available license metadata before redistribution or training use.
  • —The default dataset viewer config points to viewer/condition_plans_preview/train.parquet so the page can render quickly while still covering all domains and all per-sample plan variants.

Example

python
from datasets import load_dataset

plans = load_dataset("guhhhgu/harmonicbench-planir-main-full", split="train")
print(plans[0]["domain_id"], plans[0]["sample_id"], plans[0]["condition_id"])