inside-out-replication-v2-judge-labels

Judge labels for deduplicated sampled answers. Method is exact_match where the normalized answer equals gold, else a Qwen2.5-14B per-relation CoT judge producing grade A/B/C/D.

Dataset Info

• Rows: 2901127
• Columns: 9

Columns

Column	Type	Description
question_id	Value('string')	Question identifier
answer	Value('string')	Deduplicated answer string (full, never truncated)
count	Value('int64')	How many of the 1000 samples produced this exact answer
label	Value('string')	CORRECT or INCORRECT
method	Value('string')	exact_match or judge
grade	Value('string')	Judge grade A/B/C/D (A/B map to CORRECT/INCORRECT)
model	Value('string')	Subject model whose answers were judged
relation	Value('string')	Wikidata relation
split	Value('string')	dev or test

Generation Parameters

{
  "script_name": "03_run_judge.py",
  "model": "Qwen/Qwen2.5-14B-Instruct (judge)",
  "description": "Judge labels for deduplicated sampled answers. Method is exact_match where the normalized answer equals gold, else a Qwen2.5-14B per-relation CoT judge producing grade A/B/C/D.",
  "input_datasets": [
    "inside-out-replication-v2-samples"
  ],
  "experiment_name": "inside-out-replication-v2",
  "job_id": "mll:27608-27621",
  "cluster": "mll",
  "artifact_status": "final",
  "canary": false,
  "hyperparameters": {}
}

Usage

from datasets import load_dataset

dataset = load_dataset("latkes/inside-out-replication-v2-judge-labels", split="train")
print(f"Loaded {len(dataset)} rows")

---