GSM8K contamination (OpenCompass PPL probe)

OpenCompass perplexity probe comparing GSM8K train, GSM8K test, and a Skywork mock set to flag training-set overlap, not math accuracy.

Also known as: gsm8k_contamination_ppl, gsm8k-train-ppl, gsm8k-test-ppl, gsm8k-ref-ppl, mock_gsm8k_test

unassessed

This page is a discovery lead. Nobody has yet assessed it against the catalogue contract, so it carries no disposition. Absence of evidence here is not evidence of staleness.
Categorymath
SubcategoryOpenCompass perplexity comparison of GSM8K train, test, and a mock reference set
Page statusactive
Metricaverage_ppl
Directionlower_is_better
Dataset licenceMixed: GSM8K MIT; mock_gsm8k_test Skywork Community License (license:other)
PublisherOpenCompass (harness); Skywork / Kunlun Inc. (mock set and method); OpenAI (GSM8K)

What it measures

gsm8k_contamination is not a grade-school math test. OpenCompass feeds concatenated question-and-answer text to PPLOnlyInferencer and reports average_ppl on three corpora: the GSM8K training split, the GSM8K test split, and a reference file loaded from ./data/gsm8k-extra/mock_gsm8k_test.jsonl. Skywork (Wei et al., 2023, §5.2) treat the mock set as GSM8K-like text that should not have been in any model's training data. Lower test PPL than mock PPL is read as possible test leakage. Lower train PPL than test PPL is read as possible overfitting on the train split. The skill is distributional familiarity, not solving word problems.

Task format

Zero-shot perplexity. GSM8K splits use template "{question} {answer}" on GSM8KDataset from ./data/gsm8k. The reference uses template "{text}" on JsonlDataset. Evaluator AveragePPLEvaluator. Abbreviations gsm8k-train-ppl, gsm8k-test-ppl, gsm8k-ref-ppl. No numeric answer is extracted.

Models reporting this benchmark

No model card in ModelSpec reports this benchmark yet.

Data

This page as JSON · Edit on GitHub