MERIT: Benchmark Results for NLP Model Evaluation and Token Cost

Aggregated numerical source data supporting a manuscript's figures and tables. The dataset includes per-dataset and per-method evaluation results, quality metrics, and token cost tables. It was authored by Shuai Yu and last updated on June 2, 2026.

Use Cases

• Compare full-read versus MERIT evaluation results across multiple datasets based on the standardized splits mentioned.
• Analyze per-method quality metrics for different NLP models or techniques.
• Calculate and compare token costs across different evaluation stages or methods.
• Reproduce or verify the figures and tables from the associated manuscript.

Strengths

• Data is directly linked to a research manuscript, providing traceability for figures and tables.
• Includes multiple evaluation regimes (full-read vs. MERIT) as documented in the README.
• File size is 13.2 MB, indicating a manageable download for analysis.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The full standardized corpora (tens of GB) are not included and must be reconstructed from public sources.

Provenance

Source

figshare, author Shuai Yu

Collection Method

Aggregated from the evaluation of multiple datasets, reconstructed from public sources via a code repository.

Freshness

Last updated 2026-06-02 12:08:13; freshness should be verified.