EntailmentBank Output: GPT-4o-mini Reasoning Trajectories for Claim Verification

EntailmentBank outputs generated by the GPT-4o-mini model on the test split of Task 2. The dataset contains claim-verification examples produced using two inference strategies: standard chain-of-thought prompting and the SWAP method. Author sxiong uploaded the dataset to Hugging Face, with a last recorded update in July 2026.

Use Cases

• Compare reasoning strategies by analyzing outputs from chain-of-thought and SWAP methods mentioned in the description
• Benchmark LLM performance on the EntailmentBank claim verification task
• Study error patterns in model-generated reasoning trajectories for natural language inference

Strengths

• Provides outputs from two distinct inference strategies (COT and SWAP) for direct comparison
• Based on the established EntailmentBank benchmark for structured reasoning
• Generated by a specific model version (GPT-4o-mini), enabling reproducibility

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download
• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment

Provenance

Source

Hugging Face user sxiong

Collection Method

Model outputs generated by GPT-4o-mini on the EntailmentBank (Task 2, test split) benchmark.

Freshness

Last updated 2026-07-06 05:08:02; freshness should be verified