Prompt Variations and LLM Responses for Stability-Generalization Score Evaluation

Prompt Variations and LLM Responses contains prompt variants and model outputs used to evaluate the Stability-Generalization Score (SGS). The dataset was created by author naghamo and was last updated on June 5, 2026. It includes data from six QA and instruction benchmarks, such as TruthfulQA and Natural Questions, and covers responses from eleven large language models.

Use Cases

• Benchmarking LLM robustness based on stylistic prompt perturbations mentioned in the description
• Analyzing response consistency across open-source and closed-source models based on the described evaluation
• Studying the Stability-Generalization Score (SGS) metric across different question-answering tasks

Strengths

• Includes responses from eleven large language models, comprising eight open-source and three closed-source models
• Covers six distinct QA and instruction benchmarks, including TruthfulQA and Natural Questions
• Applies six families of stylistic perturbations to prompts for evaluation

Limitations

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

Provenance

Source

huggingface

Collection Method

Likely compiled from model inference outputs on benchmark datasets under controlled prompt variations.

Freshness

Last updated 2026-06-05 09:07:18; freshness should be verified