LLMVeriOpt Datasets for CGO 2026 Artifact Evaluation

These datasets contain all test sets and training-derived evaluation inputs used in the CGO 2026 submission LLM-VeriOpt: Verification-Guided ReinforcementLearning for LLM-Based Compiler OptimizationThey include: correctness-oriented IR transformation datasets latency-optimized IR datasets SFT training prompt datasets one-shot baseline datasets diagnostic augmented misclassification datasets All datasets are stored in HuggingFace datasets format (Arrow + metadata).These datasets are required to run the evaluation pipelines provided in the artifact repository. The corresponding code, models, and figure reproduction scripts are available at:https://github.com/carrotProgrammer/llmveriopt-AE These datasets serve as the canonical reference for reproducibility and are required for running the artifact’s sampling and full-evaluation pipelines.