Datasets:

---

lhpku20010120

/

Lite-RSI

Like 0

Adds the normalized Lite-RSI evaluation dataset as a Hugging Face-compatible Parquet shard, together with the dataset card and schema manifest. The release contains 11,273 response records with normalized model names.

Reopening to update the PR with the renamed HLE benchmark configuration and revised dataset card.

Summary

Add the HLE evaluation results to Lite-RSI as the first benchmark configuration.

Changes

• Add the hle dataset configuration with a test split.
• Include 11,273 evaluation responses covering 2,500 HLE tasks and 8 models.
• Store the normalized data in Hugging Face-compatible Parquet format.
• Add consistent model names, evaluation results, task metadata, and integrity hashes.
• Update the dataset card and manifest for future multi-benchmark expansion.

Dataset Structure

data/
└── hle/
    └── test-00000-of-00001.parquet

Loading

from datasets import load_dataset

dataset = load_dataset(
    "lhpku20010120/Lite-RSI",
    "hle",
    split="test",
)

Validation

The dataset was loaded locally with Hugging Face datasets and verified to contain 11,273 rows. Empty completions are retained and marked with has_response=false.