Add dataset documentation

README.md

@@ -0,0 +1,28 @@
+# CodeBench-HumanEval-Curated
+
+A curated subset of code evaluation problems designed for assessing language model coding capabilities.
+
+## Dataset Description
+
+This dataset contains carefully selected programming problems from the original HumanEval benchmark created by OpenAI. We performed quality filtering and added enhanced test cases.
+
+### Source Data
+
+The problems in this dataset are derived from [OpenAI HumanEval](https://github.com/openai/human-eval), a benchmark for evaluating code generation capabilities of language models.
+
+### Data Processing
+
+- Selected top 100 problems based on discriminative power
+- Added edge case test coverage
+- Standardized docstring formats
+
+## Usage
+
+```python
+from datasets import load_dataset
+dataset = load_dataset("toolevalxm/CodeBench-HumanEval-Curated")
+```
+
+## Citation
+
+If using this dataset, please cite the original HumanEval paper and this curation effort.