Datasets:

xinshuo
/

test_jsonl

@@ -1,4 +1,10 @@
 ---
 configs:
 - config_name: msb_type
   data_files:
@@ -15,213 +21,134 @@ configs:
   default: true
 ---
-# Scientific Coding Benchmark Dataset
-## Dataset Description
-This dataset contains scientific coding questions and benchmarks from multiple sources, organized into three configurations:
-- **msb_type** (244 samples): Multi-source Scientific Benchmark questions focusing on software engineering tasks
-- **et_type** (1,085 samples): Einstein Toolkit related coding questions with extensive documentation
-- **scicobench_all** (1,329 samples, **default**): Combined and unified format of all scientific coding questions
-### Dataset Summary
-The Scientific Coding Benchmark dataset is designed for evaluating AI models on real-world scientific software development tasks. The dataset provides both raw data formats and a unified format for comprehensive evaluation.
-### Supported Tasks
-- **Scientific Code Generation**: Generate code solutions for scientific computing problems
-- **Software Engineering**: Fix bugs and implement features in scientific software
-- **Documentation Understanding**: Work with complex scientific software documentation
-- **Test-Driven Development**: Write code that passes specific test cases
-## Dataset Structure
-### Data Configurations
-The dataset has three configurations with different schemas:
-| Configuration | Samples | Size | Description |
-|---------------|---------|------|-------------|
-| msb_type | 244 | ~15 MB | Software engineering tasks (raw format) |
-| et_type | 1,085 | ~677 MB | Einstein Toolkit tasks (raw format) |
-| **scicobench_all** | 1,329 | ~636 MB | **Unified format (default)** |
-**Default Configuration**: `scicobench_all`
-### Data Loading
 ```python
-from datasets import load_dataset
-# Load default configuration (scicobench_all - unified format)
-dataset = load_dataset("xinshuo/test_jsonl")
-print(f"Loaded {len(dataset['train'])} samples")
-# Load specific configurations
-msb_data = load_dataset("xinshuo/test_jsonl", "msb_type")
-et_data = load_dataset("xinshuo/test_jsonl", "et_type")
-all_data = load_dataset("xinshuo/test_jsonl", "scicobench_all")
-# Access a sample
-print(dataset['train'][0])
 ```
-### Data Fields
-#### Unified Format (scicobench_all)
-Each sample contains:
-- `question_id` (string): Unique identifier for the question
-- `question_type` (string): Type/category of the question (MSB or ET)
-- `description` (string): High-level description of the task
-- `content` (string): Detailed question content and requirements
-- `environment` (string): Environment setup and configuration details
-- `answer` (string): Expected answer or solution
-- `test` (string): Test cases and validation logic
-- `scoring_config` (string): Configuration for scoring the solution
-#### MSB Type Format (msb_type)
-Raw format from Scientific Coding SWE dataset with 33 fields including:
-- `org`, `repo`, `number`, `state`, `title`, `body`
-- `fix_patch`, `test_patch`, `doc_patch`
-- `language`, `instance_id`
-- `fixed_tests`, `run_result`, `test_patch_result`, `fix_patch_result`
-- And more... (see samples for full schema)
-#### ET Type Format (et_type)
-Raw format from Einstein Toolkit with 15 fields including:
-- `thorn_name`, `url`, `configuration`
-- `interface`, `param`, `schedule`
-- `src_filename`, `src_code`
-- `context`, `doc_tex_content`, `readme_content`
-- `combined_doc_context`
-- And more... (see samples for full schema)
-## Source Data
-### MSB Type Questions (244 samples)
-Derived from the Scientific Coding SWE dataset, focusing on:
 - **Organizations**: einsteintoolkit, openmm, pyscf, rdkit, Qiskit, AMReX-Codes
 - **Languages**: C++ (65%), Python (35%)
 - **Tasks**: Bug fixes, feature implementation, code completion
-### ET Type Questions (1,085 samples)
-Einstein Toolkit code generation tasks including:
-- Thorn implementations
-- Interface definitions
-- Parameter configurations
-- Schedule definitions
-- Extensive documentation context
-### SciCoBench All (1,329 samples)
-Unified format combining both question types with standardized structure for comprehensive evaluation.
-## Usage Examples
-### Load and Iterate Through Questions
 ```python
 from datasets import load_dataset
-# Load the unified format (recommended)
-dataset = load_dataset("xinshuo/test_jsonl")
 for sample in dataset['train']:
-    print(f"Question ID: {sample['question_id']}")
     print(f"Type: {sample['question_type']}")
-    print(f"Description: {sample['description'][:100]}...")
-    print("-" * 80)
-```
-### Filter by Question Type
-```python
-# Filter MSB-type questions from the unified dataset
-msb_questions = dataset['train'].filter(lambda x: x['question_type'] == 'MSB')
-print(f"MSB questions: {len(msb_questions)}")
-# Filter ET-type questions
-et_questions = dataset['train'].filter(lambda x: x['question_type'] == 'ET')
-print(f"ET questions: {len(et_questions)}")
-```
-### Work with Raw Formats
-```python
-# Load raw MSB format for detailed metadata
-msb_raw = load_dataset("xinshuo/test_jsonl", "msb_type")
-sample = msb_raw['train'][0]
-print(f"Organization: {sample['org']}")
-print(f"Repository: {sample['repo']}")
-print(f"Fix patch:\n{sample['fix_patch'][:200]}...")
-# Load raw ET format for full documentation
-et_raw = load_dataset("xinshuo/test_jsonl", "et_type")
-sample = et_raw['train'][0]
-print(f"Thorn: {sample['thorn_name']}")
-print(f"Documentation: {sample['combined_doc_context'][:200]}...")
 ```
-## Considerations for Using the Data
-### Use Cases
-- Training code generation models for scientific computing
-- Benchmarking AI coding assistants on scientific software tasks
-- Research in automated scientific software development
-- Education in scientific programming practices
-### Recommended Configuration
-For most use cases, we recommend using the **default `scicobench_all` configuration**, which provides:
-- Unified schema across all questions
-- Standardized fields for easier processing
-- Complete coverage of both MSB and ET question types
-Use the raw configurations (`msb_type` or `et_type`) only if you need:
-- Original metadata and detailed execution results
-- Domain-specific fields
-- Full documentation context
-### Limitations
-- Focus on specific scientific domains (physics, chemistry, quantum computing)
-- Primarily C++ and Python languages
-- May require domain-specific knowledge to evaluate correctly
-- Different configurations have different schemas
-## Additional Information
-### Dataset Curators
-This dataset combines and unifies questions from:
-- xinshuo/Scientific_Coding_SWE_dataset
-- xinshuo/ET_1k
-### Licensing Information
-Please refer to the original datasets for licensing information.
-### Citation Information
 ```bibtex
-@dataset{scicobench,
-  title={Scientific Coding Benchmark Dataset},
-  author={xinshuo},
-  year={2024},
-  url={https://huggingface.co/datasets/xinshuo/test_jsonl}
 }
 ```
-### Contributions
-For questions or issues, please refer to the dataset repository on HuggingFace.

 ---
+license: cc0-1.0
+task_categories:
+- text-generation
+tags:
+- code
+- scientific-computing
 configs:
 - config_name: msb_type
   data_files:
   default: true
 ---
+# AInsteinBench
+**AInsteinBench** is a benchmark for evaluating the capabilities of AI agents in solving scientific computing problems. It currently supports **Einstein Toolkit** and **Multi-SWE-bench** formats of coding questions.
+## 📊 Dataset Overview
+AInsteinBench contains 1,329 code generation problems from two sources:
+| Subset | Samples | Description |
+|--------|---------|-------------|
+| **et_type** | 1,085 | Einstein Toolkit code completion tasks |
+| **msb_type** | 244 | Multi-SWE-bench scientific computing tasks |
+| **scicobench_all** | 1,329 | Combined dataset in unified format (default) |
+### Data Sources
+1. **Einstein Toolkit (et_type)**: 1,085 code completion tasks from the Einstein Toolkit codebase
+2. **Multi-SWE-Bench (msb_type)**: 244 tasks from multi-swe-bench processing of multiple scientific computing repositories
+## 📋 Data Fields
+The unified format (`scicobench_all`) contains:
 ```python
+{
+    "question_id": str,        # Unique identifier
+    "question_type": str,      # "einstein_toolkit" or "multi_swe_bench"
+    "description": str,        # Task description
+    "content": str,            # Full problem context
+    "environment": str,        # Setup requirements
+    "answer": str,             # Expected solution
+    "test": str,               # Test cases
+    "scoring_config": str      # Evaluation configuration
+}
 ```
+For detailed field descriptions and raw formats, see the [AInsteinBench repository](https://github.com/ByteDance-Seed/AInsteinBench).
+## 🔬 Data Curation Process
+### Einstein Toolkit Processing (et_type)
+The Einstein Toolkit is a collection of C/C++/Fortran codes for general relativistic simulations, organized into packages called "Thorns" managed by the Cactus Computation Language (CCL).
+**Problem Definition**: Given an incomplete Thorn (missing one source file), can the model complete it and pass all tests?
+**Data Curation Pipeline**:
+1. Collected ~3,000 source files from open-sourced Einstein Toolkit Thorns
+2. Screened for 1,000 problems with runnable tests
+3. Selected 1,085 problems where models are evaluated on physics correctness, not just compilation
+**Thorn Structure**:
+```
+.
+├── doc/                       # Documentation
+│   └── documentation.tex
+├── src/                       # Source code (one file removed as target)
+│   └── *.c, *.cpp, *.f90
+├── test/                      # Test cases
+│   ├── <test_name>/          # Reference outputs
+│   └── <test_name>.par       # Test parameters
+├── README.md
+├── configuration.ccl          # Dependencies
+├── interface.ccl              # Shared variables/functions
+├── param.ccl                  # Parameters
+└── schedule.ccl               # Execution scheduling
+```
+### Multi-SWE-Bench Format (msb_type)
+244 software engineering tasks from scientific computing repositories:
 - **Organizations**: einsteintoolkit, openmm, pyscf, rdkit, Qiskit, AMReX-Codes
 - **Languages**: C++ (65%), Python (35%)
 - **Tasks**: Bug fixes, feature implementation, code completion
+Data is formatted following the Multi-SWE-Bench structure with issue descriptions, patches, and test cases.
+## 💻 Usage
+### Loading the Dataset
 ```python
 from datasets import load_dataset
+# Load the unified format (default)
+dataset = load_dataset("ByteDance-Seed/AInsteinBench", "scicobench_all")
+# Load specific subsets
+et_dataset = load_dataset("ByteDance-Seed/AInsteinBench", "et_type")
+msb_dataset = load_dataset("ByteDance-Seed/AInsteinBench", "msb_type")
+# Access samples
 for sample in dataset['train']:
+    print(f"ID: {sample['question_id']}")
     print(f"Type: {sample['question_type']}")
+    print(f"Task: {sample['description']}")
 ```
+### Evaluation
+For evaluation scripts and detailed usage, please visit the [AInsteinBench GitHub repository](https://github.com/ByteDance-Seed/AInsteinBench).
+## 📜 License
+The dataset is licensed under CC0, subject to any intellectual property rights in the dataset. The data is adapted from open source projects; your use of that data must comply with their respective licenses.
+**Source Repositories**:
+- Einstein Toolkit: https://einsteintoolkit.org
+- Scientific computing repositories: openmm, pyscf, rdkit, Qiskit, AMReX-Codes, einsteintoolkit
+All source repositories are open source with permissive licenses.
+## 🔗 Links
+- **GitHub Repository**: https://github.com/ByteDance-Seed/AInsteinBench
+- **Paper**: Coming soon
+## 🤝 Citation
+If you use AInsteinBench in your research, please cite:
 ```bibtex
+@dataset{ainsteinbench2025,
+  title={AInsteinBench: A Benchmark for AI Agents in Scientific Computing},
+  author={ByteDance Seed Team},
+  year={2025},
+  publisher={Hugging Face},
+  url={https://huggingface.co/datasets/ByteDance-Seed/AInsteinBench}
 }
 ```