Hugging Face's logo

Datasets:
Cosmoscd
/
AccelEval

Size:
1B<n<10B
Tags:
benchmark
cuda
gpu
code-generation
orbench
License:
Dataset card Files
xet
 Community
Dataset Viewer
The dataset viewer is not available for this subset.
Cannot get the split names for the config 'default' of the dataset.
Exception:    SplitsNotFoundError
Message:      The split names could not be parsed from the dataset config.
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.12/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
                  for split_generator in builder._split_generators(
                                         ^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 82, in _split_generators
                  raise ValueError(
              ValueError: The TAR archives of the dataset should be in WebDataset format, but the files in the archive don't share the same prefix or the same types.
              
              The above exception was the direct cause of the following exception:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 65, in compute_split_names_from_streaming_response
                  for split in get_dataset_split_names(
                               ^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
                  info = get_dataset_config_info(
                         ^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
                  raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
              datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

ORBench Test Data

Test data for ORBench — a benchmark for evaluating LLMs on CPU-to-CUDA code acceleration.

This repo hosts only the input data + expected outputs. The benchmark harness (CPU references, task_io adapters, evaluation scripts) lives at github.com/YOURNAME/ORBench.

Contents

File Size Description
small.tar.gz ~100 MB Small inputs — for smoke tests (43 tasks)
medium.tar.gz ~3.5 GB Medium inputs — main leaderboard
large.tar.gz ~8 GB Large inputs — stress test
manifest.json — SHA256 + task metadata

Each tarball extracts to <task>/data/<size>/ containing:

  • input.bin — binary input tensors (ORBench v2 format)
  • expected_output.txt — reference output from CPU baseline
  • cpu_time_ms.txt — CPU baseline wall time
  • requests.txt — per-call queries (if applicable)

Usage 

# Clone the harness repo
git clone https://github.com/YOURNAME/ORBench.git
cd ORBench

# Install HF client
pip install huggingface_hub

# Download data
python3 scripts/download_data.py small       # smoke
python3 scripts/download_data.py medium      # main leaderboard
python3 scripts/download_data.py large       # stress test
python3 scripts/download_data.py all         # everything

# Run the benchmark
python3 -m framework.run_all_tasks \
    --models gemini-3.1-pro-preview-openrouter \
    --levels 3 --sizes medium --yes

Verification 

python3 scripts/download_data.py medium --verify

Citation

If you use ORBench, please cite:

@misc{orbench2026,
  title={ORBench: Evaluating LLMs on CPU-to-CUDA Code Acceleration},
  author={...},
  year={2026},
  url={https://github.com/YOURNAME/ORBench}
}
Downloads last month
12
Total file size:
3.06 GB