Download croissant.json from bcui2/HPC-Bench: direct link, hf CLI and curl.
- Browser
- Download file 46.3 kB
-
https://huggingface.co/datasets/bcui2/HPC-Bench/resolve/main/croissant.json
- Command line
-
hf download hf://datasets/bcui2/HPC-Bench/croissant.json
-
curl -L -o croissant.json https://huggingface.co/datasets/bcui2/HPC-Bench/resolve/main/croissant.json
46.3 kB
{
"@context": {
"@language": "en",
"@vocab": "https://schema.org/",
"citeAs": "cr:citeAs",
"column": "cr:column",
"conformsTo": "dct:conformsTo",
"cr": "http://mlcommons.org/croissant/",
"rai": "http://mlcommons.org/croissant/RAI/",
"data": {
"@id": "cr:data",
"@type": "@json"
},
"dataType": {
"@id": "cr:dataType",
"@type": "@vocab"
},
"dct": "http://purl.org/dc/terms/",
"examples": {
"@id": "cr:examples",
"@type": "@json"
},
"extract": "cr:extract",
"field": "cr:field",
"fileProperty": "cr:fileProperty",
"fileObject": "cr:fileObject",
"fileSet": "cr:fileSet",
"format": "cr:format",
"includes": "cr:includes",
"isLiveDataset": "cr:isLiveDataset",
"jsonPath": "cr:jsonPath",
"key": "cr:key",
"md5": "cr:md5",
"parentField": "cr:parentField",
"path": "cr:path",
"recordSet": "cr:recordSet",
"references": "cr:references",
"regex": "cr:regex",
"repeated": "cr:repeated",
"replace": "cr:replace",
"sc": "https://schema.org/",
"separator": "cr:separator",
"source": "cr:source",
"subField": "cr:subField",
"transform": "cr:transform",
"wd": "https://www.wikidata.org/wiki/",
"prov": "http://www.w3.org/ns/prov#"
},
"@type": "sc:Dataset",
"name": "HPC-Bench",
"description": "HPC-Bench is a benchmark for evaluating large language models as optimizers of computational kernels embedded in complete HPC programs. This dataset provides the standardized inputs at five scales, the reference outputs used for correctness checking, the exact instrumented baseline programs, and a structured index for 92 CPU workloads (serial and OpenMP settings) and a Rodinia-derived CUDA set, drawn from PolyBench/C, Rodinia, Parboil, MachSuite, MiBench, PARSEC, and Codee and spanning 16 computational motifs. The evaluation pipeline, LLM-generated candidates, and per-trial results are in the GitHub repository.",
"conformsTo": "http://mlcommons.org/croissant/1.1",
"url": "https://huggingface.co/datasets/bcui2/HPC-Bench",
"sameAs": "https://github.com/Deep-Learning-Profiling-Tools/HPC-Bench",
"creator": [
{
"@type": "sc:Person",
"name": "Bowen Cui"
},
{
"@type": "sc:Person",
"name": "Junyu Yin"
},
{
"@type": "sc:Person",
"name": "Tejas Ramesh"
},
{
"@type": "sc:Person",
"name": "Leo Lim"
},
{
"@type": "sc:Person",
"name": "Oscar Hernandez"
},
{
"@type": "sc:Person",
"name": "Keren Zhou"
}
],
"license": [
"https://opensource.org/licenses/MIT",
"https://github.com/Deep-Learning-Profiling-Tools/HPC-Bench/blob/master/docs/BENCHMARKS.md"
],
"citeAs": "@article{cui2026hpc,\n title={HPC-Bench: A Comprehensive Benchmark for High Performance Computing Codes},\n author={Cui, Bowen and Yin, Junyu and Ramesh, Tejas and Lim, Leo and Hernandez, Oscar and Zhou, Keren},\n journal={arXiv preprint arXiv:2611.02752},\n year={2026}\n}",
"version": "1.0.1",
"datePublished": "2026-10-31",
"dateModified": "2026-11-06",
"keywords": [
"HPC",
"benchmark",
"code optimization",
"large language models",
"OpenMP",
"CUDA",
"performance"
],
"rai:dataCollection": "Benchmark programs were selected from seven established HPC benchmark suites. For each program, a profiler (HPCToolkit for CPU, Nsight Systems for CUDA) identified the dominant computational kernel, which was annotated as the optimization region and instrumented for kernel timing. Five input scales per benchmark were produced with upstream data generators or per-benchmark generators; the reference output of each scale is the output of the baseline program on that input.",
"rai:dataLimitations": "The benchmarks are kernels embedded in complete but moderately sized programs from academic suites, not full production HPC applications; libraries such as BLAS or MPI are not exercised. Reference outputs and tolerances were established with GCC 14.2 / CUDA 12.4 at -O3 on an AMD EPYC 7543 / NVIDIA A100 system; other compilers or architectures can change floating-point results within the stated tolerances. Sparse linear algebra is covered by three SpMV kernels only. The CUDA set is small (22 benchmarks in the paper; 24 directories released). A few (setting, benchmark, size) combinations have no reference output and are flagged in index/inputs.parquet. Structural complexity is a static code-complexity proxy, not a measure of optimization headroom.",
"rai:dataBiases": "Coverage is skewed toward dense linear algebra and stencil kernels (22 and 13 of 92 CPU workloads) and toward short-running kernels (61 of 92 at the lowest complexity level); irregular, sparse, and dependency-heavy workloads are under-represented relative to production HPC codes. All programs are C/C++/CUDA; Fortran and other HPC languages are absent.",
"rai:personalSensitiveInformation": "None. The data consist of program source code and synthetic or standard scientific inputs; no personal or sensitive information is included.",
"rai:dataUseCases": "Evaluating code-generating models on correctness-preserving performance optimization of HPC kernels under serial, OpenMP, and CUDA settings; studying failure modes of LLM optimizations; developing and testing compilers, autotuners, or program-optimization tools against fixed inputs and references. Not intended as a measure of end-to-end application performance or of model capability outside kernel-level optimization.",
"rai:dataSocialImpact": "Supports reproducible measurement of automated performance engineering, which can reduce the energy and cost of scientific computing. Misuse risk is low; results could be over-interpreted as evidence of general software-engineering ability, which the benchmark does not measure. Models tuned to this benchmark may overfit its specific programs.",
"rai:hasSyntheticData": true,
"prov:wasDerivedFrom": [
"https://www.cs.colostate.edu/~pouchet/software/polybench/",
"https://www.cs.virginia.edu/rodinia/",
"https://github.com/breagen/MachSuite",
"http://impact.crhc.illinois.edu/parboil/parboil.aspx",
"https://vhosts.eecs.umich.edu/mibench/",
"https://parsec.cs.princeton.edu/",
"https://www.codee.com/"
],
"prov:wasGeneratedBy": "Programs were taken from the upstream suites, normalized to a common Makefile infrastructure (gcc -O3; -fopenmp for EX2; nvcc for EX3), profiled to locate the dominant kernel, and instrumented for kernel timing. Inputs at five scales were generated with upstream or per-benchmark generators (seeded, deterministic). Reference outputs were produced by running the instrumented baseline program on each input; MachSuite reference files come from the upstream suite. The index tables were derived from the benchmark catalog, output specifications, and difficulty scores of the GitHub repository at commit 7dc9824ba10249babe9c898b7591415b79445e22 and from a scan of the archives.",
"distribution": [
{
"@type": "cr:FileObject",
"@id": "hf-repository",
"name": "hf-repository",
"description": "The Hugging Face dataset repository bcui2/HPC-Bench.",
"contentUrl": "https://huggingface.co/datasets/bcui2/HPC-Bench/tree/main",
"encodingFormat": "git+https",
"sha256": "https://github.com/mlcommons/croissant/issues/80"
},
{
"@type": "cr:FileObject",
"@id": "github-repository",
"name": "github-repository",
"description": "Evaluation pipeline, LLM candidates, and per-trial results.",
"contentUrl": "https://github.com/Deep-Learning-Profiling-Tools/HPC-Bench",
"encodingFormat": "git+https",
"sha256": "https://github.com/mlcommons/croissant/issues/80"
},
{
"@type": "cr:FileObject",
"@id": "input-archive-manifest",
"name": "manifest.csv",
"description": "Path, size, and SHA-256 of every input archive.",
"contentUrl": "https://huggingface.co/datasets/bcui2/HPC-Bench/resolve/main/manifest.csv",
"encodingFormat": "text/csv",
"sha256": "3eb741a7783c13648a0b1cb9825fb5109c9c6bee294f0b5e7a4d90c170cb0396"
},
{
"@type": "cr:FileObject",
"@id": "baseline-code-manifest",
"name": "baseline_code/manifest.csv",
"description": "Path, size, and SHA-256 of every baseline source archive.",
"contentUrl": "https://huggingface.co/datasets/bcui2/HPC-Bench/resolve/main/baseline_code/manifest.csv",
"encodingFormat": "text/csv",
"sha256": "2bd068b959b3bdb545408005a870b1e538e21db485926fe5f7cf5afe63ae5618"
},
{
"@type": "cr:FileSet",
"@id": "input-archives",
"name": "input-archives",
"containedIn": {
"@id": "hf-repository"
},
"description": "One zip archive per (setting, benchmark) holding input_data/<size>/{input,output}/ for the five input scales; output/ holds the reference outputs.",
"encodingFormat": "application/zip",
"includes": "input_data/*/*.zip"
},
{
"@type": "cr:FileSet",
"@id": "baseline-code-archives",
"name": "baseline-code-archives",
"containedIn": {
"@id": "hf-repository"
},
"description": "One zip archive per (setting, benchmark) with the instrumented baseline sources, plus build_infrastructure.zip.",
"encodingFormat": "application/zip",
"includes": "baseline_code/**/*.zip"
},
{
"@type": "cr:FileObject",
"@id": "benchmarks-parquet",
"name": "index/benchmarks.parquet",
"description": "Structured index: benchmarks.",
"contentUrl": "https://huggingface.co/datasets/bcui2/HPC-Bench/resolve/main/index/benchmarks.parquet",
"encodingFormat": "application/x-parquet",
"sha256": "a7373d4e1c11f661b8f0f74b655d294af49609c9e28c6a94e6c2ea5308f5e464"
},
{
"@type": "cr:FileObject",
"@id": "inputs-parquet",
"name": "index/inputs.parquet",
"description": "Structured index: inputs.",
"contentUrl": "https://huggingface.co/datasets/bcui2/HPC-Bench/resolve/main/index/inputs.parquet",
"encodingFormat": "application/x-parquet",
"sha256": "9a4abea627479b014a18c80f4f4bbcc045263c98af05a601ed93ccd4beaf7e44"
},
{
"@type": "cr:FileObject",
"@id": "golden_outputs-parquet",
"name": "index/golden_outputs.parquet",
"description": "Structured index: golden_outputs.",
"contentUrl": "https://huggingface.co/datasets/bcui2/HPC-Bench/resolve/main/index/golden_outputs.parquet",
"encodingFormat": "application/x-parquet",
"sha256": "17aefc7bc8a3026cda55d817204f5ffdfce6f48f68e25f67bea6f14e1a8c19d3"
}
],
"recordSet": [
{
"@type": "cr:RecordSet",
"@id": "benchmarks",
"name": "benchmarks",
"description": "One record per (setting, benchmark): classification, annotated region, baseline-code and input-archive pointers, correctness-check parameters, provenance.",
"key": [
{
"@id": "benchmarks/setting"
},
{
"@id": "benchmarks/benchmark"
}
],
"field": [
{
"@type": "cr:Field",
"@id": "benchmarks/benchmark",
"name": "benchmark",
"description": "Benchmark name",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "benchmark"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/setting",
"name": "setting",
"description": "Optimization setting: EX1 serial CPU, EX2 OpenMP CPU, EX3 CUDA GPU",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "setting"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/setting_name",
"name": "setting_name",
"description": "setting_name",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "setting_name"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/language",
"name": "language",
"description": "language",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "language"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/suite",
"name": "suite",
"description": "Source benchmark suite",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "suite"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/suite_name",
"name": "suite_name",
"description": "suite_name",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "suite_name"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/motif",
"name": "motif",
"description": "Computational motif",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "motif"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/motif_abbrev",
"name": "motif_abbrev",
"description": "Motif abbreviation",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "motif_abbrev"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/structural_complexity",
"name": "structural_complexity",
"description": "Structural complexity level d1-d4 of the annotated region",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "structural_complexity"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/complexity_loc",
"name": "complexity_loc",
"description": "complexity_loc",
"dataType": "sc:Integer",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "complexity_loc"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/complexity_functions",
"name": "complexity_functions",
"description": "complexity_functions",
"dataType": "sc:Integer",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "complexity_functions"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/complexity_branches",
"name": "complexity_branches",
"description": "complexity_branches",
"dataType": "sc:Integer",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "complexity_branches"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/complexity_loop_depth",
"name": "complexity_loop_depth",
"description": "complexity_loop_depth",
"dataType": "sc:Integer",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "complexity_loop_depth"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/complexity_score",
"name": "complexity_score",
"description": "complexity_score",
"dataType": "sc:Float",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "complexity_score"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/in_paper_benchmark_set",
"name": "in_paper_benchmark_set",
"description": "Whether the (setting, benchmark) pair is part of the paper's reported benchmark set",
"dataType": "sc:Boolean",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "in_paper_benchmark_set"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/annotated_target_file",
"name": "annotated_target_file",
"description": "Source file containing the annotated kernel region",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "annotated_target_file"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/annotated_start_line",
"name": "annotated_start_line",
"description": "First line of the annotated region",
"dataType": "sc:Integer",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "annotated_start_line"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/annotated_end_line",
"name": "annotated_end_line",
"description": "Last line of the annotated region",
"dataType": "sc:Integer",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "annotated_end_line"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/annotated_regions_json",
"name": "annotated_regions_json",
"description": "annotated_regions_json",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "annotated_regions_json"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/baseline_code_path",
"name": "baseline_code_path",
"description": "Path of the baseline source archive in this dataset",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "baseline_code_path"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/baseline_code_bytes",
"name": "baseline_code_bytes",
"description": "baseline_code_bytes",
"dataType": "sc:Integer",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "baseline_code_bytes"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/baseline_code_sha256",
"name": "baseline_code_sha256",
"description": "SHA-256 of the baseline source archive",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "baseline_code_sha256"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/baseline_code_files",
"name": "baseline_code_files",
"description": "baseline_code_files",
"dataType": "sc:Integer",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "baseline_code_files"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/baseline_kernel_files",
"name": "baseline_kernel_files",
"description": "Baseline kernel file(s) inside <EX>_optimized_codes/",
"dataType": "sc:Text",
"repeated": true,
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "baseline_kernel_files"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/source_repo",
"name": "source_repo",
"description": "source_repo",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "source_repo"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/source_commit",
"name": "source_commit",
"description": "GitHub commit the baseline sources were taken from",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "source_commit"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/source_path",
"name": "source_path",
"description": "source_path",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "source_path"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/input_archive_path",
"name": "input_archive_path",
"description": "Path of the input archive in this dataset",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "input_archive_path"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/input_archive_bytes",
"name": "input_archive_bytes",
"description": "input_archive_bytes",
"dataType": "sc:Integer",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "input_archive_bytes"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/input_archive_sha256",
"name": "input_archive_sha256",
"description": "SHA-256 of the input archive",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "input_archive_sha256"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/input_sizes",
"name": "input_sizes",
"description": "Input scales present in the archive",
"dataType": "sc:Text",
"repeated": true,
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "input_sizes"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/n_input_sizes",
"name": "n_input_sizes",
"description": "n_input_sizes",
"dataType": "sc:Integer",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "n_input_sizes"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/golden_output_sizes",
"name": "golden_output_sizes",
"description": "Input scales that carry a reference output",
"dataType": "sc:Text",
"repeated": true,
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "golden_output_sizes"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/n_golden_output_sizes",
"name": "n_golden_output_sizes",
"description": "n_golden_output_sizes",
"dataType": "sc:Integer",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "n_golden_output_sizes"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/golden_output_bytes",
"name": "golden_output_bytes",
"description": "golden_output_bytes",
"dataType": "sc:Integer",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "golden_output_bytes"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/checker",
"name": "checker",
"description": "Correctness checker used for the benchmark (unregistered: no output specification in the released checker framework)",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "checker"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/output_type",
"name": "output_type",
"description": "output_type",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "output_type"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/output_dtype",
"name": "output_dtype",
"description": "output_dtype",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "output_dtype"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/output_suffix",
"name": "output_suffix",
"description": "output_suffix",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "output_suffix"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/atol",
"name": "atol",
"description": "Absolute tolerance applied by the correctness checker",
"dataType": "sc:Float",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "atol"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/rtol",
"name": "rtol",
"description": "Relative tolerance applied by the correctness checker",
"dataType": "sc:Float",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "rtol"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/tolerance_source",
"name": "tolerance_source",
"description": "Where the applied tolerance is defined (benchmark catalog, output specification, checker constants, or program self-check)",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "tolerance_source"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/comparison_rule",
"name": "comparison_rule",
"description": "Correctness comparison rule in words",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "comparison_rule"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/runtime_args_json",
"name": "runtime_args_json",
"description": "Command-line arguments per input size (JSON)",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "runtime_args_json"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/upstream_suite",
"name": "upstream_suite",
"description": "Originating benchmark suite",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "upstream_suite"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/upstream_url",
"name": "upstream_url",
"description": "Upstream suite URL",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "upstream_url"
}
}
},
{
"@type": "cr:Field",
"@id": "benchmarks/upstream_reference",
"name": "upstream_reference",
"description": "Upstream citation",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "benchmarks-parquet"
},
"extract": {
"column": "upstream_reference"
}
}
}
]
},
{
"@type": "cr:RecordSet",
"@id": "inputs",
"name": "inputs",
"description": "One record per (setting, benchmark, input size): input files and reference-output availability inside the archive.",
"key": [
{
"@id": "inputs/setting"
},
{
"@id": "inputs/benchmark"
},
{
"@id": "inputs/input_size"
}
],
"field": [
{
"@type": "cr:Field",
"@id": "inputs/setting",
"name": "setting",
"description": "Optimization setting: EX1 serial CPU, EX2 OpenMP CPU, EX3 CUDA GPU",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "inputs-parquet"
},
"extract": {
"column": "setting"
}
}
},
{
"@type": "cr:Field",
"@id": "inputs/benchmark",
"name": "benchmark",
"description": "Benchmark name",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "inputs-parquet"
},
"extract": {
"column": "benchmark"
}
}
},
{
"@type": "cr:Field",
"@id": "inputs/input_size",
"name": "input_size",
"description": "Input scale",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "inputs-parquet"
},
"extract": {
"column": "input_size"
}
}
},
{
"@type": "cr:Field",
"@id": "inputs/size_rank",
"name": "size_rank",
"description": "size_rank",
"dataType": "sc:Integer",
"source": {
"fileObject": {
"@id": "inputs-parquet"
},
"extract": {
"column": "size_rank"
}
}
},
{
"@type": "cr:Field",
"@id": "inputs/input_archive_path",
"name": "input_archive_path",
"description": "Path of the input archive in this dataset",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "inputs-parquet"
},
"extract": {
"column": "input_archive_path"
}
}
},
{
"@type": "cr:Field",
"@id": "inputs/size_dir_in_archive",
"name": "size_dir_in_archive",
"description": "size_dir_in_archive",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "inputs-parquet"
},
"extract": {
"column": "size_dir_in_archive"
}
}
},
{
"@type": "cr:Field",
"@id": "inputs/size_present",
"name": "size_present",
"description": "size_present",
"dataType": "sc:Boolean",
"source": {
"fileObject": {
"@id": "inputs-parquet"
},
"extract": {
"column": "size_present"
}
}
},
{
"@type": "cr:Field",
"@id": "inputs/input_files",
"name": "input_files",
"description": "input_files",
"dataType": "sc:Integer",
"source": {
"fileObject": {
"@id": "inputs-parquet"
},
"extract": {
"column": "input_files"
}
}
},
{
"@type": "cr:Field",
"@id": "inputs/input_bytes",
"name": "input_bytes",
"description": "input_bytes",
"dataType": "sc:Integer",
"source": {
"fileObject": {
"@id": "inputs-parquet"
},
"extract": {
"column": "input_bytes"
}
}
},
{
"@type": "cr:Field",
"@id": "inputs/golden_output_available",
"name": "golden_output_available",
"description": "Whether a reference output exists for this input size",
"dataType": "sc:Boolean",
"source": {
"fileObject": {
"@id": "inputs-parquet"
},
"extract": {
"column": "golden_output_available"
}
}
},
{
"@type": "cr:Field",
"@id": "inputs/golden_output_files",
"name": "golden_output_files",
"description": "golden_output_files",
"dataType": "sc:Text",
"repeated": true,
"source": {
"fileObject": {
"@id": "inputs-parquet"
},
"extract": {
"column": "golden_output_files"
}
}
},
{
"@type": "cr:Field",
"@id": "inputs/golden_output_bytes",
"name": "golden_output_bytes",
"description": "golden_output_bytes",
"dataType": "sc:Integer",
"source": {
"fileObject": {
"@id": "inputs-parquet"
},
"extract": {
"column": "golden_output_bytes"
}
}
},
{
"@type": "cr:Field",
"@id": "inputs/runtime_args",
"name": "runtime_args",
"description": "runtime_args",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "inputs-parquet"
},
"extract": {
"column": "runtime_args"
}
}
}
]
},
{
"@type": "cr:RecordSet",
"@id": "golden_outputs",
"name": "golden_outputs",
"description": "One record per reference-output file inside the input archives, with the comparison parameters of its benchmark.",
"key": [
{
"@id": "golden_outputs/setting"
},
{
"@id": "golden_outputs/benchmark"
},
{
"@id": "golden_outputs/input_size"
},
{
"@id": "golden_outputs/path_in_archive"
}
],
"field": [
{
"@type": "cr:Field",
"@id": "golden_outputs/setting",
"name": "setting",
"description": "Optimization setting: EX1 serial CPU, EX2 OpenMP CPU, EX3 CUDA GPU",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "golden_outputs-parquet"
},
"extract": {
"column": "setting"
}
}
},
{
"@type": "cr:Field",
"@id": "golden_outputs/benchmark",
"name": "benchmark",
"description": "Benchmark name",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "golden_outputs-parquet"
},
"extract": {
"column": "benchmark"
}
}
},
{
"@type": "cr:Field",
"@id": "golden_outputs/input_size",
"name": "input_size",
"description": "Input scale",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "golden_outputs-parquet"
},
"extract": {
"column": "input_size"
}
}
},
{
"@type": "cr:Field",
"@id": "golden_outputs/input_archive_path",
"name": "input_archive_path",
"description": "Path of the input archive in this dataset",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "golden_outputs-parquet"
},
"extract": {
"column": "input_archive_path"
}
}
},
{
"@type": "cr:Field",
"@id": "golden_outputs/path_in_archive",
"name": "path_in_archive",
"description": "Path of the reference-output file inside the input archive",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "golden_outputs-parquet"
},
"extract": {
"column": "path_in_archive"
}
}
},
{
"@type": "cr:Field",
"@id": "golden_outputs/bytes",
"name": "bytes",
"description": "bytes",
"dataType": "sc:Integer",
"source": {
"fileObject": {
"@id": "golden_outputs-parquet"
},
"extract": {
"column": "bytes"
}
}
},
{
"@type": "cr:Field",
"@id": "golden_outputs/file_suffix",
"name": "file_suffix",
"description": "file_suffix",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "golden_outputs-parquet"
},
"extract": {
"column": "file_suffix"
}
}
},
{
"@type": "cr:Field",
"@id": "golden_outputs/checker",
"name": "checker",
"description": "Correctness checker used for the benchmark (unregistered: no output specification in the released checker framework)",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "golden_outputs-parquet"
},
"extract": {
"column": "checker"
}
}
},
{
"@type": "cr:Field",
"@id": "golden_outputs/output_type",
"name": "output_type",
"description": "output_type",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "golden_outputs-parquet"
},
"extract": {
"column": "output_type"
}
}
},
{
"@type": "cr:Field",
"@id": "golden_outputs/output_dtype",
"name": "output_dtype",
"description": "output_dtype",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "golden_outputs-parquet"
},
"extract": {
"column": "output_dtype"
}
}
},
{
"@type": "cr:Field",
"@id": "golden_outputs/atol",
"name": "atol",
"description": "Absolute tolerance applied by the correctness checker",
"dataType": "sc:Float",
"source": {
"fileObject": {
"@id": "golden_outputs-parquet"
},
"extract": {
"column": "atol"
}
}
},
{
"@type": "cr:Field",
"@id": "golden_outputs/rtol",
"name": "rtol",
"description": "Relative tolerance applied by the correctness checker",
"dataType": "sc:Float",
"source": {
"fileObject": {
"@id": "golden_outputs-parquet"
},
"extract": {
"column": "rtol"
}
}
},
{
"@type": "cr:Field",
"@id": "golden_outputs/tolerance_source",
"name": "tolerance_source",
"description": "Where the applied tolerance is defined (benchmark catalog, output specification, checker constants, or program self-check)",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "golden_outputs-parquet"
},
"extract": {
"column": "tolerance_source"
}
}
},
{
"@type": "cr:Field",
"@id": "golden_outputs/comparison_rule",
"name": "comparison_rule",
"description": "Correctness comparison rule in words",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "golden_outputs-parquet"
},
"extract": {
"column": "comparison_rule"
}
}
}
]
},
{
"@type": "cr:RecordSet",
"@id": "input_archives",
"name": "input_archives",
"description": "Input archive manifest.",
"key": [
{
"@id": "input_archives/path"
}
],
"field": [
{
"@type": "cr:Field",
"@id": "input_archives/ex",
"name": "ex",
"description": "Setting",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "input-archive-manifest"
},
"extract": {
"column": "ex"
}
}
},
{
"@type": "cr:Field",
"@id": "input_archives/benchmark",
"name": "benchmark",
"description": "Benchmark name",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "input-archive-manifest"
},
"extract": {
"column": "benchmark"
}
}
},
{
"@type": "cr:Field",
"@id": "input_archives/path",
"name": "path",
"description": "Archive path in this dataset",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "input-archive-manifest"
},
"extract": {
"column": "path"
}
}
},
{
"@type": "cr:Field",
"@id": "input_archives/bytes",
"name": "bytes",
"description": "Archive size in bytes",
"dataType": "sc:Integer",
"source": {
"fileObject": {
"@id": "input-archive-manifest"
},
"extract": {
"column": "bytes"
}
}
},
{
"@type": "cr:Field",
"@id": "input_archives/sha256",
"name": "sha256",
"description": "SHA-256 of the archive",
"dataType": "sc:Text",
"source": {
"fileObject": {
"@id": "input-archive-manifest"
},
"extract": {
"column": "sha256"
}
}
}
]
}
]
}
{
"@context": {
"@language": "en",
"@vocab": "https://schema.org/",
"citeAs": "cr:citeAs",
"column": "cr:column",
"conformsTo": "dct:conformsTo",
"cr": "http://mlcommons.org/croissant/",
"rai": "http://mlcommons.org/croissant/RAI/",
"data": {
"@id": "cr:data",
"@type": "@json"
},
"dataType": {
"@id": "cr:dataType",
"@type": "@vocab"
},
"dct": "http://purl.org/dc/terms/",
"examples": {
"@id": "cr:examples",
"@type": "@json"
},
"extract": "cr:extract",
"field": "cr:field",
"fileProperty": "cr:fileProperty",
"fileObject": "cr:fileObject",
"fileSet": "cr:fileSet",
"format": "cr:format",
"includes": "cr:includes",
"isLiveDataset": "cr:isLiveDataset",
"jsonPath": "cr:jsonPath",
"key": "cr:key",
"md5": "cr:md5",
"parentField": "cr:parentField",
"path": "cr:path",
"recordSet": "cr:recordSet",
"references": "cr:references",
"regex": "cr:regex",
"repeated": "cr:repeated",
"replace": "cr:replace",
"sc": "https://schema.org/",
"separator": "cr:separator",
"source": "cr:source",
"subField": "cr:subField",
"transform": "cr:transform",
"wd": "https://www.wikidata.org/wiki/",
"prov": "http://www.w3.org/ns/prov#"
},
"@type": "sc:Dataset",
"name": "HPC-Bench",
"description": "HPC-Bench is a benchmark for evaluating large language models as optimizers of computational kernels embedded in complete HPC programs. This dataset provides the standardized inputs at five scales, the reference outputs used for correctness checking, the exact instrumented baseline programs, and a structured index for 92 CPU workloads (serial and OpenMP settings) and a Rodinia-derived CUDA set, drawn from PolyBench/C, Rodinia, Parboil, MachSuite, MiBench, PARSEC, and Codee and spanning 16 computational motifs. The evaluation pipeline, LLM-generated candidates, and per-trial results are in the GitHub repository.",
"conformsTo": "http://mlcommons.org/croissant/1.1",
"url": "https://huggingface.co/datasets/bcui2/HPC-Bench",
"sameAs": "https://github.com/Deep-Learning-Profiling-Tools/HPC-Bench",
"creator": [
{
"@type": "sc:Person",
"name": "Bowen Cui"
},
{
"@type": "sc:Person",
"name": "Junyu Yin"
},
{
"@type": "sc:Person",
"name": "Tejas Ramesh"
},
{
"@type": "sc:Person",
"name": "Leo Lim"
},
{
"@type": "sc:Person",
"name": "Oscar Hernandez"
},
{
"@type": "sc:Person",
"name": "Keren Zhou"
}
],
"license": [
"https://opensource.org/licenses/MIT",
"https://github.com/Deep-Learning-Profiling-Tools/HPC-Bench/blob/master/docs/BENCHMARKS.md"
],
"citeAs": "@article{cui2026hpc,\n title={HPC-Bench: A Comprehensive Benchmark for High Performance Computing Codes},\n author={Cui, Bowen and Yin, Junyu and Ramesh, Tejas and Lim, Leo and Hernandez, Oscar and Zhou, Keren},\n journal={arXiv preprint arXiv:2611.02752},\n year={2026}\n}",
"version": "1.0.1",
"datePublished": "2026-10-31",
"dateModified": "2026-11-06",
"keywords": [
"HPC",
"benchmark",
"code optimization",
"large language models",
"OpenMP",
"CUDA",
"performance"
],
"rai:dataCollection": "Benchmark programs were selected from seven established HPC benchmark suites. For each program, a profiler (HPCToolkit for CPU, Nsight Systems for CUDA) identified the dominant computational kernel, which was annotated as the optimization region and instrumented for kernel timing. Five input scales per benchmark were produced with upstream data generators or per-benchmark generators; the reference output of each scale is the output of the baseline program on that input.",
"rai:dataLimitations": "The benchmarks are kernels embedded in complete but moderately sized programs from academic suites, not full production HPC applications; libraries such as BLAS or MPI are not exercised. Reference outputs and tolerances were established with GCC 14.2 / CUDA 12.4 at -O3 on an AMD EPYC 7543 / NVIDIA A100 system; other compilers or architectures can change floating-point results within the stated tolerances. Sparse linear algebra is covered by three SpMV kernels only. The CUDA set is small (22 benchmarks in the paper; 24 directories released). A few (setting, benchmark, size) combinations have no reference output and are flagged in index/inputs.parquet. Structural complexity is a static code-complexity proxy, not a measure of optimization headroom.",
"rai:dataBiases": "Coverage is skewed toward dense linear algebra and stencil kernels (22 and 13 of 92 CPU workloads) and toward short-running kernels (61 of 92 at the lowest complexity level); irregular, sparse, and dependency-heavy workloads are under-represented relative to production HPC codes. All programs are C/C++/CUDA; Fortran and other HPC languages are absent.",