Dataset Viewer
Auto-converted to Parquet Duplicate
problem_id
int64
101
108
name
stringclasses
8 values
level
int64
1
1
domain
stringclasses
6 values
subdomain
stringclasses
8 values
task_type
stringclasses
1 value
backend
stringclasses
1 value
precision
stringclasses
1 value
filter
bool
1 class
tags
listlengths
3
3
source
dict
task_key
stringclasses
8 values
task_version
int64
1
1
status
stringclasses
1 value
reference_hash
stringclasses
8 values
code
stringclasses
8 values
101
pairwise_rbf
1
statistics
kernel methods
pytorch_reference
cuda
fp32
true
[ "rbf", "gaussian-process", "distance" ]
{ "origin": "original reference implementation", "license": "AGPL-3.0", "references": [ "SchΓΆlkopf & Smola, Learning with Kernels (2002)", "Rasmussen & Williams, Gaussian Processes for Machine Learning (2006)" ] }
statistics.pairwise_rbf
1
active
sha256:75d713ec5c91c18718397262fc396ebb1b134c88ac1c36cf119b46800dbc3784
"""Pairwise Gaussian (RBF) kernel matrix. Domain: statistics / machine learning (kernel methods). Solves: evaluates ``exp(-gamma * ||x_i - y_j||^2)`` for every pair. Source: SchΓΆlkopf & Smola, *Learning with Kernels* (2002); Rasmussen & Williams, *Gaussian Processes for Machine Learning* (2006). Level: 1 (pairwise dis...
102
laplacian_2d
1
physics
numerical PDEs
pytorch_reference
cuda
fp32
true
[ "stencil", "laplacian", "pde" ]
{ "origin": "original reference implementation", "license": "AGPL-3.0", "references": [ "LeVeque, Finite Difference Methods for ODEs and PDEs (2007)", "Press et al., Numerical Recipes (2007)" ] }
physics.laplacian_2d
1
active
sha256:211fefee4e71a24062d139b47aa124beed98f2b6560bf5059330d79b6630ff2c
"""Five-point Laplacian stencil on a 2D field. Domain: computational physics / numerical PDEs. Solves: applies the discrete Laplacian used by diffusion, Poisson and wave-equation solvers to a 2D field. Source: LeVeque, *Finite Difference Methods for Ordinary and Partial Differential Equations* (2007); Press et al., *N...
103
heat_step_2d
1
physics
computational physics
pytorch_reference
cuda
fp32
true
[ "heat-equation", "diffusion", "pde" ]
{ "origin": "original reference implementation", "license": "AGPL-3.0", "references": [ "Fourier, ThΓ©orie analytique de la chaleur (1822)", "LeVeque (2007)" ] }
physics.heat_step_2d
1
active
sha256:c79695876d1b442c1b9c97547fbada1ef9f8b351befb8f068262bf9b9b4dec64
"""Explicit heat-equation step on a 2D field. Domain: computational physics / partial differential equations. Solves: advances the heat equation ``u_t = alpha * laplacian(u)`` by one explicit finite-difference timestep with zero-flux (replicate) boundaries. Source: Fourier's heat equation (1822); the explicit scheme i...
104
lennard_jones_energy
1
chemistry
molecular dynamics
pytorch_reference
cuda
fp32
true
[ "lennard-jones", "potential-energy", "md" ]
{ "origin": "original reference implementation", "license": "AGPL-3.0", "references": [ "Lennard-Jones (1924)", "Allen & Tildesley, Computer Simulation of Liquids (1987)" ] }
chemistry.lennard_jones_energy
1
active
sha256:fd51a04549d41e017487d4d33a0280993692d8475b3080e18ead80fd4c64261b
"""Total Lennard-Jones potential energy of a small particle cluster. Domain: computational chemistry / molecular dynamics. Solves: evaluates ``U = 4 * epsilon * sum_{i<j} [(sigma/r_ij)^12 - (sigma/r_ij)^6]`` for all distinct particle pairs. Source: Lennard-Jones, "On the determination of molecular fields" (1924); the ...
105
lotka_volterra_step
1
biology
population ecology
pytorch_reference
cuda
fp32
true
[ "ode", "predator-prey", "rk4" ]
{ "origin": "original reference implementation", "license": "AGPL-3.0", "references": [ "Lotka, Elements of Physical Biology (1925)", "Volterra (1926)" ] }
biology.lotka_volterra_step
1
active
sha256:f377d3997ab136dbd90579d410f3d6899ac62f5a80940dab3112a7b198389816
"""One classical Runge-Kutta step of the Lotka-Volterra predator-prey system. Domain: mathematical biology / population ecology. Solves: advances the ODE ``dx/dt = alpha*x - beta*x*y`` and ``dy/dt = delta*x*y - gamma*y`` by one RK4 step for a batch of population states. Source: Lotka, *Elements of Physical Biology* (1...
106
gaussian_kde
1
statistics
density estimation
pytorch_reference
cuda
fp32
true
[ "kde", "gaussian", "kernel" ]
{ "origin": "original reference implementation", "license": "AGPL-3.0", "references": [ "Rosenblatt (1956)", "Parzen (1962)" ] }
statistics.gaussian_kde
1
active
sha256:0603e3b94276410103aadab03db3ec89ce1b34672820871852ca0e7e2081abe2
"""Gaussian kernel density estimate at a set of query points. Domain: statistics / non-parametric density estimation. Solves: evaluates ``p(x) = 1/(N h^d (2*pi)^(d/2)) * sum_i exp(-||x - x_i||^2 / (2 h^2))``. Source: Rosenblatt, "Remarks on some nonparametric estimates of a density function" (1956); Parzen, "On estima...
107
horner_eval
1
mathematics
numerical methods
pytorch_reference
cuda
fp32
true
[ "polynomial", "horner", "evaluation" ]
{ "origin": "original reference implementation", "license": "AGPL-3.0", "references": [ "Horner (1819)", "Higham, Accuracy and Stability of Numerical Algorithms (2002)" ] }
mathematics.horner_eval
1
active
sha256:04bb76c93f87a94f2772f6542c4ac3aac842f251ca5679598a568af21711cd64
"""Horner evaluation of a polynomial at many points. Domain: numerical mathematics / scientific computing. Solves: evaluates ``p(x) = c_0 + c_1 x + ... + c_{K-1} x^{K-1}`` using Horner's recurrence, which requires K multiplications and additions instead of computing every power separately. Source: Horner, "A new metho...
108
black_scholes_call
1
economics
quantitative finance
pytorch_reference
cuda
fp32
true
[ "option-pricing", "black-scholes", "finance" ]
{ "origin": "original reference implementation", "license": "AGPL-3.0", "references": [ "Black & Scholes (1973)" ] }
economics.black_scholes_call
1
active
sha256:e77ae049b753a404fb6d29f922c037812c80f7a3e0365379039240474c45ceec
"""Vectorised Black-Scholes price of a European call option. Domain: economics / quantitative finance. Solves: evaluates ``C = S * Phi(d1) - K * exp(-rT) * Phi(d2)``, ``d1 = (log(S/K) + (r + sigma^2/2)T) / (sigma sqrt(T))``, ``d2 = d1 - sigma sqrt(T)``, where ``Phi`` is the standard normal CDF. Source: Black & Scholes...

Science KernelBench

A science-focused task set for robust-kernelbench. Each task provides a PyTorch reference implementation of a numerical primitive; the candidate must produce a faster, numerically equivalent GPU kernel.

This directory is the source of truth. to_hf.py expands the manifest into a Hugging Face dataset; the generated export is not committed.

Current starter set: 22 tasks across physics, chemistry, biology, statistics, mathematics and economics.

level focus tasks
level_1 single stencil, reduction or short elementwise expression 8
level_2 all-pairs interactions, batched linear algebra, multi-state updates 7
level_3 sequential dynamic programs, FFTs, tree backward induction 7

Each task carries domain, subdomain, level, tags and a source block. The reference files contain a docstring with the domain, the problem solved and the canonical paper/source.

Task list

id task domain level
101 pairwise_rbf statistics 1
102 laplacian_2d physics 1
103 heat_step_2d physics 1
104 lennard_jones_energy chemistry 1
105 lotka_volterra_step biology 1
106 gaussian_kde statistics 1
107 horner_eval mathematics 1
108 black_scholes_call economics 1
201 nbody_acceleration physics 2
202 coulomb_energy chemistry 2
203 hodgkin_huxley_step biology 2
204 kalman_update statistics 2
205 graph_laplacian mathematics 2
206 portfolio_variance economics 2
207 wave_step_2d physics 2
301 needleman_wunsch_score biology 3
302 fft_convolution mathematics 3
303 binomial_option_crr economics 3
304 split_step_schrodinger physics 3
305 ewald_real_space chemistry 3
306 gaussian_process_posterior statistics 3
307 floyd_warshall mathematics 3

Versioning

Current release: 0.1.0 (schema_version: kernelbench-v1), compatible with robust-kernelbench >= 0.2.0.

The dataset version is independent of the package version:

  • PATCH β€” documentation, metadata or provenance fixes with no scoring change.
  • MINOR β€” new tasks, domains or levels; existing tasks unchanged.
  • MAJOR β€” removal/renumbering, reference-output changes, or benchmark-split changes.

Every task has a permanent problem_id, a stable task_key, a task_version and a reference_hash. Removed tasks are marked retired; IDs are never reused. Pin production runs to an HF tag or commit:

robust-kernelbench \
  --dataset ai-nikolai/ScienceKernelBench \
  --dataset_revision v0.1.0 \
  --dataset_split level_1 \
  ...

Schema

Required columns:

column type meaning
problem_id int unique task id
name string filesystem-safe task name
code string self-contained reference source

The code must define:

class Model(nn.Module): ...
def get_inputs(): ...
def get_init_inputs(): ...

Optional columns understood by the harness:

column meaning
level dataset split / difficulty grouping (1-3 here)
domain physics, chemistry, biology, statistics, mathematics, economics
subdomain finer label, e.g. molecular dynamics, PDE, bioinformatics
filter include in the representative subset
task_type pytorch_reference or kernel_reference
backend cuda, triton, ...
precision fp32, fp16, bf16
tags searchable labels
source provenance / license metadata

Local layout

data/science_kernels/
β”œβ”€β”€ README.md                  # this dataset card
β”œβ”€β”€ manifest.json              # versioned task list and provenance
β”œβ”€β”€ CHANGELOG.md               # dataset release history
β”œβ”€β”€ requirements.txt           # minimal HF/export dependencies
β”œβ”€β”€ kernels/
β”‚   β”œβ”€β”€ level_1/               # 8 tasks
β”‚   β”œβ”€β”€ level_2/               # 7 tasks
β”‚   └── level_3/               # 7 tasks
└── to_hf.py                   # per-level export / optional push_to_hub

Minimal dependencies

pip install -r requirements.txt

torch is optional and only needed for --execute smoke tests; it is listed as a comment in the requirements file so the dataset can be loaded and exported on a lightweight machine.

Validate locally

# Schema only
robust-kernelbench-dataset validate \
  --dataset_path data/science_kernels \
  --dataset_format directory

# Counts by level and domain, and a browsable list
robust-kernelbench-dataset stats \
  --dataset_path data/science_kernels \
  --dataset_format directory
robust-kernelbench-dataset list \
  --dataset_path data/science_kernels \
  --dataset_format directory --domain physics

# Also run one forward pass per task
robust-kernelbench-dataset validate \
  --dataset_path data/science_kernels \
  --dataset_format directory \
  --execute --device cpu

Use with robust-kernelbench

Evaluation only:

robust-kernelbench-eval \
  --dataset_path data/science_kernels \
  --dataset_format directory \
  --dataset_split level_1 \
  --experiment science_exp --trial 1

Agentic generate -> compile -> evaluate:

robust-kernelbench \
  --dataset_path data/science_kernels \
  --dataset_format directory \
  --dataset_split level_1 \
  --dataset_name science-kernels \
  --version science_v0 \
  --model_name Qwen/Qwen3-Coder-30B-A3B-Instruct \
  "1 1 kernelbench"

Export for Hugging Face

python data/science_kernels/to_hf.py --output /tmp/science_kernels_hf

# huggingface_hub >= 1 uses the `hf` command:
hf upload ai-nikolai/ScienceKernelBench /tmp/science_kernels_hf --repo-type dataset

# Works with any huggingface_hub version:
python data/science_kernels/upload_to_hf.py \
  --path /tmp/science_kernels_hf

Or export and push in one step:

python data/science_kernels/to_hf.py \
  --output /tmp/science_kernels_hf \
  --push

Provenance and licensing

Only add kernels you have the right to redistribute. Record the source repository, commit, and license in each task's source block. Original tasks written for this dataset are released under AGPL-3.0, matching the repository. If you combine sources with different licenses, use the strictest compatible license for the combined dataset and document it in this card.

Citation

If you use this dataset, please cite robust-kernelbench:

@misc{rozanov2026robustkernelbench,
  title        = {Robust KernelBench: A Robust Harness for GPU Kernel Generation and Evaluation},
  author       = {Rozanov, Nikolai},
  year         = {2026},
  howpublished = {\url{https://github.com/ai-nikolai/robust_kernelbench}},
  note         = {ScienceKernelBench dataset, version 0.1.0}
}

The dataset card and version should also be cited when reporting benchmark results, together with the exact dataset_revision used.

Downloads last month
69