datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
lin-alg-kernels-coretitans-memory-kernels
Titans Memory Kernel - Nova Lineage
Evolved memory kernels for the Titans Memory system.
Structure
nova_prime.safetensors - The Prime kernel (16M params, 4096 dim)
nova_v1.safetensors - Production-hardened V1 (identity preserved)
REGISTRY.jsonl - Lineage registry with metadata
variants/ - 137 generation snapshots from evolution
Parameters
Dimensions: 4096
Memory Size: ~65MB per kernel
Evolution: Hardened with decay=0.99999, lr=0.0001
Usage
from… See the full description on the dataset page: https://huggingface.co/datasets/ADAPT-Chase/titans-memory-kernels.kern-kernels
kern-kernels
Reproducible attention kernel recipes, ABI manifests, checksums and measured
results. First profile: GB300 / Qwen3.8-27B / BF16 / Q24-KV4-D256 / page 64.
Uses unmodified TRTLLM-GEN full attention, not MLA. The model's GDN layers are
unchanged. Model weights and the base export's other kernels are not included.
NVIDIA binaries are downloaded directly from pinned upstream URLs and verified
by SHA256; this repository does not mirror them. The small Apache-2.0 vLLM KV… See the full description on the dataset page: https://huggingface.co/datasets/susun-123/kern-kernels.dynamic_kernels_rirkernel_synth_annotated
KernelSynth (annotated)
One million synthetic univariate time series, each 1024 points long, drawn from a Gaussian
process prior whose kernel is a random composition of up to five base kernels. This is the
KernelSynth procedure from Chronos with one addition:
the generating kernel is kept alongside each series. The ground-truth structure behind
every series is therefore known, which makes the corpus usable for interpretability work
rather than only for pretraining.… See the full description on the dataset page: https://huggingface.co/datasets/felixdivo/kernel_synth_annotated.kernelstrain
KernelStrain
Long-horizon GPU kernel optimization trajectories for training small models to iterate on CUDA and Triton kernels.
KernelStrain is a large synthetic dataset of kernel-optimization episodes: given a kernel task (shapes, dtype, GPU target, baseline code + timing), a model proposes successive complete kernel candidates, observes simulated benchmark / correctness / compile feedback, and keeps improving over many steps — structural rewrites, parameter sweeps, joint… See the full description on the dataset page: https://huggingface.co/datasets/Akahsizrr/kernelstrain.Kernel-Smith-SFT-71KIf this work is useful to you, please cite:
@article{DBLP:journals/corr/abs-2603-28342,
author = {He Du and
Qiming Ge and
Jiakai Hu and
Aijun Yang and
Zheng Cai and
Zixian Huang and
Sheng Yuan and
Qinxiu Cheng and
Xinchen Xie and
Yicheng Chen and
Yining Li and
Jiaxing Xie and… See the full description on the dataset page: https://huggingface.co/datasets/CoopReason/Kernel-Smith-SFT-71K.cuda-triton-gpu-kernels-2026
⚡ Complete 2026 CUDA & OpenAI Triton High-Performance GPU Kernel Engineering SFT/DPO Suite
The definitive, production-grade synthetic alignment dataset engineered for training and fine-tuning open-weights Large Language Models (Qwen 2.5 Coder, DeepSeek-Coder, Llama 3.1) on ultra-high-throughput GPU kernel programming: NVIDIA Hopper H100 / Blackwell B200 TMA async transfers, OpenAI Triton 3.1+ FlashAttention-3, 32-bank conflict elimination, and low-bit FP8 / INT4 GEMM… See the full description on the dataset page: https://huggingface.co/datasets/beatsprom/cuda-triton-gpu-kernels-2026.ParallelKernelBench_Kernels
ParallelKernelBench Kernels
Net-new multi-GPU CUDA kernels generated by LLMs for ParallelKernelBench.
Each subdirectory under solutions/ is one model run. File names match the benchmark problem stems (e.g. 17_rope_allgather_cuda.py ↔ problem 17_rope_allgather in willychan21/ParallelKernelBench_Problems).
Layout
solutions/
<run_id>/
<stem>_cuda.py
...
Runs (1 run(s), 87 kernel files)
run_id
kernels
path… See the full description on the dataset page: https://huggingface.co/datasets/willychan21/ParallelKernelBench_Kernels.ga104-cuda-kernels
GA104 Hand-Optimized CUDA Kernel Corpus
A measurement corpus of hand-optimized CUDA / SASS kernels targeting the
RTX 3070 Ti (GA104, sm_86, Ampere). Every kernel is written without
cuBLAS, cuDNN, or PyTorch in the optimized path; vendor libraries are
linked only for measured comparison under kernels/reference/. This
dataset is for SASS and GPU-optimization researchers — it pairs each
.cu source with its compiled machine code and its disassembly, so the
exact instruction stream a… See the full description on the dataset page: https://huggingface.co/datasets/pjt222/ga104-cuda-kernels.kernelstrain2
kernelstrain2 — kernelinfer
Synthetic long-horizon GPU kernel-optimization trajectories, inference-only.
Each row is a complete optimization session (~30–200 steps, simulating
1–5 hours of work): the model receives an inference-kernel task, emits a
full CUDA or Triton candidate, and gets simulated compile / correctness /
benchmark feedback — iterating, self-correcting, and tuning launch
parameters until convergence. No reasoning traces; assistant turns are
kernel code only.… See the full description on the dataset page: https://huggingface.co/datasets/Akahsizrr/kernelstrain2.kernelswarm
kernelswarm — multi-agent kernel-optimization episodes (RL)
Synthetic dataset of multi-agent long-horizon optimization campaigns on
AI-inference kernels. Each row is one whole episode: a lead orchestrator plus
2–20 specialist agents split the work, exchange dispatches/statuses/handoffs,
submit full kernel candidates, and receive simulated compile/verify/bench
feedback — with a reward attached to every environment result.
Inference-focused only. All ops are inference primitives… See the full description on the dataset page: https://huggingface.co/datasets/Akahsizrr/kernelswarm.litmus-kernels
Litmus Kernel Verification Corpus
Correct and deliberately-broken Triton kernels, each broken one shipped with
the input that exposes it.
The corpus exists to measure one thing: how much of what a fixed-shape
torch.rand() allclose test calls "correct" actually is. On this corpus the
answer is that 88% of the planted bugs pass that test.
Columns
column
meaning
name
kernel identifier
family
elementwise / reduction / softmax / layernorm / matmul /… See the full description on the dataset page: https://huggingface.co/datasets/NagaYu/litmus-kernels.nanode-cuda-kernels
Nanode CUDA kernels — prebuilt wheels
Prebuilt CUDA extension wheels used by Nanode,
a node-based AI canvas. Hosted here because most of these builds exist nowhere
else — PyPI carries source-only distributions (or nothing at all) for these
packages, and the upstream binary channels don't cover these ABI combinations.
Windows — Microsoft TRELLIS.2 kernels (cp311)
The custom CUDA extensions behind TRELLIS.2
image-to-3D generation, compiled for Windows. To our… See the full description on the dataset page: https://huggingface.co/datasets/pandafood/nanode-cuda-kernels.KDA_kernelsb-sides-v2-kernels
B-Sides kernel search index
Published runtime artifacts for the B-Sides kernel corpus.
meta.sqlite: searchable kernel metadata
embeddings.f16.npy: row-aligned 256-dimensional query matrix
The kernels.row values align one-to-one with the embedding matrix rows.
SYNTHETIC-2-Base-Kernelskernelsight
KernelSight v4
Per-timestep workload labels for GPU execution traces.
KernelSight pairs every GPU workload trace with a dense, per-timestep workload
labeling. Each snapshot is a [24, 512] counter image — 24 hardware-counter
channels sampled across 512 equal-width time bins — paired with per-bin labels
drawn from a two-level hierarchy of 12 coarse (L1) and 73 fine (L2)
workload classes. The goal is to label what a kernel is doing at each instant
(matmul, attention, reduction… See the full description on the dataset page: https://huggingface.co/datasets/williamhtan/kernelsight.Kernel-Smith-RL-2KIf this work is useful to you, please cite:
@article{DBLP:journals/corr/abs-2603-28342,
author = {He Du and
Qiming Ge and
Jiakai Hu and
Aijun Yang and
Zheng Cai and
Zixian Huang and
Sheng Yuan and
Qinxiu Cheng and
Xinchen Xie and
Yicheng Chen and
Yining Li and
Jiaxing Xie and… See the full description on the dataset page: https://huggingface.co/datasets/CoopReason/Kernel-Smith-RL-2K.cuda_kernelsFECA_ECOLI_Tsuboyama_2023_2D1U_substitutions_singles_stability_PE_REGRKernel-Smith-Seed-59KIf this work is useful to you, please cite:
@article{DBLP:journals/corr/abs-2603-28342,
author = {He Du and
Qiming Ge and
Jiakai Hu and
Aijun Yang and
Zheng Cai and
Zixian Huang and
Sheng Yuan and
Qinxiu Cheng and
Xinchen Xie and
Yicheng Chen and
Yining Li and
Jiaxing Xie and… See the full description on the dataset page: https://huggingface.co/datasets/CoopReason/Kernel-Smith-Seed-59K.m3-mxfp8-kernelspyc-kernels
PyC CUDA kernel lab
This repository documents 19 CUDA kernel-lab entries from PyC. It is a source
and evidence release, not a compiled binary distribution and not a claim that
all entries are wired into PyC runtime dispatch.
Contents
kernels/prototypes/: standalone CUDA prototype sources.
manifests/lab_kernels.json: the 19-entry lab catalog, including build/run commands.
manifests/registry_kernels.json: the catalog mirrored into the registry release.… See the full description on the dataset page: https://huggingface.co/datasets/AethronPhantom/pyc-kernels.styleforge-kernels
StyleForge Pre-compiled CUDA Kernels
This repository contains pre-compiled CUDA kernels for the StyleForge neural style transfer project.
Files
fused_instance_norm.so
Usage
These kernels are automatically downloaded by StyleForge when running on Hugging Face Spaces.
Compilation
Kernels are compiled for multiple GPU architectures:
sm_70 (V100)
sm_75 (T4)
sm_80 (A100)
For local compilation, see compile_kernels.py in the main repository.
krea2-aoti-kernelskernel-syzfix-qwen35
syzfix Kernel Crash Fixes (sample)
Description
Teaches domain-specific instruction following and code generation for this expert.
Source
xiaoguangwang/syzfix-dataset - 500 MB streamed sample of 5.1 GB
Formatted for the MoE-orchestrator project
(https://github.com/michaelowusuntim6/MoE-orchestrator). Expert target:
linux_kernel.
Format
Each record is a JSON object with a messages field formatted for Qwen3.5's
native chat template:… See the full description on the dataset page: https://huggingface.co/datasets/michaelowusuntim6/kernel-syzfix-qwen35.SYNTHETIC-2-Base-v2-Kernelskernel-skill-sourceopencl-kernels