GPUMODE/kernelbot-data

1---2configs:3- config_name: amd_submissions4  data_files: "submissions.parquet"5- config_name: amd_successful_submissions6  data_files: "successful_submissions.parquet"7- config_name: nvidia_nvfp4_submissions8  data_files: "nvidia_nvfp4_submissions.parquet"9- config_name: leaderboards10  data_files: "leaderboards.parquet"11tags:12- code13license: mit14---15 16# KernelBot Competition Data17 18This dataset contains GPU kernel submissions from the KernelBot competition platform. Submissions are optimized GPU kernels written for specific hardware targets.19 20## Data Files21 22### AMD MI300 Submissions23| File | Description |24|------|-------------|25| `submissions.parquet` | All AMD competition submissions |26| `successful_submissions.parquet` | AMD submissions that passed correctness tests |27| `deduplicated_submissions.parquet` | AMD submissions deduplicated by (user, code) |28| `deduplicated_successful_submissions.parquet` | Deduplicated passing AMD submissions |29 30**AMD Problems:** fp8-gemm, moe (mixture of experts), mla-decode, all2all, gemm+reducescatter, allgather+gemm31 32### NVIDIA Blackwell NVFP4 Submissions33| File | Size | Description |34|------|------|-------------|35| `nvidia_nvfp4_submissions.parquet` | ~1.4 GB | NVFP4 submissions deduplicated by (user, code), with full code content |36 37**NVIDIA NVFP4 Problems:**38| Problem | Submissions | Unique Users | Passing |39|---------|-------------|--------------|---------|40| nvfp4_gemv | 35,559 | 281 | 4,860 |41| nvfp4_gemm | 48,682 | 160 | 26,123 |42| nvfp4_dual_gemm | 113,353 | 160 | 73,802 |43 44**Note:** Scores are execution time in seconds. **Lower is better.**45 46## Helper Scripts47 48- `analyze_submissions.py` - Python functions for analyzing submissions49- `skills.md` - Documentation for data processing workflows50 51### Quick Start52 53```python54from analyze_submissions import load_submissions, top_contestants, author_progression55 56# Load NVIDIA NVFP4 data57df = load_submissions()58 59# Get top 20 for a problem60leaders = top_contestants(df, problem_name='nvfp4_gemm', n=20)61 62# See a user's progression over time63progression = author_progression(df, user_name='username', problem_name='nvfp4_gemm')64```65 66## Learn More67 68- Competition platform: [gpumode.com](https://gpumode.com)69- Reference kernels and problem specs: [github.com/gpu-mode/reference-kernels](https://github.com/gpu-mode/reference-kernels)70 71## Citation72 73If you use this dataset in your work, please cite:74 75```bibtex76@inproceedings{77  zhang2025kernelbot,78  title={KernelBot: A Competition Platform for Writing Heterogeneous {GPU} Code},79  author={Alex L Zhang and Matej Sirovatka and Erik Schultheis and Benjamin Horowitz and Mark Saroufim},80  booktitle={Championing Open-source DEvelopment in ML Workshop @ ICML25},81  year={2025},82  url={https://openreview.net/forum?id=bq9U4dmuyJ}83}84```85