Ouroboros Kernel Corpus

OUROBOROS Verified Kernel Corpus is a set of Triton GPU kernels written almost entirely by open-weight models within an AI loop and then checked by a verifier the models cannot fool. Every kernel compiled, matched PyTorch on an adversarial correctness sweep, and outperformed torch.compile max-autotune before inclusion. The dataset serves as the training and evidence data for the Kernel Mint Space, with its last update recorded on 2026-06-15.

Use Cases

• Training AI models for code generation based on verified, high-performance kernel examples.
• Benchmarking compiler optimization techniques against the adversarial correctness sweep results mentioned in the description.
• Studying the capabilities and limitations of AI-generated code within a closed-loop verification system.
• Developing new verification methods for machine-generated code artifacts.

Strengths

• Kernels are verified by an adversarial checker the generating models cannot fool.
• Every included kernel compiled and matched PyTorch on an adversarial correctness sweep.
• Kernels reportedly beat torch.compile max-autotune before inclusion.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last updated 2026-06-15 21:02:37; freshness should be verified.

Provenance

Source

huggingface, author YMRohit

Collection Method

Generated by open-weight models inside the OUROBOROS loop and verified by an adversarial checker.

Freshness

Last updated 2026-06-15 21:02:37.