jtatman/kev-code-verify-data

Kev code-verify: execution-labelled coder attempts 4920 attempts by nine coder models, plus the reference solutions, at the 164 HumanEvalPack Python tasks, each labelled by running the task's hidden test suite. Built to train and evaluate decision models that tell an agent router whether a cheap model's code can be trusted or should be escalated. Every attempt also carries evidence a router can compute without the hidden tests: does the code compile, define the requested… See the full description on the dataset page: https://huggingface.co/datasets/jtatman/kev-code-verify-data.

<iframe src="https://silytics.si/embed/item/2c02053a-eaea-4fb7-9a85-c2b3396b757d" width="420" height="190" style="border:0" loading="lazy" title="Silytics entry"></iframe>

Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 17 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.