track-models

You can track new open-source LLMs and update kernel coverage docs by running the discovery command to identify untracked models, then refreshing the status to update FlashInfer-Bench documentation with per-kernel details.

To validate kernel coverage, the skill scans model definitions and uses frontmatter metadata to check whether a model's kernels are already covered, then updates the model coverage documentation accordingly.

Yes, you can discover newly released open-source LLMs automatically by running the discovery command, which identifies new models and determines their current kernel coverage status for tracking.

You refresh coverage status for already-tracked models by running the status update command, which scans existing definitions and updates the docs with a concise summary and detailed per-kernel status.

No specific dependencies are required to update model coverage, as the skill operates standalone using frontmatter metadata and definition scanning to validate and refresh the documentation.

The updated model coverage documentation uses the mdx format, specifically updating the docs/model_coverage.mdx file with a concise summary and detailed per-kernel status for tracked LLMs.