onboard-model

To onboard new LLM kernels into FlashInfer-Bench, initiate the full pipeline via /onboard-model --discover or specify a target with /onboard-model --model-name. The process automates kernel discovery, definition generation, workload collection, and PR submission.

End-to-end model onboarding orchestration coordinates model discovery, kernel definition generation, workload collection, and PR submission. It manages repository updates across tmp/ directories and tracks kernel availability and issues throughout the integration pipeline.

Yes, you can automate kernel definition generation and workload collection pipelines. The orchestration generates definitions and collects workloads automatically, optionally proceeding to PR submission workflows for seamless integration into the benchmark suite.

Yes, the onboarding pipeline explicitly performs kernel availability checks and handles issue tracking. It coordinates multiple sub-skills to ensure new models with unknown kernels are defined, tested, and integrated into the benchmark suite.

The best way to integrate a new model with unknown kernels is using an automated onboarding orchestration pipeline. It coordinates discovery, generates kernel definitions, collects workloads, and submits PRs to ensure complete integration into the benchmark suite.