Participant GuideAdvanced
Advanced — Deeper tooling
Sourced from whest-starterkit @
aaa3882.
Advanced — Deeper tooling
You don't need either of these to ship a submission. They become useful once you're tuning aggressively and want better visibility into where FLOPs and time go.
| Doc | When to read |
|---|---|
| profile-simulation.md | Profile the FLOP and time breakdown of your predict() call — identify the dominant op before you optimize. |
| use-whestbench-explorer.md | Open the hosted WhestBench Explorer (now an external app at aicrowd.github.io/whestbench-explorer) to inspect MLPs and ground-truth activations layer-by-layer. |
➡️ Where to look next
- Need a faster local iteration loop instead? → How-to: use evaluation datasets.
- Looking for the FLOP cost of specific ops? → Reference: flopscope primer.