flashinfer.experimental.cake_stepfun_moe.requant

flashinfer.experimental.cake_stepfun_moe.requant(**kwargs: Any) → None

Warning

cake_stepfun_moe.requant is experimental: it provides no compatibility guarantees and may change or be removed without deprecation.

Run the exported NVFP4 per-token requantization stage once on

caller-owned buffers (cake_backend.prepare_requant(**kwargs).launch()).