flashinfer.norm.supports_rmsnorm_sigmoid_gate

flashinfer.norm.supports_rmsnorm_sigmoid_gate(dtype: dtype, head_dim: int, device: device | str | None = None, *, backend: Literal['auto', 'cake', 'cute'] = 'auto', enable_pdl: bool = True) → bool

Whether a BF16, head-dimension-64 call can run on device (the current one when None) through backend ("auto": the Cake package or the CuTe DSL kernel) in the launch form enable_pdl selects. The package routes on the head count and the token count of a call as well and serves 32-byte aligned storage alone; rmsnorm_sigmoid_gate falls to the CuTe DSL kernel where they lie outside its windows or its alignment.