flashinfer.fused_moe.CuTileBf16Config

class flashinfer.fused_moe.CuTileBf16Config

cuTile BF16 backend.

Expert parallelism and fused shared experts are not supported.

__init__() None

Methods

__init__()

prepare_weights(w1_bf16, w2_bf16, *, ...[, ...])

Build the cutile_bf16 weight view from canonical BF16 weights.

supported(arch)