Skip to main content

Mojo function

moe_create_indices_kernel

moe_create_indices_kernel[input_type: DType, num_threads: Int, TokenExpertOrderLayoutType: TensorLayout, ExpertStartIndicesLayoutType: TensorLayout, RestoreTokenOrderLayoutType: TensorLayout, ExpertIdsLayoutType: TensorLayout, ExpertUsageStatsLayoutType: TensorLayout, IndicesPaddedLayoutType: TensorLayout, PaddedInputLayoutType: TensorLayout, TopkIdsLayoutType: TensorLayout](token_expert_order: TileTensor[DType.uint32, TokenExpertOrderLayoutType, MutAnyOrigin], expert_start_indices: TileTensor[DType.uint32, ExpertStartIndicesLayoutType, MutAnyOrigin], restore_token_order: TileTensor[DType.uint32, RestoreTokenOrderLayoutType, MutAnyOrigin], expert_ids: TileTensor[DType.int32, ExpertIdsLayoutType, MutAnyOrigin], expert_usage_stats: TileTensor[DType.uint32, ExpertUsageStatsLayoutType, MutAnyOrigin], indices_padded: TileTensor[DType.uint32, IndicesPaddedLayoutType, MutAnyOrigin], topk_ids_padded: TileTensor[input_type, PaddedInputLayoutType, MutAnyOrigin], topk_ids: TileTensor[input_type, TopkIdsLayoutType, MutAnyOrigin])

Was this page helpful?