Structs

  • | Inverse operation of Partition. | | Takes the original, full ‘keys’ tensor | followed by sharded value tensors, | and returns the full value tensor, combined | using the same hash used in | Partition. |
  • | LengthsPartition splits the input | int tensor into multiple ones according | to the second tensor. The first dimension | is expected to be the tensor that describes | lengths of the elements. | | Takes the second input and partitions | it to shards according to the remainder | of values modulo the number of partitions. | It requires the second tensor to be a | 1D-tensor of the integral type. The | first tensor should be 1D-tensor of | int32 that would represent the lengths | of the elements in the input. The number | of partitions is derived as (num_output | / num_input). | | If additional inputs are present they | must have the same shape as the first | input, optionally with extra trailing | dimensions. They will be partitioned | accordingly to the first input. | | Optional arg ‘pack_first_input’ transforms | the first tensor values as | | X_ij / num_partitions. | | Outputs are ordered as | | X_0_part_0, X_1_part_0, …, X_N-1_part_0, | X_0_part_1, …, X_N-1_part_K-1 |
  • | Splits the input int tensor into multiple | ones according to the first tensor. | | Takes the first input and partitions | it to shards according to the remainder | of values modulo the number of partitions. | It requires that the first tensor is | of integral type. The number of partitions | is derived as (num_output / num_input). | | If additional inputs are present they | must have the same shape as the first | input, optionally with extra trailing | dimensions. They will be partitioned | accordingly to the first input. | | Optional arg ‘pack_first_input’ transforms | the first tensor values as | | X_ij / num_partitions. | | Outputs are ordered as | | X_0_part_0, X_1_part_0, …, X_N-1_part_0, | X_0_part_1, …, X_N-1_part_K-1 |

Functions