megatron.core.tensor_parallel.utils#
Description
Classes
Split the vocabulary into world_size chunks and return the first and last index of the vocabulary belonging to the rank partition: Note that indices in [fist, last) |
Functions
|
Opposite of split_tensor_into_1d_equal_chunks. |
|
Split a tensor along its last dimension. |
|
Break a tensor into equal 1D chunks across tensor parallel ranks. |