mindspore.ops.moe_token_permute

View Source On Gitee
mindspore.ops.moe_token_permute(tokens, indices, num_out_tokens=None, padded_mode=False)[source]

Permute the tokens based on the indices. Token with the same index will be grouped together.

Warning

  • It is only supported on Atlas A2 Training Series Products.

  • The input tokens only supports the bfloat16 data type in the current version.

  • This is an experimental API that is subject to change or deletion.

  • When indices is 2-D, the size of the second dim must be less than or equal to 512.

Parameters

  • tokens (Tensor) – The input token tensor to be permuted. The dtype is bfloat16. The shape is \((num\_tokens, hidden\_size)\) , where num_tokens and hidden_size are positive integers.

  • indices (Tensor) – The tensor specifies indices used to permute the tokens. The dtype is int32 or int64. The shape is \((num\_tokens, topk)\) or \((num\_tokens,)\), where num_tokens and topk are positive integers. If the shape is the latter case, topk is implied to be 1.

  • num_out_tokens (int, optional) – The effective output token count, when enabling the capacity factor, should equal the number of tokens not dropped. By default, set to None, meaning no tokens are dropped.

  • padded_mode (bool, optional) – If True, indicating the indices are padded to denote selected tokens per expert. It can only be False currently. Default: False .

Returns

tuple (Tensor), tuple of 2 tensors, containing the permuted tokens and sorted indices.

  • permuted_tokens (Tensor) - The permuted tensor of the same dtype as tokens.

  • sorted_indices (Tensor) - The indices Tensor of dtype int32, corresponds to permuted tensor.

Raises

  • TypeError – If tokens or indices is not a Tensor.

  • TypeError – If dtype of tokens is not bfloat16.

  • TypeError – If dtype of indices is not int32 or int64.

  • TypeError – If specified num_out_tokens is not an integer.

  • TypeError – If specified padded_mode is not a bool.

  • ValueError – If second dim of indices is greater than 512 when exists.

  • ValueError – If padded_node is set to True.

  • ValueError – If tokens is not 2-D or indices is not 1-D or 2-D Tensor.

  • RuntimeError – If first dimensions of tokens and indices are not consistent.

Supported Platforms:

Ascend

Examples

>>> import mindspore
>>> from mindspore import Tensor, ops
>>> tokens = Tensor([[1, 1, 1],
...                  [7, 7, 7],
...                  [2, 2, 2],
...                  [1, 1, 1],
...                  [2, 2, 2],
...                  [3, 3, 3]], dtype=mindspore.bfloat16)
>>> indices = Tensor([5, 0, 3, 1, 2, 4], dtype=mindspore.int32)
>>> out = ops.moe_untoken_permute(tokens, indices)
>>> print(out)
(Tensor(shape=[6, 3], dtype=BFloat16, value=
[[7, 7, 7],
 [1, 1, 1],
 [2, 2, 2],
 [2, 2, 2],
 [3, 3, 3],
 [1, 1, 1]]), Tensor(shape=[6], dtype=Int32, value= [5, 0, 3, 1, 2, 4]))