mindflow.cell.ConditionDiffusionTransformer

查看源文件
class mindflow.cell.ConditionDiffusionTransformer(in_channels, out_channels, cond_channels, hidden_channels, layers, heads, time_token_cond=True, cond_as_token=True, compute_dtype=mstype.float32)[源代码]

以Transformer作为骨干网络的条件控制扩散模型。

参数:
  • in_channels (int) - 输入特征维度。

  • out_channels (int) - 输出特征维度。

  • hidden_channels (int) - 隐藏层特征维度。

  • cond_channels (int) - 条件特征维度。

  • layers (int) - Transformer 层数。

  • heads (int) - 注意力头数。

  • time_token_cond (bool) - 是否将时间作为条件token。Default: True

  • cond_as_token (bool) - 是否将条件作为token。Default: True

  • compute_dtype (mindspore.dtype) - 计算数据类型。支持 mstype.float32 or mstype.float16 。默认值: mstype.float32 ,表示 mindspore.float32

输入:
  • x (Tensor) - 网络输入。shape为 \((batch\_size, sequence\_len, in\_channels)\) 的Tensor。

  • timestep (Tensor) - 时间步。shape为 \((batch\_size,)\) 的Tensor。

  • condition (Tensor) - 控制条件。shape为 \((batch\_size, cond\_channels)\) 的Tensor。

输出:
  • output (Tensor) - shape为 \((batch\_size, sequence\_len, out\_channels)\) 的Tensor。

支持平台:

Ascend

样例:

>>> from mindspore import ops
>>> from mindflow.cell import ConditionDiffusionTransformer
>>> in_channels, out_channels, cond_channels, hidden_channels = 16, 16, 10, 256
>>> layers, heads, batch_size, seq_len = 3, 4, 8, 256
>>> model = ConditionDiffusionTransformer(in_channels=in_channels,
...                                       out_channels=out_channels,
...                                       cond_channels=cond_channels,
...                                       hidden_channels=hidden_channels,
...                                       layers=layers,
...                                       heads=heads)
>>> x = ops.rand((batch_size, seq_len, in_channels))
>>> cond = ops.rand((batch_size, cond_channels))
>>> timestep = ops.randint(0, 1000, (batch_size,))
>>> output = model(x, timestep, cond)
>>> print(output.shape)
(8, 256, 16)