HINT: We don't support broadcasting, please use `expand` yourself before calling `memory_efficient_attention`

[nianchu1]

ValueError: Incompatible shapes for attention inputs: query.shape: torch.Size([32, 14400, 8, 40]) key.shape : torch.Size([2, 77, 8, 40]) value.shape: torch.Size([2, 77, 8, 40]) HINT: We don't support broadcasting, please use expand yourself before calling memory_efficient_attention if you need to

When i use animatediff ,it occur above quesiton.what does it mens?

[qooyvonne]

I also have the same problem when i use "Reference" in ControlNet v1.1.443.

ValueError: Incompatible shapes for attention inputs: query.shape: torch.Size([2, 7225, 8, 40]) key.shape : torch.Size([1, 77, 8, 40]) value.shape: torch.Size([1, 77, 8, 40]) HINT: We don't support broadcasting, please use expand yourself before calling memory_efficient_attention if you need to

[DenSckriva]

Same problem for me with Tiled Diffusion & VAE extension. :(

[zhichaoLii]

Same problem for me with Tiled Diffusion & VAE extension. :(

Have you solve the problem in the multidiffusion webui extension?

[Wangbenzhi]

I solve this issue by setting drop_last=False in the dataloder.

[liyuantsao]

Have the same question.

[jxrnxx]

有同样的疑问。

[zz13526585541]

当我在 ControlNet v1.1.443 中使用“参考”时，我也遇到了同样的问题。

ValueError：注意力输入的形状不兼容：query.shape：torch。size（[2， 7225， 8， 40]） key.shape ： 火把。size（[1， 77， 8， 40]） value.shape： 火炬。Size（[1， 77， 8， 40]）提示：我们不支持广播，如果需要，请在致电前自行使用expand``memory_efficient_attention

In optimization Settings :[PR] reverse cue word leads to minimum sigma (when the image is close to completion...) Just change the value to 0