File "/usr/local/python/lib/python3.8/site-packages/diffusers/models/attention_processor.py", line 3074, in __call__ [rank0]: batch_size, key_tokens, _ = ( [rank0]: ValueError: too many values to unpack (expected 3)

### Describe the bug

添加完xformer加速之后出现了这个问题

### Reproduction

def __init__():

self.unet = pipeline.unet
self.set_diffusers_xformers_flag(self.unet,True)

def set_diffusers_xformers_flag( model, valid):

def fn_recursive_set_mem_eff(module: torch.nn.Module):
if hasattr(module, "set_use_memory_efficient_attention_xformers"):
module.set_use_memory_efficient_attention_xformers(valid)
print("="*100)
print(hasattr(module, "set_use_memory_efficient_attention_xformers"))

for child in module.children():
fn_recursive_set_mem_eff(child)
fn_recursive_set_mem_eff(model)

def forward():
self.attn1(
mda_norm_hidden_states,
encoder_hidden_states=encoder_hidden_states if self.only_cross_attention else None,
attention_mask=attention_mask,
**cross_attention_kwargs,
)

### Logs
# ps：The rest of the code is not convenient to provide
```shell

```

### System Info

- 🤗 Diffusers version: 0.33.1
- Platform: Linux-5.4.241-1-tlinux4-0017.7-x86_64-with-glibc2.2.5
- Running on Google Colab?: No
- Python version: 3.8.12
- PyTorch version (GPU?): 2.4.0+cu118 (True)
- Flax version (CPU?/GPU?/TPU?): not installed (NA)
- Jax version: not installed
- JaxLib version: not installed
- Huggingface_hub version: 0.30.2
- Transformers version: 4.46.3
- Accelerate version: 1.0.1
- PEFT version: 0.13.2
- Bitsandbytes version: 0.42.0
- Safetensors version: 0.4.3
- xFormers version: 0.0.27.post2+cu118
- Accelerator: NVIDIA H20, 97871 MiB

### Who can help?

_No response_

Inizia in diffusers/models/attention_processor.py intorno alla riga 3074, dove si verifica il ValueError segnalato durante la chiamata di attention con xFormers abilitato. Riproduci il problema usando la configurazione set_use_memory_efficient_attention_xformers fornita e ispeziona i valori che vengono spacchettati; il lavoro è completato quando la configurazione segnalata non genera più questo errore relativo alla forma.

Scritto dal modello di indicizzazione a partire dal testo della issue.