fix fastercache implementation

Author: a-r-r-o-w

src/diffusers/hooks/faster_cache.py

@@ -18,6 +18,7 @@
 
 import torch
 
+from ..models.attention import AttentionModuleMixin
 from ..models.attention_processor import Attention, MochiAttention
 from ..models.modeling_outputs import Transformer2DModelOutput
 from ..utils import logging
@@ -567,7 +568,7 @@ def high_frequency_weight_callback(module: torch.nn.Module) -> float:
     _apply_faster_cache_on_denoiser(module, config)
 
     for name, submodule in module.named_modules():
-        if not isinstance(submodule, _ATTENTION_CLASSES):
+        if not isinstance(submodule, (*_ATTENTION_CLASSES, AttentionModuleMixin)):
             continue
         if any(re.search(identifier, name) is not None for identifier in _TRANSFORMER_BLOCK_IDENTIFIERS):
             _apply_faster_cache_on_attention_class(name, submodule, config)