[Bugfix] Fix shape mismatch in LongCatAudioDiTTransformer conversion (#13494)

Signed-off-by: Lancer <maruixiang6688@gmail.com>

Author: RuixiangMa

scripts/convert_longcat_audio_dit_to_diffusers.py

@@ -131,6 +131,7 @@ def convert_longcat_audio_dit(
         cross_attn_norm=config.get("dit_cross_attn_norm", False),
         eps=config.get("dit_eps", 1e-6),
         use_latent_condition=config.get("dit_use_latent_condition", True),
+        ff_mult=config.get("dit_ff_mult", 4),
     )
     transformer.load_state_dict(transformer_state_dict, strict=True)
     transformer = transformer.to(dtype=torch_dtype)

src/diffusers/models/transformers/transformer_longcat_audio_dit.py

@@ -475,6 +475,7 @@ def __init__(
         cross_attn_norm: bool = False,
         eps: float = 1e-6,
         use_latent_condition: bool = True,
+        ff_mult: float = 4.0,
     ):
         super().__init__()
         dim = dit_dim
@@ -498,7 +499,7 @@ def __init__(
                     cross_attn_norm=cross_attn_norm,
                     adaln_type=adaln_type,
                     adaln_use_text_cond=adaln_use_text_cond,
-                    ff_mult=4.0,
+                    ff_mult=ff_mult,
                 )
                 for _ in range(dit_depth)
             ]