Refactor: Rename Magi1AttnProcessor2_0 and Magi1TransformerBlock classes for clarity

Refactor: Rename VAE components for clarity

Renames the attention processor and transformer block classes to be more specific to the VAE architecture.

This improves code readability by making the purpose of these components more explicit.

Author: tolgacangoz

src/diffusers/models/autoencoders/autoencoder_kl_magi1.py

@@ -34,7 +34,7 @@
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
 
 
-class Magi1AttnProcessor2_0:
+class Magi1VAEAttnProcessor2_0:
     def __init__(self, dim, num_heads=8):
         if not hasattr(F, "scaled_dot_product_attention"):
             raise ImportError("WanAttnProcessor2_0 requires PyTorch 2.0. To use it, please upgrade PyTorch to 2.0.")
@@ -80,7 +80,7 @@ def __call__(
         return hidden_states + identity
 
 
-class Magi1TransformerBlock(nn.Module):
+class Magi1VAETransformerBlock(nn.Module):
     def __init__(
         self,
         dim: int,
@@ -99,7 +99,7 @@ def __init__(
             bias=True,
             cross_attention_dim=None,
             out_bias=True,
-            processor=Magi1AttnProcessor2_0(dim, num_heads),
+            processor=Magi1VAEAttnProcessor2_0(dim, num_heads),
         )
 
         self.drop_path = nn.Identity()
@@ -154,7 +154,7 @@ def __init__(
         # 3. Transformer blocks
         self.blocks = nn.ModuleList(
             [
-                Magi1TransformerBlock(
+                Magi1VAETransformerBlock(
                     inner_dim,
                     num_attention_heads,
                     ffn_dim,
@@ -241,7 +241,7 @@ def __init__(
         # 3. Transformer blocks
         self.blocks = nn.ModuleList(
             [
-                Magi1TransformerBlock(
+                Magi1VAETransformerBlock(
                     inner_dim,
                     num_attention_heads,
                     ffn_dim,
@@ -317,7 +317,7 @@ class AutoencoderKLMagi1(ModelMixin, ConfigMixin, FromOriginalModelMixin):
 
     _supports_gradient_checkpointing = False
     _skip_layerwise_casting_patterns = ["patch_embedding", "norm"]
-    _no_split_modules = ["Magi1TransformerBlock"]
+    _no_split_modules = ["Magi1VAETransformerBlock"]
     _keep_in_fp32_modules = ["qkv_norm", "norm1", "norm2"]
     _keys_to_ignore_on_load_unexpected = ["norm_added_q"]