Added custom multihead attention code.

docs/reference/node_reference.md

@@ -176,6 +176,7 @@
 | [Ptn Masked Mean Pooling](ptn_masked_mean_pooling.md) | Constructs a masked mean pooling layer. |
 | [Ptn Max Pool 2d](ptn_max_pool2d.md) | A maxpool layer.   |
 | [Ptn Multihead Attention](ptn_multihead_attention.md) | A Multihead attention model. |
+| [Ptn Multihead Attention Custom](ptn_multihead_attention_custom.md) | A Multihead attention model. |
 | [Ptn Pre Add Channel Axis](ptn_pre_add_channel_axis.md) | Adds a channel axis after the batch axis if the input is rank 3 (bs, h, w) |
 | [Ptn Pre Flatten](ptn_pre_flatten.md) | Flattens the input tensor before processing the tensor in the specified model. |
 | [Ptn RNN](ptn_rnn.md) | A recurrent neural network (RNN) model consisting of one or more of a recurrent layer.   |

docs/reference/ptn_multihead_attention_custom.md

@@ -0,0 +1,32 @@
+# Ptn Multihead Attention Custom
+A Multihead attention model.
+
+    Args:
+        embed_dim (int): Hidden dimension (d_model) for each token of the Transformer.
+        num_heads  (int): Number of attention heads.
+        dropout (float): Dropout rate for output.
+        bias (bool): If True, bias will be added to the linear projections of query, key, value, and the final output.
+        kdim (int): Dimension of k.
+        vdim (int): Dimension of v.
+        batch_first (bool): Input shape format: True for (B, Seq, Feature), False for (Seq, B, Feature).
+
+## Input
+| Name | Data type |
+|---|---|
+| embed_dim | Int |
+| num_heads | Int |
+| dropout | Float |
+| bias | Boolean |
+| kdim | Int |
+| vdim | Int |
+| batch_first | Boolean |
+
+## Output
+| Data type |
+|---|
+| Ptmodel |
+
+<HR>
+Category: PyTorch wrapper - Model
+
+ComfyUI Pt Wrapper Node Reference. © 2025 Hide Inada (HowToSD.com). All rights reserved.