Use regular numbers for rope in lumina model.

comfy/ldm/lumina/model.py

@@ -9,6 +9,7 @@
 
 from comfy.ldm.modules.diffusionmodules.mmdit import TimestepEmbedder, RMSNorm
 from comfy.ldm.modules.attention import optimized_attention_masked
+from comfy.ldm.flux.layers import EmbedND
 
 
 def modulate(x, scale):
@@ -92,10 +93,9 @@ def apply_rotary_emb(
                 and key tensor with rotary embeddings.
         """
 
-        x = torch.view_as_complex(x_in.float().reshape(*x_in.shape[:-1], -1, 2))
-        freqs_cis = freqs_cis.unsqueeze(2)
-        x_out = torch.view_as_real(x * freqs_cis).flatten(3)
-        return x_out.type_as(x_in)
+        t_ = x_in.reshape(*x_in.shape[:-1], -1, 1, 2).float()
+        t_out = freqs_cis[..., 0] * t_[..., 0] + freqs_cis[..., 1] * t_[..., 1]
+        return t_out.reshape(*x_in.shape).type_as(x_in)
 
     def forward(
         self,
@@ -130,6 +130,7 @@ def forward(
 
         xq = self.q_norm(xq)
         xk = self.k_norm(xk)
+
         xq = JointAttention.apply_rotary_emb(xq, freqs_cis=freqs_cis)
         xk = JointAttention.apply_rotary_emb(xk, freqs_cis=freqs_cis)
 
@@ -480,7 +481,8 @@ def __init__(
         assert (dim // n_heads) == sum(axes_dims)
         self.axes_dims = axes_dims
         self.axes_lens = axes_lens
-        self.rope_embedder = RopeEmbedder(axes_dims=axes_dims, axes_lens=axes_lens)
+        # self.rope_embedder = RopeEmbedder(axes_dims=axes_dims, axes_lens=axes_lens)
+        self.rope_embedder = EmbedND(dim=dim // n_heads, theta=10000.0, axes_dim=axes_dims)
         self.dim = dim
         self.n_heads = n_heads
 
@@ -550,7 +552,7 @@ def patchify_and_embed(
             position_ids[i, cap_len:cap_len+img_len, 1] = row_ids
             position_ids[i, cap_len:cap_len+img_len, 2] = col_ids
 
-        freqs_cis = self.rope_embedder(position_ids)
+        freqs_cis = self.rope_embedder(position_ids).movedim(1, 2)
 
         # build freqs_cis for cap and image individually
         cap_freqs_cis_shape = list(freqs_cis.shape)