fix scale_shift_factor being on cpu for wan and ltx

src/diffusers/models/transformers/transformer_ltx.py

@@ -353,7 +353,9 @@ def forward(
         norm_hidden_states = self.norm1(hidden_states)
 
         num_ada_params = self.scale_shift_table.shape[0]
-        ada_values = self.scale_shift_table[None, None] + temb.reshape(batch_size, temb.size(1), num_ada_params, -1)
+        ada_values = self.scale_shift_table[None, None].to(temb.device) + temb.reshape(
+            batch_size, temb.size(1), num_ada_params, -1
+        )
         shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = ada_values.unbind(dim=2)
         norm_hidden_states = norm_hidden_states * (1 + scale_msa) + shift_msa
 

src/diffusers/models/transformers/transformer_wan.py

@@ -682,12 +682,12 @@ def forward(
         # 5. Output norm, projection & unpatchify
         if temb.ndim == 3:
             # batch_size, seq_len, inner_dim (wan 2.2 ti2v)
-            shift, scale = (self.scale_shift_table.unsqueeze(0) + temb.unsqueeze(2)).chunk(2, dim=2)
+            shift, scale = (self.scale_shift_table.unsqueeze(0).to(temb.device) + temb.unsqueeze(2)).chunk(2, dim=2)
             shift = shift.squeeze(2)
             scale = scale.squeeze(2)
         else:
             # batch_size, inner_dim
-            shift, scale = (self.scale_shift_table + temb.unsqueeze(1)).chunk(2, dim=1)
+            shift, scale = (self.scale_shift_table.to(temb.device) + temb.unsqueeze(1)).chunk(2, dim=1)
 
         # Move the shift and scale tensors to the same device as hidden_states.
         # When using multi-GPU inference via accelerate these will be on the

src/diffusers/models/transformers/transformer_wan_vace.py

@@ -103,7 +103,7 @@ def forward(
             control_hidden_states = control_hidden_states + hidden_states
 
         shift_msa, scale_msa, gate_msa, c_shift_msa, c_scale_msa, c_gate_msa = (
-            self.scale_shift_table + temb.float()
+            self.scale_shift_table.to(temb.device) + temb.float()
         ).chunk(6, dim=1)
 
         # 1. Self-attention
@@ -361,7 +361,7 @@ def forward(
                     hidden_states = hidden_states + control_hint * scale
 
         # 6. Output norm, projection & unpatchify
-        shift, scale = (self.scale_shift_table + temb.unsqueeze(1)).chunk(2, dim=1)
+        shift, scale = (self.scale_shift_table.to(temb.device) + temb.unsqueeze(1)).chunk(2, dim=1)
 
         # Move the shift and scale tensors to the same device as hidden_states.
         # When using multi-GPU inference via accelerate these will be on the