デバッグ用の出力を追加して、プロジェクション出力のデバイスを確認

Author: hndrr

diffusers_helper/models/hunyuan_video_packed.py

@@ -591,7 +591,14 @@ def forward(
 
         # 3. Modulation and residual connection
         hidden_states = torch.cat([attn_output, mlp_hidden_states], dim=2)
-        hidden_states = gate * self.proj_out(hidden_states)
+        print(f"DEBUG: Before proj_out - hidden_states device: {hidden_states.device}")
+        print(f"DEBUG: Before proj_out - gate device: {gate.device}")
+        # Check proj_out layer's device (assuming it's a nn.Module with parameters)
+        proj_out_device = next(self.proj_out.parameters()).device if list(self.proj_out.parameters()) else "No parameters"
+        print(f"DEBUG: Before proj_out - self.proj_out device: {proj_out_device}")
+        proj_out_result = self.proj_out(hidden_states)
+        print(f"DEBUG: After proj_out - proj_out_result device: {proj_out_result.device}")
+        hidden_states = gate * proj_out_result # Error likely occurs here
         hidden_states = hidden_states + residual
 
         hidden_states, encoder_hidden_states = (