Update modeling_qwen2_nomic_vision.py

modeling_qwen2_nomic_vision.py

@@ -933,7 +933,7 @@ class Qwen2NomicVisionModel(Qwen2NomicVisionPreTrainedModel):
             img_emb = self.vision_model(image).last_hidden_state
             img_embeddings = F.normalize(img_emb[:, 0], p=2, dim=1)
             mix = self.mix_linear(img_embeddings).unsqueeze(1)
-            hidden_states = torch.cat((mix, hidden_states), dim=1)
+            hidden_states = torch.cat((hidden_states, mix), dim=1)
             # hidden_states = hidden_states.concat()
 
         # add hidden states from the last decoder layer