diff --git a/fastvideo/models/hunyuan/modules/attenion.py b/fastvideo/models/hunyuan/modules/attenion.py
index 4329834d1..432d75697 100644
--- a/fastvideo/models/hunyuan/modules/attenion.py
+++ b/fastvideo/models/hunyuan/modules/attenion.py
@@ -172,7 +172,7 @@ def parallel_attention(q, k, v, img_q_len, img_kv_len, text_mask, mask_strategy=
                 for i, head_idx in enumerate(heads):
                     hidden_states[:, head_idx:head_idx + 1, :, :] = processed_heads[:, i:i + 1, :, :]
                 
-                hidden_states = hidden_states.transpose(1, 2)
+            hidden_states = hidden_states.transpose(1, 2)
     else:
         query = torch.cat([query, encoder_query], dim=1)
         key = torch.cat([key, encoder_key], dim=1)
diff --git a/scripts/inference/inference_hunyuan_STA.sh b/scripts/inference/inference_hunyuan_STA.sh
index e17f284b2..efde6992b 100644
--- a/scripts/inference/inference_hunyuan_STA.sh
+++ b/scripts/inference/inference_hunyuan_STA.sh
@@ -45,5 +45,4 @@ CUDA_VISIBLE_DEVICES=1 torchrun --nnodes=1 --nproc_per_node=$num_gpus --master_p
     --model_path $MODEL_BASE \
     --mask_strategy_file_path $mask_strategy_file_path \
     --dit-weight ${MODEL_BASE}/hunyuan-video-t2v-720p/transformers/mp_rank_00_model_states.pt \
-    --vae-sp \
-    --enable_torch_compile
\ No newline at end of file
+    --vae-sp
\ No newline at end of file