diff --git a/.azure-pipelines/scripts/ut/run_itrex.sh b/.azure-pipelines/scripts/ut/run_itrex.sh
index 111d7175f76..2f9f4ed79c0 100644
--- a/.azure-pipelines/scripts/ut/run_itrex.sh
+++ b/.azure-pipelines/scripts/ut/run_itrex.sh
@@ -11,8 +11,7 @@ bash /intel-extension-for-transformers/.github/workflows/script/prepare_env.sh
 bash /intel-extension-for-transformers/.github/workflows/script/install_binary.sh
 
 # prepare test env
-# tmp install transformers for incompatible issue
-pip install transformers
+sed -i '/neural-compressor.git/d' /intel-extension-for-transformers/tests/requirements.txt
 pip install -r /intel-extension-for-transformers/tests/requirements.txt
 LOG_DIR=/neural-compressor/log_dir
 mkdir -p ${LOG_DIR}
diff --git a/neural_compressor/adaptor/pytorch.py b/neural_compressor/adaptor/pytorch.py
index 314cd800043..3892a659b33 100644
--- a/neural_compressor/adaptor/pytorch.py
+++ b/neural_compressor/adaptor/pytorch.py
@@ -1813,7 +1813,8 @@ def smooth_quant(
             kwargs["percentile"] = percentile
         if scales_per_op is not None:
             kwargs["scales_per_op"] = scales_per_op
-        auto_alpha_args["init_alpha"] = default_alpha
+        if alpha == "auto":
+            auto_alpha_args["init_alpha"] = default_alpha
         model._model = self.sq.transform(
             alpha=alpha,
             folding=folding,