diff --git a/.github/workflows/unit_tests.yml b/.github/workflows/unit_tests.yml
index 921f5083a..aba4e5b5c 100644
--- a/.github/workflows/unit_tests.yml
+++ b/.github/workflows/unit_tests.yml
@@ -51,8 +51,8 @@ on:
         description: 'Wheel Build Server'
         type: choice
         options:
-          - '["self-hosted", "zen4"]'
           - '["self-hosted", "xeon5"]'
+          - '["self-hosted", "zen4"]'
 
 env:
   CUDA_DEVICE_ORDER: PCI_BUS_ID
@@ -347,20 +347,20 @@ jobs:
           echo "sha256=$sha256"
           echo "SOURCE_DOWNLOADED=1" >> $GITHUB_ENV
 
-      - name: Download source from github
-        if: env.SOURCE_DOWNLOADED == '' && !cancelled()
-        uses: actions/download-artifact@v4
-        with:
-          name: source
-          path: dist
-          run-id: ${{ github.run_id }}
-
-      - name: Uncompress source
-        continue-on-error: true
-        run: |
-          find . -mindepth 1 ! -name "gptqmodel_source.tar.gz" -exec rm -rf {} +
-          ls -ahl .
-          tar -zxf gptqmodel_source.tar.gz
+      # - name: Download source from github
+      #   if: env.SOURCE_DOWNLOADED == '' && !cancelled()
+      #   uses: actions/download-artifact@v4
+      #   with:
+      #     name: source
+      #     path: dist
+      #     run-id: ${{ github.run_id }}
+
+#      - name: Uncompress source
+ #       continue-on-error: true
+  #      run: |
+   #       find . -mindepth 1 ! -name "gptqmodel_source.tar.gz" -exec rm -rf {} +
+    #      ls -ahl .
+     #     tar -zxf gptqmodel_source.tar.gz
 
       - name: Download wheel from local
         continue-on-error: true
@@ -415,11 +415,13 @@ jobs:
           timestamp=$(date +%s%3N)
           gpu_id=-1
 
+          url="http://$XEON5/gpu/get?id=${{ github.run_id }}&timestamp=$timestamp&test=${{ matrix.test_script }}&runner=${RUNNER_NAME}&exclusive=${{ github.event.inputs.exclusive-gpu }}"
+          echo "$url"
           while [ "$gpu_id" -lt 0 ]; do
-            gpu_id=$(curl -s "http://$RUNNER/gpu/get?id=${{ github.run_id }}&timestamp=$timestamp&test=${{ matrix.test_script }}&runner=${RUNNER_NAME}&exclusive=${{ github.event.inputs.exclusive-gpu }}")
+            gpu_id=$(curl -s "$url")
 
             if [ "$gpu_id" -lt 0 ]; then
-              echo "http://$RUNNER/gpu/get?id=${{ github.run_id }}&timestamp=$timestamp&test=${{ matrix.test_script }}&runner=${RUNNER_NAME}&exclusive=${{ github.event.inputs.exclusive-gpu }} returned $gpu_id"
+              echo "http://$XEON5/gpu/get?id=${{ github.run_id }}&timestamp=$timestamp&test=${{ matrix.test_script }}&runner=${RUNNER_NAME}&exclusive=${{ github.event.inputs.exclusive-gpu }} returned $gpu_id"
               echo "No available GPU, waiting 5 seconds..."
               curl http://$XEON5/gpu/status2
               sleep 5
@@ -446,7 +448,7 @@ jobs:
 
       - name: Release GPU
         if: always() && !contains(matrix.test_script, 'ipex')
-        run: curl -X GET "http://$RUNNER/gpu/release?id=${{ github.run_id }}&gpu=${{ env.CUDA_VISIBLE_DEVICES }}&timestamp=${{ env.STEP_TIMESTAMP }}&test=${{ matrix.test_script }}&runner=${RUNNER_NAME}"
+        run: curl -X GET "http://$XEON5/gpu/release?id=${{ github.run_id }}&gpu=${{ env.CUDA_VISIBLE_DEVICES }}&timestamp=${{ env.STEP_TIMESTAMP }}&test=${{ matrix.test_script }}&runner=${RUNNER_NAME}"
 
       - name: Clean cache
         if: always()
@@ -510,20 +512,20 @@ jobs:
           echo "sha256=$sha256"
           echo "SOURCE_DOWNLOADED=1" >> $GITHUB_ENV
 
-      - name: Download source from github
-        if: env.SOURCE_DOWNLOADED == '' && !cancelled()
-        uses: actions/download-artifact@v4
-        with:
-          name: source
-          path: dist
-          run-id: ${{ github.run_id }}
-
-      - name: Uncompress source
-        continue-on-error: true
-        run: |
-          find . -mindepth 1 ! -name "gptqmodel_source.tar.gz" -exec rm -rf {} +
-          ls -ahl .
-          tar -zxf gptqmodel_source.tar.gz
+      # - name: Download source from github
+      #   if: env.SOURCE_DOWNLOADED == '' && !cancelled()
+      #   uses: actions/download-artifact@v4
+      #   with:
+      #     name: source
+      #     path: dist
+      #     run-id: ${{ github.run_id }}
+
+ #     - name: Uncompress source
+  #      continue-on-error: true
+   #     run: |
+    #      find . -mindepth 1 ! -name "gptqmodel_source.tar.gz" -exec rm -rf {} +
+     #     ls -ahl .
+      #    tar -zxf gptqmodel_source.tar.gz
 
       - name: Download wheel from local
         continue-on-error: true
@@ -627,11 +629,12 @@ jobs:
           timestamp=$(date +%s%3N)
           gpu_id=-1
 
+          url="http://$XEON5/gpu/get?id=${{ github.run_id }}&timestamp=$timestamp&test=${{ matrix.test_script }}&runner=${RUNNER_NAME}&exclusive=${{ github.event.inputs.exclusive-gpu }}"
+          echo "$url"
           while [ "$gpu_id" -lt 0 ]; do
-            gpu_id=$(curl -s "http://$RUNNER/gpu/get?id=${{ github.run_id }}&timestamp=$timestamp&test=${{ matrix.test_script }}&runner=${RUNNER_NAME}&exclusive=${{ github.event.inputs.exclusive-gpu }}")
-
+            gpu_id=$(curl -s "$url")
             if [ "$gpu_id" -lt 0 ]; then
-              echo "http://$RUNNER/gpu/get?id=${{ github.run_id }}&timestamp=$timestamp&test=${{ matrix.test_script }}&runner=${RUNNER_NAME}&exclusive=${{ github.event.inputs.exclusive-gpu }} returned $gpu_id"
+              echo "http://$XEON5/gpu/get?id=${{ github.run_id }}&timestamp=$timestamp&test=${{ matrix.test_script }}&runner=${RUNNER_NAME}&exclusive=${{ github.event.inputs.exclusive-gpu }} returned $gpu_id"
               echo "No available GPU, waiting 5 seconds..."
               curl http://$XEON5/gpu/status2
               sleep 5
@@ -664,7 +667,7 @@ jobs:
           pytest --durations=0 tests/${{ matrix.test_script }}.py || { echo "ERROR=1" >> $GITHUB_ENV; exit 1; }
           execution_time=$(( $(date +%s) - start_time ))
           echo "$((execution_time / 60))m $((execution_time % 60))s"
-          curl "http://$RUNNER/gpu/log_test_vram?id=${{ github.run_id }}&gpu=${{ env.CUDA_VISIBLE_DEVICES }}&range=$execution_time&unit=second&test=${{ matrix.test_script }}"
+          curl "http://$XEON5/gpu/log_test_vram?id=${{ github.run_id }}&gpu=${{ env.CUDA_VISIBLE_DEVICES }}&range=$execution_time&unit=second&test=${{ matrix.test_script }}"
 
       - name: Release GPU
         if: always() && !contains(matrix.test_script, 'ipex') && !contains(matrix.test_script, 'xpu')