Wrap python around with no exit

Author: winskuo-quic

.ci/scripts/test_qnn_static_llama.sh

@@ -33,13 +33,15 @@ cp -f build-x86/backends/qualcomm/PyQnnManagerAdaptor.cpython-310-x86_64-linux-g
 cp -f build-x86/backends/qualcomm/PyQnnWrapperAdaptor.cpython-310-x86_64-linux-gnu.so backends/qualcomm/python
 
 
+set +e
 # Compile only as weight sharing is not applicable on x86
 $PYTHON_EXECUTABLE backends/qualcomm/tests/test_qnn_delegate.py -k TestExampleScript.test_stories_single_llama --model SM8650 --build_folder build-android/ --executorch_root . --artifact_dir . --compile_only
 exit_code1=$?
 
 # Checks accuracy with weight sharing disabled since x86 does not support weight sharing.
 $PYTHON_EXECUTABLE backends/qualcomm/tests/test_qnn_delegate.py -k TestExampleScript.test_stories_single_llama --model SM8650 --build_folder build-android/ --executorch_root . --artifact_dir . --compile_only
 exit_code2=$?
+set -e
 
 # Check the exit codes and print messages
 if [ $exit_code1 -ne 0 ]; then

backends/qualcomm/tests/test_qnn_delegate.py

@@ -3804,7 +3804,7 @@ def test_stories_single_llama(self):
         if self.enable_x86_64:
             cmds.extend(["--enable_x86_64"])
 
-        golden_start_with = "wOnce upon a time,"
+        golden_start_with = "sOnce upon a time,"
         p = subprocess.Popen(cmds, stdout=subprocess.DEVNULL)
         with Listener((self.ip, self.port)) as listener:
             conn = listener.accept()
@@ -3819,7 +3819,7 @@ def test_stories_single_llama(self):
                 # x86 does not allow weight sharing, so we don't check pte size
                 if not self.enable_x86_64:
                     pte_size = msg["pte_size"]
-                    self.assertLessEqual(pte_size, 130000000)
+                    self.assertLessEqual(pte_size, 13) # 130000000
 
     @unittest.skip("dynamic shape inputs appear in recent torch.export.export")
     def test_mobilebert(self):