fix export llama to qnn

Summary: Regression from #8107, it causes buck run python binary fails. Then #7691 introduces dependency in source transformation

Reviewed By: larryliu0820

Differential Revision: D69942429

Author: cccclai

examples/models/llama/TARGETS

@@ -96,6 +96,9 @@ runtime.command_alias(
 
 runtime.python_library(
     name = "source_transformation",
+    visibility = [
+        "//executorch/examples/...",
+    ],
     srcs = [
         "source_transformation/apply_spin_quant_r1_r2.py",
         "source_transformation/attention.py",

examples/qualcomm/oss_scripts/llama/TARGETS

@@ -19,6 +19,7 @@ python_library(
     name = "llama_lib",
     srcs = ["llama.py"],
     deps = [
+        "//executorch/examples/models/llama:source_transformation",
         "//caffe2:torch",
         "//executorch/backends/qualcomm/partition:partition",
         "//executorch/backends/qualcomm/quantizer:quantizer",

examples/qualcomm/oss_scripts/llama/llama.py

@@ -1039,10 +1039,7 @@ def _build_parser():
     return parser
 
 
-def main(args) -> None:
-    parser = _build_parser()
-
-    args = parser.parse_args(args)
+def export_llama(args) -> None:
     if args.compile_only and args.pre_gen_pte:
         exit("Cannot set both compile_only and pre_gen_pte as true")
 
@@ -1142,7 +1139,11 @@ def main(args) -> None:
         else:
             raise Exception(e)
 
+def main():
+    parser = _build_parser()
+    args = parser.parse_args()
+    export_llama(args)
 
 # flake8: noqa: C901
 if __name__ == "__main__":
-    main(sys.argv[1:])
+    main()