Refactor and add Llama Python library build

examples/qualcomm/oss_scripts/llama/TARGETS

@@ -15,6 +15,24 @@ python_library(
     ],
 )
 
+python_library(
+    name = "llama_lib",
+    srcs = ["llama.py"],
+    deps = [
+        "//caffe2:torch",
+        "//executorch/backends/qualcomm/partition:partition",
+        "//executorch/backends/qualcomm/quantizer:quantizer",
+        "//executorch/devtools:lib",
+        "//executorch/examples/models:models",
+        "//executorch/examples/qualcomm/oss_scripts/llama:static_llama",
+        "//executorch/examples/qualcomm:utils",
+        "//executorch/extension/export_util:export_util",
+        "//executorch/extension/llm/custom_ops:model_sharding_py",
+        "//executorch/extension/llm/export:export_lib",
+        "//executorch/extension/pybindings:aten_lib",
+    ],
+)
+
 python_binary(
     name = "llama",
     srcs = ["llama.py"],

examples/qualcomm/oss_scripts/llama/llama.py

@@ -847,7 +847,7 @@ def post_process():
             logging.info(f"Results[{idx}]:\n{output}")
 
 
-def main():
+def _build_parser():
     parser = setup_common_args_and_variables()
     parser.add_argument(
         "-a",
@@ -980,7 +980,13 @@ def main():
         help="Fallback to cpu embedding operator and type of embedding quantization, '<bitwidth>,<groupsize>', e.g., '4,32'.",
     )
 
-    args = parser.parse_args()
+    return parser
+
+
+def main(args) -> None:
+    parser = _build_parser()
+
+    args = parser.parse_args(args)
     if args.compile_only and args.pre_gen_pte:
         exit("Cannot set both compile_only and pre_gen_pte as true")
 
@@ -1071,4 +1077,4 @@ def main():
 
 # flake8: noqa: C901
 if __name__ == "__main__":
-    main()
+    main(sys.argv[1:])