MCPcopy Create free account
hub / github.com/RightNow-AI/autokernel / parse_args

Function parse_args

profile.py:826–932  ·  view source on GitHub ↗
()

Source from the content-addressed store, hash-verified

824# ---------------------------------------------------------------------------
825
826def parse_args() -> argparse.Namespace:
827 parser = argparse.ArgumentParser(
828 description=(
829 "AutoKernel Model Profiler -- identify GPU kernel bottlenecks "
830 "in any PyTorch model."
831 ),
832 formatter_class=argparse.RawDescriptionHelpFormatter,
833 epilog=(
834 "Examples:\n"
835 " uv run profile.py --model models/llama_7b.py "
836 "--class-name LlamaModel --input-shape 1,2048 --dtype float16\n"
837 " uv run profile.py --module transformers "
838 "--class-name AutoModelForCausalLM "
839 "--pretrained meta-llama/Llama-2-7b-hf --input-shape 1,2048\n"
840 " uv run profile.py --model my_net.py "
841 "--class-name MyNet --input-shape 8,3,224,224 --dtype float32\n"
842 ),
843 )
844
845 # Model source
846 parser.add_argument(
847 "--model",
848 type=str,
849 default=None,
850 help="Path to a Python file containing the model class.",
851 )
852 parser.add_argument(
853 "--module",
854 type=str,
855 default=None,
856 help="Python module to import the model from (e.g. 'transformers').",
857 )
858 parser.add_argument(
859 "--class-name",
860 type=str,
861 required=True,
862 help="Name of the model class to instantiate.",
863 )
864 parser.add_argument(
865 "--pretrained",
866 type=str,
867 default=None,
868 help="Pretrained model name/path for HuggingFace from_pretrained().",
869 )
870
871 # Input configuration
872 parser.add_argument(
873 "--input-shape",
874 type=str,
875 required=True,
876 help="Comma-separated input shape, e.g. '1,2048' or '8,3,224,224'.",
877 )
878 parser.add_argument(
879 "--dtype",
880 type=str,
881 default="float16",
882 help="Data type: float16, bfloat16, float32 (default: float16).",
883 )

Callers 1

mainFunction · 0.85

Calls

no outgoing calls

Tested by

no test coverage detected