main() — pytorch Function Reference

Source Code

benchmarks/dynamo/check_memory_compression_ratio.py lines 8–48

def main(args):
    actual = pd.read_csv(args.actual)
    expected = pd.read_csv(args.expected)
    failed = []

    for name in actual["name"]:
        actual_memory_compression = float(
            actual.loc[actual["name"] == name]["compression_ratio"]
        )
        try:
            expected_memory_compression = float(
                expected.loc[expected["name"] == name]["compression_ratio"]
            )
        except TypeError:
            print(f"{name:34} is missing from {args.expected}")
            continue
        if actual_memory_compression >= expected_memory_compression * 0.95:
            status = "PASS"
        else:
            status = "FAIL"
            failed.append(name)
        print(
            f"""
            {name:34}:
                actual_memory_compression={actual_memory_compression:.2f},
                expected_memory_compression={expected_memory_compression:.2f},
                {status}
            """
        )

    if failed:
        print(
            textwrap.dedent(
                f"""
                Error: {len(failed)} models below expected memory compression ratio:
                    {" ".join(failed)}
                If this drop is expected, you can update `{args.expected}`.
                """
            )
        )
        sys.exit(1)