Files
gpt-engineer/scripts/benchmark.py
Anton Osika 7f16abecb3 Make sure benchmark runs and be more strict about collecting feedback (#397)
* Update benchmark script

* Bump version
2023-06-25 15:53:34 +02:00

78 lines
2.0 KiB
Python

# list all folders in benchmark folder
# for each folder, run the benchmark
import contextlib
import os
import subprocess
from itertools import islice
from pathlib import Path
from typing import Iterable, Union
from typer import run
def main(
n_benchmarks: Union[int, None] = None,
):
path = Path("benchmark")
folders: Iterable[Path] = path.iterdir()
if n_benchmarks:
folders = islice(folders, n_benchmarks)
benchmarks = []
for bench_folder in folders:
if os.path.isdir(bench_folder):
print(f"Running benchmark for {bench_folder}")
log_path = bench_folder / "log.txt"
log_file = open(log_path, "w")
process = subprocess.Popen(
[
"python",
"-u", # Unbuffered output
"-m",
"gpt_engineer.main",
bench_folder,
"--steps",
"benchmark",
],
stdout=log_file,
stderr=log_file,
bufsize=0,
)
benchmarks.append((bench_folder, process, log_file))
print("You can stream the log file by running:")
print(f"tail -f {log_path}")
print()
for bench_folder, process, file in benchmarks:
process.wait()
file.close()
print("process", bench_folder.name, "finished with code", process.returncode)
print("Running it. Original benchmark prompt:")
print()
with open(bench_folder / "prompt") as f:
print(f.read())
print()
with contextlib.suppress(KeyboardInterrupt):
subprocess.run(
[
"python",
"-m",
"gpt_engineer.main",
bench_folder,
"--steps",
"evaluate",
],
)
if __name__ == "__main__":
run(main)