dearwolf/aider
1
0
Fork 0
mirror of https://github.com/Aider-AI/aider.git synced 2025-05-29 08:44:59 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 3

prettier test report

Browse source
This commit is contained in:
Paul Gauthier 2023-06-24 20:54:47 -07:00
parent 436d0dd774
commit e0e680e781
1 changed files with 20 additions and 27 deletions
Download patch file Download diff file Expand all files Collapse all files

39
scripts/benchmark.py
View file

@ -109,6 +109,7 @@ def main():
args.verbose,
)
all_results = run_test_threaded.gather(tqdm=True)
print()
print()
print()
@ -136,33 +137,25 @@ def summarize_results(all_results, total_tests=None):
for i in range(len(results["tests_outcomes"]) - 1, retries):
passed_tests[i] += 1
dump(completed_tests, total_tests)
for i in range(retries):
pass_rate = 100 * passed_tests[i] / completed_tests
dump(i, pass_rate)
total_cost += results["cost"]
dump(total_cost)
avg_cost = total_cost / completed_tests
dump(avg_cost)
projected_cost = avg_cost * total_tests
dump(projected_cost)
print(
f"Cost: ${avg_cost:.4f} average, ${total_cost:.2f} total,"
f" ${projected_cost:.2f} projected"
)
duration += results["duration"]
avg_duration = duration / completed_tests
dump(avg_duration)
min_left = (total_tests - completed_tests) * avg_duration / 60
dump(min_left)
print()
print(f"{completed_tests} test-cases")
for i in range(retries):
pass_rate = 100 * passed_tests[i] / completed_tests
print(f"{pass_rate:.1f}% correct after try {i}")
avg_duration = duration / completed_tests
print(f"{avg_duration:.1f} sec/test-case")
avg_cost = total_cost / completed_tests
projected_cost = avg_cost * total_tests
print(
f"Cost: ${avg_cost:.4f} average, ${total_cost:.2f} total, ${projected_cost:.2f} projected"
)
def run_test(testdir, model_name, edit_format, retries, no_test, verbose):