ayo/ipex-llm
1
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 1

bigdl-llm stable version: let the perf test fail if the difference between perf and baseline is greater than 5% (#9750)

Browse source 
* test

* test

* test

* update

* revert
This commit is contained in:
Shaojun Liu 2023-12-25 13:47:11 +08:00 committed by GitHub
parent 986f65cea9
commit b6222404b8
2 changed files with 21 additions and 4 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
.github/workflows/llm_performance_tests_stable_version.yml vendored
View file

@ -150,6 +150,6 @@ jobs:
cp ./*.csv /models/nightly_perf_cpu/
cd ../../../test/benchmark
python -m pip install pandas==1.5.3
python csv_to_html.py -f /models/nightly_perf_cpu/
python csv_to_html.py -f /models/nightly_perf_cpu/ -b /models/nightly_perf_cpu/transformer_int4-results-1baseline.csv -t 5.0

17
python/llm/test/benchmark/csv_to_html.py
View file

@ -30,12 +30,17 @@ def highlight_vals(val, max=3.0):
else:
return ''
def is_diffs_within_normal_range(diff1, diff2, threshold=5.0):
return not any(diff < (-threshold) for diff in diff1 + diff2)
def main():
parser = argparse.ArgumentParser(description="convert .csv file to .html file")
parser.add_argument("-f", "--folder_path", type=str, dest="folder_path",
help="The directory which stores the .csv file", default="/mnt/disk1/nightly_perf_gpu/")
parser.add_argument("-t", "--threshold", type=float, dest="threshold",
help="the threshold of highlight values", default=3.0)
parser.add_argument("-b", "--baseline_path", type=str, dest="baseline_path",
help="the baseline path which stores the baseline.csv file")
args = parser.parse_args()
csv_files = []
@ -50,7 +55,12 @@ def main():
latest_csv = pd.read_csv(csv_files[0], index_col=0)
daily_html=csv_files[0].split(".")[0]+".html"
diffs_within_normal_range = True
if len(csv_files)>1:
if args.baseline_path:
previous_csv = pd.read_csv(args.baseline_path, index_col=0)
else:
previous_csv = pd.read_csv(csv_files[1], index_col=0)
last1=['']*len(latest_csv.index)
@ -96,6 +106,8 @@ def main():
latest_csv.insert(loc=5,column='last2',value=last2)
latest_csv.insert(loc=6,column='diff2(%)',value=diff2)
diffs_within_normal_range = is_diffs_within_normal_range(diff1, diff2, highlight_threshold)
subset=['diff1(%)','diff2(%)']
columns={'1st token avg latency (ms)': '{:.2f}', '2+ avg latency (ms/token)': '{:.2f}', 'last1': '{:.2f}', 'diff1(%)': '{:.2f}',
'last2': '{:.2f}', 'diff2(%)': '{:.2f}', 'encoder time (ms)': '{:.2f}', 'peak mem (GB)': '{:.2f}'}
@ -106,5 +118,10 @@ def main():
else:
latest_csv.to_html(daily_html)
if args.baseline_path and not diffs_within_normal_range:
print("The diffs are outside the normal range: %" + highlight_threshold)
return 1
return 0
if __name__ == "__main__":
sys.exit(main())