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

Enable fp8e5 harness (#9761)

Browse source 
* fix precision format like fp8e5

* match fp8_e5m2
This commit is contained in:
Chen, Zhentao 2023-12-22 16:59:48 +08:00 committed by GitHub
parent 4c487313f2
commit 7fd7c37e1b
1 changed files with 1 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
python/llm/dev/benchmark/harness/harness_to_leaderboard.py
View file

@ -49,7 +49,7 @@ task_to_metric = dict(
) )
def parse_precision(precision, model="bigdl-llm"): def parse_precision(precision, model="bigdl-llm"):
result = match(r"([a-zA-Z_]*)(\d+)", precision) result = match(r"([a-zA-Z_]+)(\d+)([a-zA-Z_\d]*)", precision)
datatype = result.group(1) datatype = result.group(1)
bit = int(result.group(2)) bit = int(result.group(2))
if bit >= 16: if bit >= 16: