From 7181117f3ceb3f969e89e7add9bbe535f686fece Mon Sep 17 00:00:00 2001 From: Andrew Song <40076917+a-ys@users.noreply.github.com> Date: Wed, 18 Sep 2024 14:02:07 -0700 Subject: [PATCH] [cherry-pick] [neo] Fix calib_size dtype bug (#2397) (#2398) --- serving/docker/partition/partition.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/serving/docker/partition/partition.py b/serving/docker/partition/partition.py index 05ffcb749..768fbb364 100644 --- a/serving/docker/partition/partition.py +++ b/serving/docker/partition/partition.py @@ -328,7 +328,7 @@ def autofp8_quantize(self): # If using dynamic activation scales, a calibration dataset is not required examples = [] else: - calib_size = self.properties.get("option.calib_size", 512) + calib_size = int(self.properties.get("option.calib_size", 512)) # Tokenize dataset for calibrating static activation scales ds = load_dataset("abisee/cnn_dailymail", "3.0.0",