Skip to content

Commit d4129b7

Browse files
Arm backend: Updated how generic evaluator is handled (#14940)
Currently using the generic evaluator leads to a dataset of "None" being used. This patch reverts the dataset to being defaulted to example inputs for the generic evaluator. Change-Id: I7bdb3161d6339eadc50a1d3a73d592119fa87587 cc @freddan80 @per @zingo @oscarandersson8218 @digantdesai
1 parent bdc526b commit d4129b7

File tree

1 file changed

+3
-1
lines changed

1 file changed

+3
-1
lines changed

examples/arm/aot_arm_compiler.py

Lines changed: 3 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -285,7 +285,9 @@ def get_calibration_data(
285285
):
286286
# Firstly, if the model is being evaluated, take the evaluators calibration function if it has one
287287
if evaluator_name is not None:
288-
return evaluator_calibration_data(evaluator_name, evaluator_config)
288+
evaluator_data = evaluator_calibration_data(evaluator_name, evaluator_config)
289+
if evaluator_data is not None:
290+
return evaluator_data
289291

290292
# If the model is in the calibration_data dictionary, get the data from there
291293
# This is used for the simple model examples provided

0 commit comments

Comments
 (0)