Fix onnxrt calibration issue (#1666)

Signed-off-by: Mengni Wang <mengni.wang@intel.com>
intel · Mar 14, 2024 · cd6d244 · cd6d244
1 parent 01d966b
commit cd6d244
Showing 1 changed file with 3 additions and 2 deletions.
diff --git a/neural_compressor/adaptor/ox_utils/calibration.py b/neural_compressor/adaptor/ox_utils/calibration.py
@@ -440,8 +440,9 @@ def get_weight_tensors_calib_range(self):
                 ),
             )
             calibrator = CALIBRATOR["minmax"]()  # use minmax method to calibrate initializer tensors
-            calibrator.collect(initializer_tensor)
-            weight_tensors_calib_range[initializer_tensor_name] = [list(calibrator.calib_range)]
+            if initializer_tensor.flatten().size > 0:
+                calibrator.collect(initializer_tensor)
+                weight_tensors_calib_range[initializer_tensor_name] = [list(calibrator.calib_range)]
             calibrator.clear()
             del calibrator
         return weight_tensors_calib_range