Reduce computation time for ASYM mode

openvinotoolkit · Jun 11, 2024 · d754a97 · d754a97
1 parent 0c51517
commit d754a97
Showing 1 changed file with 2 additions and 0 deletions.
diff --git a/nncf/quantization/algorithms/weight_compression/scale_estimation.py b/nncf/quantization/algorithms/weight_compression/scale_estimation.py
@@ -165,6 +165,8 @@ def apply(
             original_weight = fns.zeros_like(weight) + weight
 
             compressed_weights, scale, zp = do_integer_quantization(original_weight, reduction_axis, config)
+            if zp is not None:
+                zp = zp.astype(scale.dtype)
             q_weights = do_dequantization(compressed_weights, scale, zp, reduction_axis)
 
             s = fns.unsqueeze(s, 0)