Fix inception_v3 Accuracy Regression Issue (#1200)

Signed-off-by: zehao-intel <[email protected]>
intel · Aug 30, 2023 · db5dfa0 · db5dfa0
1 parent e33a753
commit db5dfa0
Showing 1 changed file with 2 additions and 1 deletion.
diff --git a/...ples/tensorflow/image_recognition/tensorflow_models/inception_v3/quantization/ptq/main.py b/...ples/tensorflow/image_recognition/tensorflow_models/inception_v3/quantization/ptq/main.py
@@ -113,7 +113,8 @@ def run(self):
             eval_dataloader = create_dataloader('tensorflow', eval_dataloader_args)
             op_name_dict = {'v0/cg/conv0/conv2d/Conv2D': {
                'activation':  {'dtype': ['fp32']}}}
-            conf = PostTrainingQuantConfig(calibration_sampling_size=[50, 100], op_name_dict=op_name_dict)
+            conf = PostTrainingQuantConfig(calibration_sampling_size=[50, 100], \
+                                op_name_dict=op_name_dict, excluded_precisions=['bf16'])
             from neural_compressor import Metric
             top1 = Metric(name="topk", k=1)
             q_model = quantization.fit(args.input_graph, conf=conf, calib_dataloader=calib_dataloader,