Missing corrections

Author: kawakami-masaki0

tests/pytorch_tests/test_activation_quantizer_holder.py

@@ -72,8 +72,10 @@ def test_activation_quantization_holder_save_and_load(self):
                 quantizer = quantizer_class(**quantizer_args)
                 model = PytorchActivationQuantizationHolder(quantizer)
 
-                # Initialize a random input to quantize between -50 to 50.
-                x = torch.from_numpy(np.random.rand(1, 3, 50, 50). astype(np.float32) * 100 - 50, )
+                # Initialize a random input to quantize between -50 to 50. Input includes positive and negative values.
+                x = torch.rand(1, 3, 50, 50) * 50
+                signs = torch.from_numpy(np.where(np.indices((1, 3, 50, 50)).sum(axis=0) % 2 == 0, 1, -1).astype(np.int8))
+                x = x * signs
                 exp_output_tensor = model(x)
 
                 fx_model = symbolic_trace(model)

tests/pytorch_tests/test_fln_activation_quantizer_holder.py

@@ -110,8 +110,10 @@ def test_fln_activation_quantization_holder_save_and_load(self):
                 quantizer = quantizer_class(**quantizer_args)
                 model = PytorchFLNActivationQuantizationHolder(quantizer, quantization_bypass)
 
-                # Initialize a random input to quantize between -50 to 50.
-                x = torch.from_numpy(np.random.rand(1, 3, 50, 50). astype(np.float32) * 100 - 50, )
+                # Initialize a random input to quantize between -50 to 50. Input includes positive and negative values.
+                x = torch.rand(1, 3, 50, 50) * 50
+                signs = torch.from_numpy(np.where(np.indices((1, 3, 50, 50)).sum(axis=0) % 2 == 0, 1, -1).astype(np.int8))
+                x = x * signs
                 exp_output_tensor = model(x)
 
                 fx_model = symbolic_trace(model)

tests/pytorch_tests/test_preserving_activation_quantizer_holder.py

@@ -110,8 +110,10 @@ def test_preserving_activation_quantization_holder_save_and_load(self):
                 quantizer = quantizer_class(**quantizer_args)
                 model = PytorchPreservingActivationQuantizationHolder(quantizer, quantization_bypass)
 
-                # Initialize a random input to quantize between -50 to 50.
-                x = torch.from_numpy(np.random.rand(1, 3, 50, 50). astype(np.float32) * 100 - 50, )
+                # Initialize a random input to quantize between -50 to 50. Input includes positive and negative values.
+                x = torch.rand(1, 3, 50, 50) * 50
+                signs = torch.from_numpy(np.where(np.indices((1, 3, 50, 50)).sum(axis=0) % 2 == 0, 1, -1).astype(np.int8))
+                x = x * signs
                 exp_output_tensor = model(x)
 
                 fx_model = symbolic_trace(model)