Revert "add comparison of pcc/atol of all outputs"

This reverts commit 8bb73bc.

tests/models/MobileNetV2/test_MobileNetV2.py

@@ -47,7 +47,7 @@ def test_MobileNetV2(record_property, mode, nightly):
     else:
         cc.compile_depth = CompileDepth.TTNN_IR
 
-    tester = ThisTester(model_name, mode, compiler_config=cc, required_atol=0.03)
+    tester = ThisTester(model_name, mode, compiler_config=cc)
     results = tester.test_model()
     if mode == "eval":
         # Print the top 5 predictions

tests/models/autoencoder_linear/test_autoencoder_linear.py

@@ -97,9 +97,7 @@ def test_autoencoder_linear(record_property, mode, nightly):
     if nightly:
         cc.compile_depth = CompileDepth.EXECUTE_OP_BY_OP
 
-    tester = ThisTester(
-        model_name, mode, pcc=0.1, required_atol=0.2, compiler_config=cc
-    )
+    tester = ThisTester(model_name, mode, compiler_config=cc)
     results = tester.test_model()
 
     if mode == "eval":

tests/models/beit/test_beit_image_classification.py

@@ -34,9 +34,6 @@ def append_fake_loss_function(self, outputs):
     def get_results_train(self, model, inputs, outputs):
         return inputs["pixel_values"].grad
 
-    def _extract_outputs(self, output_object):
-        return (output_object.logits,)
-
 
 @pytest.mark.parametrize(
     "mode",
@@ -60,18 +57,14 @@ def test_beit_image_classification(record_property, model_name, mode, nightly):
     else:
         cc.compile_depth = CompileDepth.TTNN_IR
 
-    atol = 0.05
-    tester = ThisTester(model_name, mode, compiler_config=cc, required_atol=atol)
+    tester = ThisTester(model_name, mode, compiler_config=cc)
     results = tester.test_model()
 
     if mode == "eval":
         logits = results.logits
 
         # model predicts one of the 1000 ImageNet classes
         predicted_class_idx = logits.argmax(-1).item()
-        print(
-            "Predicted class:",
-            tester.framework_model.config.id2label[predicted_class_idx],
-        )
+        print("Predicted class:", tester.model.config.id2label[predicted_class_idx])
 
     record_property("torch_ttnn", (tester, results))

tests/models/mgp-str-base/test_mgp_str_base.py

@@ -32,9 +32,6 @@ def _load_inputs(self):
         inputs["pixel_values"] = inputs["pixel_values"].to(torch.bfloat16)
         return inputs
 
-    def _extract_outputs(self, output_object):
-        return output_object.logits
-
 
 @pytest.mark.parametrize(
     "mode",
@@ -55,7 +52,7 @@ def test_mgp_str_base(record_property, mode, nightly):
     else:
         cc.compile_depth = CompileDepth.TTNN_IR
 
-    tester = ThisTester(model_name, mode, relative_atol=0.02, compiler_config=cc)
+    tester = ThisTester(model_name, mode, compiler_config=cc)
     results = tester.test_model()
 
     if mode == "eval":

tests/models/openpose/test_openpose_v2.py

@@ -32,19 +32,17 @@ def get_image_tensor():
     return input_batch
 
 
-# NOTE: If we are to decompose BatchNorm2d (which we must in order to lower to SHLO)
-# we need to run in float32 to maintain precision.
 class ThisTester(ModelTester):
     def _load_model(self):
         # Create PyBuda module from PyTorch model
         model = ptcv_get_model("lwopenpose2d_mobilenet_cmupan_coco", pretrained=True)
-        model = model.to(torch.float32)
+        model = model.to(torch.bfloat16)
         return model
 
     def _load_inputs(self):
         input_batch = [get_image_tensor()]
         batch_input = torch.cat(input_batch, dim=0)
-        batch_input = batch_input.to(torch.float32)
+        batch_input = batch_input.to(torch.bfloat16)
         return batch_input
 
 

tests/models/perceiver_io/test_perceiver_io.py

@@ -34,9 +34,6 @@ def _load_inputs(self):
         }
         return arguments
 
-    def _extract_outputs(self, output_object):
-        return (output_object.logits,)
-
 
 @pytest.mark.parametrize(
     "mode",
@@ -55,7 +52,7 @@ def test_perceiver_io(record_property, mode, nightly):
     else:
         cc.compile_depth = CompileDepth.TTNN_IR
 
-    tester = ThisTester(model_name, mode, relative_atol=0.02, compiler_config=cc)
+    tester = ThisTester(model_name, mode, compiler_config=cc)
     results = tester.test_model()
     if mode == "eval":
         logits = results.logits

tests/models/squeeze_bert/test_squeeze_bert.py

@@ -24,9 +24,6 @@ def _load_inputs(self):
         inputs = self.tokenizer("Hello, my dog is cute", return_tensors="pt")
         return inputs
 
-    def _extract_outputs(self, output_object):
-        return (output_object.logits,)
-
 
 @pytest.mark.parametrize(
     "mode",