Support Int32 datatype

[brataTT]

Problem description

Follow up of tt-mlir Commit a1af497

After tt-mlir Commit a1af497 added support for Int32 dtype, tt-torch tests fail when it can't convert the new type to a pytorch dtype.

What's changed

Added conversion support between TT::Int32 to Pytorch::Int cherry-picked from @jnie-TT's fix branch

Checklist

• [-] New/Existing tests provide coverage for changes
• Wait for referenced commit to be uplifted

[codecov-commenter]

❌ 17 Tests Failed:

Tests completed	Failed	Passed	Skipped
483	17	466	7

View the full list of 3 ❄️ flaky tests

tests.models.torchvision.test_torchvision_image_classification::test_torchvision_image_classification[full-eval-mobilenet_v2]

Flake rate in main: 28.57% (Passed 5 times, Failed 2 times)

Stack Traces | 11.6s run time

record_property = <function record_property.<locals>.append_property at 0x7fb6a9947740>
model_info = ('mobilenet_v2', 'MobileNet_V2_Weights'), mode = 'eval'
op_by_op = False

    @pytest.mark.parametrize(
        "model_info", model_info_list, ids=[info[0] for info in model_info_list]
    )
    @pytest.mark.parametrize("mode", ["train", "eval"])
    @pytest.mark.parametrize("op_by_op", [True, False], ids=["op_by_op", "full"])
    def test_torchvision_image_classification(record_property, model_info, mode, op_by_op):
        if mode == "train":
            pytest.skip()
    
        cc = CompilerConfig()
        cc.enable_consteval = True
        cc.consteval_parameters = True
        if op_by_op:
            cc.compile_depth = CompileDepth.EXECUTE_OP_BY_OP
    
        tester = ThisTester(
            model_info,
            mode,
            required_pcc=0.98,
            assert_pcc=True,
            assert_atol=False,
            compiler_config=cc,
            record_property_handle=record_property,
        )
>       results = tester.test_model()

.../models/torchvision/test_torchvision_image_classification.py:117: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
tests/utils.py:274: in test_model
    return self.test_model_eval(on_device)
.../venv/lib/python3.11.../torch/utils/_contextlib.py:116: in decorate_context
    return func(*args, **kwargs)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <test_torchvision_image_classification.ThisTester object at 0x7fb739f4e110>
on_device = True

    @torch.no_grad()
    def test_model_eval(self, on_device=True):
        model = (
            self.framework_model.eval()
            if hasattr(self.framework_model, "eval")
            else self.framework_model
        )
        golden = self.get_golden_outputs(model, self.inputs)
    
        if on_device == True:
            model = self.compile_model(model, self.compiler_config)
    
        outputs = self.run_model(model, self.inputs)
        assert type(outputs) == type(
            golden
        ), "Expecting the type of both calculated and golden to be identical. Whether that be a tensor, list, dictonary, etc."
    
        passed_pcc, passed_atol, msg, err_msg, pccs, atols = verify_against_golden(
            self._extract_outputs(golden),
            self._extract_outputs(outputs),
            self.required_pcc,
            self.required_atol,
            self.relative_atol,
        )
        self.record_tag_cache["pccs"] = pccs
        self.record_tag_cache["atols"] = atols
    
        if self.assert_pcc and self.assert_atol:
            if passed_pcc and passed_atol:
                print(msg)
            else:
                assert False, err_msg
        elif not self.assert_pcc and self.assert_atol:
            print("Ignoring PCC check\n")
            if passed_atol:
                print(msg)
            else:
                assert False, err_msg
        elif self.assert_pcc and not self.assert_atol:
            print("Ignoring ATOL check\n")
            if passed_pcc:
                print(msg)
            else:
>               assert False, err_msg
E               AssertionError: PCC of output 0: 0.0439, threshold: 0.98 ❌
E               ATOL of output 0: 7.0625, threshold: 0.01 ❌

tests/utils.py:263: AssertionError

tests.models.torchvision.test_torchvision_image_classification::test_torchvision_image_classification[full-eval-resnext50_32x4d]

Flake rate in main: 28.57% (Passed 5 times, Failed 2 times)

Stack Traces | 30.3s run time

record_property = <function record_property.<locals>.append_property at 0x7fb70c3859e0>
model_info = ('resnext50_32x4d', 'ResNeXt50_32X4D_Weights'), mode = 'eval'
op_by_op = False

    @pytest.mark.parametrize(
        "model_info", model_info_list, ids=[info[0] for info in model_info_list]
    )
    @pytest.mark.parametrize("mode", ["train", "eval"])
    @pytest.mark.parametrize("op_by_op", [True, False], ids=["op_by_op", "full"])
    def test_torchvision_image_classification(record_property, model_info, mode, op_by_op):
        if mode == "train":
            pytest.skip()
    
        cc = CompilerConfig()
        cc.enable_consteval = True
        cc.consteval_parameters = True
        if op_by_op:
            cc.compile_depth = CompileDepth.EXECUTE_OP_BY_OP
    
        tester = ThisTester(
            model_info,
            mode,
            required_pcc=0.98,
            assert_pcc=True,
            assert_atol=False,
            compiler_config=cc,
            record_property_handle=record_property,
        )
>       results = tester.test_model()

.../models/torchvision/test_torchvision_image_classification.py:117: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
tests/utils.py:274: in test_model
    return self.test_model_eval(on_device)
.../venv/lib/python3.11.../torch/utils/_contextlib.py:116: in decorate_context
    return func(*args, **kwargs)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <test_torchvision_image_classification.ThisTester object at 0x7fb739bcaad0>
on_device = True

    @torch.no_grad()
    def test_model_eval(self, on_device=True):
        model = (
            self.framework_model.eval()
            if hasattr(self.framework_model, "eval")
            else self.framework_model
        )
        golden = self.get_golden_outputs(model, self.inputs)
    
        if on_device == True:
            model = self.compile_model(model, self.compiler_config)
    
        outputs = self.run_model(model, self.inputs)
        assert type(outputs) == type(
            golden
        ), "Expecting the type of both calculated and golden to be identical. Whether that be a tensor, list, dictonary, etc."
    
        passed_pcc, passed_atol, msg, err_msg, pccs, atols = verify_against_golden(
            self._extract_outputs(golden),
            self._extract_outputs(outputs),
            self.required_pcc,
            self.required_atol,
            self.relative_atol,
        )
        self.record_tag_cache["pccs"] = pccs
        self.record_tag_cache["atols"] = atols
    
        if self.assert_pcc and self.assert_atol:
            if passed_pcc and passed_atol:
                print(msg)
            else:
                assert False, err_msg
        elif not self.assert_pcc and self.assert_atol:
            print("Ignoring PCC check\n")
            if passed_atol:
                print(msg)
            else:
                assert False, err_msg
        elif self.assert_pcc and not self.assert_atol:
            print("Ignoring ATOL check\n")
            if passed_pcc:
                print(msg)
            else:
>               assert False, err_msg
E               AssertionError: PCC of output 0: -0.0611, threshold: 0.98 ❌
E               ATOL of output 0: 768.0000, threshold: 0.01 ❌

tests/utils.py:263: AssertionError

tests.models.torchvision.test_torchvision_image_classification::test_torchvision_image_classification[full-eval-regnet_x_400mf]

Flake rate in main: 28.57% (Passed 5 times, Failed 2 times)

Stack Traces | 33.2s run time

record_property = <function record_property.<locals>.append_property at 0x7fb6aac43240>
model_info = ('regnet_x_400mf', 'RegNet_X_400MF_Weights'), mode = 'eval'
op_by_op = False

    @pytest.mark.parametrize(
        "model_info", model_info_list, ids=[info[0] for info in model_info_list]
    )
    @pytest.mark.parametrize("mode", ["train", "eval"])
    @pytest.mark.parametrize("op_by_op", [True, False], ids=["op_by_op", "full"])
    def test_torchvision_image_classification(record_property, model_info, mode, op_by_op):
        if mode == "train":
            pytest.skip()
    
        cc = CompilerConfig()
        cc.enable_consteval = True
        cc.consteval_parameters = True
        if op_by_op:
            cc.compile_depth = CompileDepth.EXECUTE_OP_BY_OP
    
        tester = ThisTester(
            model_info,
            mode,
            required_pcc=0.98,
            assert_pcc=True,
            assert_atol=False,
            compiler_config=cc,
            record_property_handle=record_property,
        )
>       results = tester.test_model()

.../models/torchvision/test_torchvision_image_classification.py:117: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
tests/utils.py:274: in test_model
    return self.test_model_eval(on_device)
.../venv/lib/python3.11.../torch/utils/_contextlib.py:116: in decorate_context
    return func(*args, **kwargs)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <test_torchvision_image_classification.ThisTester object at 0x7fb6a2192f10>
on_device = True

    @torch.no_grad()
    def test_model_eval(self, on_device=True):
        model = (
            self.framework_model.eval()
            if hasattr(self.framework_model, "eval")
            else self.framework_model
        )
        golden = self.get_golden_outputs(model, self.inputs)
    
        if on_device == True:
            model = self.compile_model(model, self.compiler_config)
    
        outputs = self.run_model(model, self.inputs)
        assert type(outputs) == type(
            golden
        ), "Expecting the type of both calculated and golden to be identical. Whether that be a tensor, list, dictonary, etc."
    
        passed_pcc, passed_atol, msg, err_msg, pccs, atols = verify_against_golden(
            self._extract_outputs(golden),
            self._extract_outputs(outputs),
            self.required_pcc,
            self.required_atol,
            self.relative_atol,
        )
        self.record_tag_cache["pccs"] = pccs
        self.record_tag_cache["atols"] = atols
    
        if self.assert_pcc and self.assert_atol:
            if passed_pcc and passed_atol:
                print(msg)
            else:
                assert False, err_msg
        elif not self.assert_pcc and self.assert_atol:
            print("Ignoring PCC check\n")
            if passed_atol:
                print(msg)
            else:
                assert False, err_msg
        elif self.assert_pcc and not self.assert_atol:
            print("Ignoring ATOL check\n")
            if passed_pcc:
                print(msg)
            else:
>               assert False, err_msg
E               AssertionError: PCC of output 0: 0.7462, threshold: 0.98 ❌
E               ATOL of output 0: 3.8438, threshold: 0.01 ❌

tests/utils.py:263: AssertionError

To view more test analytics, go to the Test Analytics Dashboard
_{📋 Got 3 mins? Take this short survey to help us improve Test Analytics.}