Merge pull request #589 from InfiniTensor/issue/566

issue/566: 添加 Python 算子测试

test/infinicore/ops/bitwise_right_shift.py

+import sys
+import os
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
+
+import torch
+import infinicore
+from framework.base import BaseOperatorTest, TensorSpec, TestCase
+from framework.runner import GenericTestRunner
+from framework.utils import is_broadcast
+
+# Test cases format: (shape, a_strides_or_None, b_strides_or_None, out_strides_or_None)
+
+_TEST_CASES_DATA = [
+    ((8, 8), None, None, None),
+    ((8, 8), (16, 1), None, None),
+    ((8, 8), None, (0, 1), None),
+    ((4, 1), None, None, (8, 1)),
+    ((2, 3, 4), None, None, None),
+    ((16, 512), (1024, 1), (0, 1), None),
+]
+
+_TOLERANCE_MAP = {
+    infinicore.int32: {"atol": 0, "rtol": 0},
+    infinicore.int64: {"atol": 0, "rtol": 0},
+    infinicore.uint8: {"atol": 0, "rtol": 0},
+}
+
+_TENSOR_DTYPES = [infinicore.int32, infinicore.int64, infinicore.uint8]
+
+
+def parse_test_cases():
+    test_cases = []
+    for data in _TEST_CASES_DATA:
+        shape, a_strides, b_strides, out_strides = data[0], data[1], data[2], data[3]
+
+        a_supports_inplace = not is_broadcast(a_strides)
+        b_supports_inplace = not is_broadcast(b_strides)
+        out_supports_inplace = not is_broadcast(out_strides)
+
+        for dtype in _TENSOR_DTYPES:
+            tol = _TOLERANCE_MAP.get(dtype, {"atol": 0, "rtol": 0})
+            a_spec = TensorSpec.from_tensor(shape, a_strides, dtype)
+            b_spec = TensorSpec.from_tensor(shape, b_strides, dtype)
+            out_spec = TensorSpec.from_tensor(shape, out_strides, dtype)
+
+            test_cases.append(
+                TestCase(
+                    inputs=[a_spec, b_spec],
+                    kwargs={},
+                    output_spec=None,
+                    comparison_target=None,
+                    tolerance=tol,
+                    description="Bitwise right shift - OUT_OF_PLACE",
+                )
+            )
+
+            if out_supports_inplace:
+                test_cases.append(
+                    TestCase(
+                        inputs=[a_spec, b_spec],
+                        kwargs=None,
+                        output_spec=out_spec,
+                        comparison_target="out",
+                        tolerance=tol,
+                        description="Bitwise right shift - INPLACE(out)",
+                    )
+                )
+
+            if a_supports_inplace:
+                test_cases.append(
+                    TestCase(
+                        inputs=[a_spec, b_spec],
+                        kwargs={"out": 0},
+                        output_spec=None,
+                        comparison_target=0,
+                        tolerance=tol,
+                        description="Bitwise right shift - INPLACE(a)",
+                    )
+                )
+
+            if b_supports_inplace:
+                test_cases.append(
+                    TestCase(
+                        inputs=[a_spec, b_spec],
+                        kwargs={"out": 1},
+                        output_spec=None,
+                        comparison_target=1,
+                        tolerance=tol,
+                        description="Bitwise right shift - INPLACE(b)",
+                    )
+                )
+
+    return test_cases
+
+
+class OpTest(BaseOperatorTest):
+    """BitwiseRightShift operator test with simplified implementation"""
+
+    def __init__(self):
+        super().__init__("BitwiseRightShift")
+
+    def get_test_cases(self):
+        return parse_test_cases()
+
+    def torch_operator(self, *args, **kwargs):
+        return torch.bitwise_right_shift(*args, **kwargs)
+
+    # def infinicore_operator(self, *args, **kwargs):
+    #     """InfiniCore implementation (operator not yet available)."""
+    #     return infinicore.bitwise_right_shift(*args, **kwargs)
+
+
+def main():
+    """Main entry point"""
+    runner = GenericTestRunner(OpTest)
+    runner.run_and_exit()
+
+
+if __name__ == "__main__":
+    main()

test/infinicore/ops/bitwise_xor.py

@@ -31,7 +31,7 @@ _TEST_CASES_DATA = [
     ((16, 5632), (13312, 1), (13312, 1), None),
 ]
 
-# Tolerance configuration - exact match required for bitwise operations
+# Tolerance configuration
 _TOLERANCE_MAP = {
     infinicore.int8: {"atol": 0, "rtol": 0},
     infinicore.int16: {"atol": 0, "rtol": 0},
@@ -41,14 +41,14 @@ _TOLERANCE_MAP = {
     infinicore.bool: {"atol": 0, "rtol": 0},
 }
 
-# Data types to test - integer types for bitwise operations
+# Data types to test
 _TENSOR_DTYPES = [
     infinicore.int8,
     infinicore.int16,
     infinicore.int32,
     infinicore.int64,
     infinicore.uint8,
-    infinicore.bool,  # XOR also supports boolean tensors
+    infinicore.bool,
 ]
 
 

test/infinicore/ops/block_diag.py

+import sys
+import os
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
+
+import torch
+import infinicore
+from framework.base import BaseOperatorTest, TensorSpec, TestCase
+from framework.runner import GenericTestRunner
+
+# Test cases format: (list_of_matrix_shapes, list_of_strides_or_None, dtype)
+
+_TEST_CASES_DATA = [
+    ([(3, 4), (2, 2)], [None, None], None),
+    ([(1, 1), (1, 1), (1, 1)], [None, None, None], None),
+    ([(4, 4), (2, 3), (3, 2)], [None, (6, 1), (0, 3)], None),
+    ([(8, 8)], [None], None),
+    ([(5, 2), (2, 5)], [(10, 1), None], None),
+    ([(6, 6), (6, 6), (6, 6)], [None, None, None], None),
+]
+
+_TOLERANCE_MAP = {
+    infinicore.float16: {"atol": 0, "rtol": 1e-2},
+    infinicore.float32: {"atol": 0, "rtol": 1e-3},
+    infinicore.bfloat16: {"atol": 0, "rtol": 5e-2},
+}
+
+_TENSOR_DTYPES = [infinicore.float16, infinicore.bfloat16, infinicore.float32]
+
+
+def parse_test_cases():
+    test_cases = []
+
+    for shapes, strides_list, _ in _TEST_CASES_DATA:
+        # prepare TensorSpec list for inputs
+        for dtype in _TENSOR_DTYPES:
+            tol = _TOLERANCE_MAP.get(dtype, {"atol": 0, "rtol": 1e-3})
+            input_specs = []
+            for s, st in zip(shapes, strides_list):
+                input_specs.append(TensorSpec.from_tensor(s, st, dtype))
+
+            test_cases.append(
+                TestCase(
+                    inputs=input_specs,
+                    kwargs={},
+                    output_spec=None,
+                    comparison_target=None,
+                    tolerance=tol,
+                    description="block_diag - OUT_OF_PLACE",
+                )
+            )
+
+    return test_cases
+
+
+class OpTest(BaseOperatorTest):
+    """block_diag operator test with simplified implementation"""
+
+    def __init__(self):
+        super().__init__("block_diag")
+
+    def get_test_cases(self):
+        return parse_test_cases()
+
+    def torch_operator(self, *args, **kwargs):
+        return torch.block_diag(*args, **kwargs)
+
+    # def infinicore_operator(self, *args, **kwargs):
+    #     """InfiniCore implementation (operator not yet available)."""
+    #     return infinicore.block_diag(*args, **kwargs)
+
+
+def main():
+    """Main entry point"""
+    runner = GenericTestRunner(OpTest)
+    runner.run_and_exit()
+
+
+if __name__ == "__main__":
+    main()

test/infinicore/ops/broadcast_to.py

+import sys
+import os
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
+
+import torch
+import infinicore
+from framework.base import BaseOperatorTest, TensorSpec, TestCase
+from framework.runner import GenericTestRunner
+from framework.utils import is_broadcast
+
+# Test cases format: (input_shape, input_strides_or_None, target_shape)
+_TEST_CASES_DATA = [
+    ((3, 1), None, (3, 4)),
+    ((1,), None, (5,)),
+    ((2, 1, 4), (8, 1, 1), (2, 3, 4)),
+    ((4, 1), None, (4, 6)),
+    ((1, 1), None, (2, 3)),
+    ((2,), None, (2, 2)),
+]
+
+_TOLERANCE_MAP = {infinicore.float32: {"atol": 1e-5, "rtol": 1e-4}}
+_TENSOR_DTYPES = [infinicore.float32]
+
+
+def parse_test_cases():
+    test_cases = []
+    for data in _TEST_CASES_DATA:
+        shape, strides, target = data
+
+        for dtype in _TENSOR_DTYPES:
+            tol = _TOLERANCE_MAP.get(dtype)
+            input_spec = TensorSpec.from_tensor(shape, strides, dtype)
+
+            kwargs = {"size": target}
+
+            test_cases.append(
+                TestCase(
+                    inputs=[input_spec],
+                    kwargs=kwargs,
+                    output_spec=None,
+                    comparison_target=None,
+                    tolerance=tol,
+                    description="broadcast_to - OUT_OF_PLACE",
+                )
+            )
+
+    return test_cases
+
+
+class OpTest(BaseOperatorTest):
+    """BroadcastTo operator test with simplified implementation"""
+
+    def __init__(self):
+        super().__init__("BroadcastTo")
+
+    def get_test_cases(self):
+        return parse_test_cases()
+
+    def torch_operator(self, *args, **kwargs):
+        return torch.broadcast_to(*args, **kwargs)
+
+    # def infinicore_operator(self, *args, **kwargs):
+    #     """InfiniCore implementation (operator not yet available)."""
+    #     return infinicore.broadcast_to(*args, **kwargs)
+
+
+def main():
+    """Main entry point"""
+    runner = GenericTestRunner(OpTest)
+    runner.run_and_exit()
+
+
+if __name__ == "__main__":
+    main()

test/infinicore/ops/bucketize.py

+import sys
+import os
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
+
+import torch
+import infinicore
+from framework.base import BaseOperatorTest, TensorSpec, TestCase
+from framework.runner import GenericTestRunner
+
+# Test cases format: (input_shape, input_strides_or_None, boundaries_len)
+_TEST_CASES_DATA = [
+    ((5,), None, 3),
+    ((4, 3), None, 4),
+    ((2,), None, 5),
+    ((6,), None, 2),
+    ((3, 3), None, 6),
+    ((1,), None, 1),
+]
+
+_TOLERANCE_MAP = {infinicore.int64: {"atol": 0, "rtol": 0}}
+_TENSOR_DTYPES = [infinicore.float32]
+
+
+def parse_test_cases():
+    test_cases = []
+    for shape, strides, b_len in _TEST_CASES_DATA:
+        input_spec = TensorSpec.from_tensor(shape, strides, infinicore.float32)
+        boundaries_spec = TensorSpec.from_tensor((b_len,), None, infinicore.float32)
+
+        test_cases.append(
+            TestCase(
+                inputs=[input_spec, boundaries_spec],
+                kwargs={},
+                output_spec=None,
+                comparison_target=None,
+                tolerance=_TOLERANCE_MAP[infinicore.int64],
+                description="bucketize - OUT_OF_PLACE",
+            )
+        )
+
+    return test_cases
+
+
+class OpTest(BaseOperatorTest):
+    """Bucketize operator test with simplified implementation"""
+
+    def __init__(self):
+        super().__init__("Bucketize")
+
+    def get_test_cases(self):
+        return parse_test_cases()
+
+    def torch_operator(self, *args, **kwargs):
+        return torch.bucketize(*args, **kwargs)
+
+    # def infinicore_operator(self, *args, **kwargs):
+    #     """InfiniCore implementation (operator not yet available)."""
+    #     return infinicore.bucketize(*args, **kwargs)
+
+
+def main():
+    """Main entry point"""
+    runner = GenericTestRunner(OpTest)
+    runner.run_and_exit()
+
+
+if __name__ == "__main__":
+    main()

test/infinicore/ops/cat.py

@@ -89,7 +89,7 @@ def parse_test_cases():
                     output_spec=None,
                     comparison_target=None,
                     tolerance=tolerance,
-                    description=f"Cat - OUT_OF_PLACE",
+                    description="Cat - OUT_OF_PLACE",
                 )
             )
 
@@ -102,7 +102,7 @@ def parse_test_cases():
                         output_spec=output_spec,
                         comparison_target="out",
                         tolerance=tolerance,
-                        description=f"Cat - INPLACE(out)",
+                        description="Cat - INPLACE(out)",
                     )
                 )
 

test/infinicore/ops/cdist.py

+import sys
+import os
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
+
+import torch
+import infinicore
+from framework.base import BaseOperatorTest, TensorSpec, TestCase
+from framework.runner import GenericTestRunner
+
+# Test cases format: (x1_shape, x2_shape, x1_strides_or_None, x2_strides_or_None, p_or_None)
+
+_TEST_CASES_DATA = [
+    ((5, 3), (6, 3), None, None, None),
+    ((1, 4), (2, 4), None, None, 1.0),
+    ((8, 16), (8, 16), (128, 16), (128, 16), 2.0),
+    ((3, 2), (4, 2), None, (0, 2), None),
+    ((10, 5), (7, 5), None, None, float("inf")),
+    ((2, 1), (3, 1), None, None, None),
+]
+
+_TOLERANCE_MAP = {
+    infinicore.float32: {"atol": 1e-5, "rtol": 1e-4},
+}
+
+_TENSOR_DTYPES = [infinicore.float32]
+
+
+def parse_test_cases():
+    test_cases = []
+    for x1_shape, x2_shape, x1_strides, x2_strides, p in _TEST_CASES_DATA:
+        for dtype in _TENSOR_DTYPES:
+            tol = _TOLERANCE_MAP.get(dtype, {"atol": 1e-5, "rtol": 1e-4})
+            x1_spec = TensorSpec.from_tensor(x1_shape, x1_strides, dtype)
+            x2_spec = TensorSpec.from_tensor(x2_shape, x2_strides, dtype)
+
+            kwargs = {}
+            if p is not None:
+                kwargs["p"] = p
+
+            test_cases.append(
+                TestCase(
+                    inputs=[x1_spec, x2_spec],
+                    kwargs=kwargs,
+                    output_spec=None,
+                    comparison_target=None,
+                    tolerance=tol,
+                    description="cdist - OUT_OF_PLACE",
+                )
+            )
+
+    return test_cases
+
+
+class OpTest(BaseOperatorTest):
+    """cdist operator test with simplified implementation"""
+
+    def __init__(self):
+        super().__init__("cdist")
+
+    def get_test_cases(self):
+        return parse_test_cases()
+
+    def torch_operator(self, *args, **kwargs):
+        return torch.cdist(*args, **kwargs)
+
+    # def infinicore_operator(self, *args, **kwargs):
+    #     """InfiniCore implementation (operator not yet available)."""
+    #     return infinicore.cdist(*args, **kwargs)
+
+
+def main():
+    """Main entry point"""
+    runner = GenericTestRunner(OpTest)
+    runner.run_and_exit()
+
+
+if __name__ == "__main__":
+    main()

test/infinicore/ops/celu.py

+import sys
+import os
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
+
+import torch
+import infinicore
+from framework.base import BaseOperatorTest, TensorSpec, TestCase
+from framework.runner import GenericTestRunner
+from framework.utils import is_broadcast
+
+# Test cases format: (in_shape, in_strides_or_None, alpha_or_None)
+
+# ==============================================================================
+# Operator-specific configuration
+# ==============================================================================
+
+_TEST_CASES_DATA = [
+    ((13, 4), None, None),
+    ((13, 4), (10, 1), None),
+    ((8, 8), None, 0.5),
+    ((16, 16), (256, 1), 1.5),
+    ((32, 8), None, 1.0),
+]
+
+# Tolerance configuration
+_TOLERANCE_MAP = {
+    infinicore.float16: {"atol": 1e-3, "rtol": 1e-2},
+    infinicore.float32: {"atol": 1e-5, "rtol": 1e-4},
+    infinicore.bfloat16: {"atol": 1e-2, "rtol": 5e-2},
+}
+
+# Data types to test
+_TENSOR_DTYPES = [infinicore.float16, infinicore.bfloat16, infinicore.float32]
+
+
+def parse_test_cases():
+
+    test_cases = []
+
+    for data in _TEST_CASES_DATA:
+        shape = data[0]
+        in_strides = data[1] if len(data) > 1 else None
+        alpha = data[2] if len(data) > 2 else None
+
+        input_supports_inplace = not is_broadcast(in_strides)
+
+        for dtype in _TENSOR_DTYPES:
+            tolerance = _TOLERANCE_MAP.get(dtype, {"atol": 1e-5, "rtol": 1e-4})
+
+            input_spec = TensorSpec.from_tensor(shape, in_strides, dtype)
+
+            # Out-of-place
+            kwargs = {}
+            if alpha is not None:
+                kwargs["alpha"] = alpha
+
+            test_cases.append(
+                TestCase(
+                    inputs=[input_spec],
+                    kwargs=kwargs,
+                    output_spec=None,
+                    comparison_target=None,
+                    tolerance=tolerance,
+                    description=f"CELU - OUT_OF_PLACE",
+                )
+            )
+
+            # In-place
+            if input_supports_inplace:
+                inplace_kwargs = {"inplace": True}
+                if alpha is not None:
+                    inplace_kwargs["alpha"] = alpha
+
+                test_cases.append(
+                    TestCase(
+                        inputs=[input_spec],
+                        kwargs=inplace_kwargs,
+                        output_spec=None,
+                        comparison_target=0,
+                        tolerance=tolerance,
+                        description=f"CELU - INPLACE",
+                    )
+                )
+
+    return test_cases
+
+
+class OpTest(BaseOperatorTest):
+    """CELU operator test with simplified implementation"""
+
+    def __init__(self):
+        super().__init__("CELU")
+
+    def get_test_cases(self):
+        return parse_test_cases()
+
+    def torch_operator(self, *args, **kwargs):
+        return torch.nn.functional.celu(*args, **kwargs)
+
+    # def infinicore_operator(self, *args, **kwargs):
+    #     """InfiniCore implementation (operator not yet available)."""
+    #     return infinicore.nn.functional.celu(*args, **kwargs)
+
+
+def main():
+    """Main entry point"""
+    runner = GenericTestRunner(OpTest)
+    runner.run_and_exit()
+
+
+if __name__ == "__main__":
+    main()

test/infinicore/ops/clone.py

+import sys
+import os
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
+
+import torch
+import infinicore
+from framework.base import BaseOperatorTest, TensorSpec, TestCase
+from framework.runner import GenericTestRunner
+
+# Test cases format: (shape, input_strides_or_None)
+_TEST_CASES_DATA = [
+    ((3, 4), None),
+    ((6, 2), (12, 1)),
+    ((5, 5), None),
+    ((1, 7), None),
+    ((8, 3), (24, 1)),
+    ((2, 2, 2), None),
+]
+
+_TOLERANCE_MAP = {infinicore.float32: {"atol": 1e-5, "rtol": 1e-4}}
+_TENSOR_DTYPES = [infinicore.float32]
+
+
+def parse_test_cases():
+    test_cases = []
+    for data in _TEST_CASES_DATA:
+        shape, strides = data
+
+        for dtype in _TENSOR_DTYPES:
+            tol = _TOLERANCE_MAP.get(dtype)
+            input_spec = TensorSpec.from_tensor(shape, strides, dtype)
+
+            # clone is out-of-place; also test clone with memory_format not None if supported
+            test_cases.append(
+                TestCase(
+                    inputs=[input_spec],
+                    kwargs={},
+                    output_spec=None,
+                    comparison_target=None,
+                    tolerance=tol,
+                    description=f"clone - OUT_OF_PLACE",
+                )
+            )
+
+    return test_cases
+
+
+class OpTest(BaseOperatorTest):
+    """Clone operator test with simplified implementation"""
+
+    def __init__(self):
+        super().__init__("Clone")
+
+    def get_test_cases(self):
+        return parse_test_cases()
+
+    def torch_operator(self, *args, **kwargs):
+        return torch.clone(*args, **kwargs)
+
+    # def infinicore_operator(self, *args, **kwargs):
+    #     """InfiniCore implementation (operator not yet available)."""
+    #     return infinicore.clone(*args, **kwargs)
+
+
+def main():
+    """Main entry point"""
+    runner = GenericTestRunner(OpTest)
+    runner.run_and_exit()
+
+
+if __name__ == "__main__":
+    main()

test/infinicore/ops/combinations.py

+import sys
+import os
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
+
+import torch
+import infinicore
+from framework.base import BaseOperatorTest, TensorSpec, TestCase
+from framework.runner import GenericTestRunner
+
+# Test cases format: (input_shape, input_strides_or_None, r)
+# combinations operates on 1-D inputs (combinations of elements). We keep inputs small.
+_TEST_CASES_DATA = [
+    ((5,), None, 2),
+    ((6,), None, 3),
+    ((4,), None, 1),
+    ((7,), None, 2),
+    ((8,), None, 3),
+    ((3,), None, 2),
+]
+
+_TOLERANCE_MAP = {infinicore.float32: {"atol": 1e-5, "rtol": 1e-4}}
+
+_TENSOR_DTYPES = [infinicore.float32]
+
+
+def parse_test_cases():
+    test_cases = []
+    for data in _TEST_CASES_DATA:
+        shape, strides, r = data
+
+        for dtype in _TENSOR_DTYPES:
+            tol = _TOLERANCE_MAP.get(dtype)
+            input_spec = TensorSpec.from_tensor(shape, strides, dtype)
+
+            kwargs = {"r": r}
+
+            test_cases.append(
+                TestCase(
+                    inputs=[input_spec],
+                    kwargs=kwargs,
+                    output_spec=None,
+                    comparison_target=None,
+                    tolerance=tol,
+                    description=f"combinations - OUT_OF_PLACE",
+                )
+            )
+
+    return test_cases
+
+
+class OpTest(BaseOperatorTest):
+    """Combinations operator test with simplified implementation"""
+
+    def __init__(self):
+        super().__init__("Combinations")
+
+    def get_test_cases(self):
+        return parse_test_cases()
+
+    def torch_operator(self, *args, **kwargs):
+        return torch.combinations(*args, **kwargs)
+
+    # def infinicore_operator(self, *args, **kwargs):
+    #     """InfiniCore implementation (operator not yet available)."""
+    #     return infinicore.combinations(*args, **kwargs)
+
+
+def main():
+    """Main entry point"""
+    runner = GenericTestRunner(OpTest)
+    runner.run_and_exit()
+
+
+if __name__ == "__main__":
+    main()

test/infinicore/ops/conv1d.py

+import sys
+import os
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
+
+import torch
+import infinicore
+from framework.base import BaseOperatorTest, TensorSpec, TestCase
+from framework.runner import GenericTestRunner
+from framework.utils import is_broadcast
+
+# Test cases: (in_shape, in_strides_or_None, weight_shape, bias_shape_or_None, stride, padding, dilation, groups)
+
+_TEST_CASES_DATA = [
+    ((2, 4, 16), None, (8, 4, 3), None, 1, 0, 1, 1),
+    ((1, 6, 15), (90, 15, 1), (4, 6, 5), (4,), 2, 2, 1, 1),
+    ((2, 16, 32), None, (8, 8, 1), None, 1, 0, 1, 2),
+    ((3, 3, 7), (21, 7, 1), (6, 3, 3), None, 1, 0, 1, 1),
+    ((2, 2, 31), None, (4, 2, 4), (4,), 2, 1, 1, 1),
+    ((1, 8, 9), (72, 9, 1), (8, 8, 3), None, 1, 1, 2, 1),
+]
+
+_TOLERANCE_MAP = {
+    infinicore.float16: {"atol": 1e-2, "rtol": 1e-2},
+    infinicore.float32: {"atol": 1e-4, "rtol": 1e-4},
+}
+
+_TENSOR_DTYPES = [infinicore.float16, infinicore.float32]
+
+
+def parse_test_cases():
+    tests = []
+    for (
+        in_shape,
+        in_strides,
+        w_shape,
+        b_shape,
+        stride,
+        padding,
+        dilation,
+        groups,
+    ) in _TEST_CASES_DATA:
+        for dtype in _TENSOR_DTYPES:
+            tol = _TOLERANCE_MAP.get(dtype, {"atol": 1e-5, "rtol": 1e-3})
+            in_spec = TensorSpec.from_tensor(in_shape, in_strides, dtype)
+            weight_spec = TensorSpec.from_tensor(w_shape, None, dtype)
+            if b_shape is not None:
+                bias_spec = TensorSpec.from_tensor(b_shape, None, dtype)
+            else:
+                bias_spec = None
+
+            kwargs = {
+                "stride": stride,
+                "padding": padding,
+                "dilation": dilation,
+                "groups": groups,
+            }
+            inputs = [in_spec, weight_spec]
+            if bias_spec is not None:
+                inputs.append(bias_spec)
+
+            tests.append(
+                TestCase(
+                    inputs=inputs,
+                    kwargs=kwargs,
+                    output_spec=None,
+                    comparison_target=None,
+                    tolerance=tol,
+                    description="Conv1d - OUT_OF_PLACE",
+                )
+            )
+
+    return tests
+
+
+class OpTest(BaseOperatorTest):
+    """Conv1d operator test with simplified implementation"""
+
+    def __init__(self):
+        super().__init__("Conv1d")
+
+    def get_test_cases(self):
+        return parse_test_cases()
+
+    def torch_operator(self, *args, **kwargs):
+        return torch.nn.functional.conv1d(*args, **kwargs)
+
+    # def infinicore_operator(self, *args, **kwargs):
+    #     """InfiniCore implementation (operator not yet available)."""
+    #     return infinicore.nn.functional.conv1d(*args, **kwargs)
+
+
+def main():
+    """Main entry point"""
+    runner = GenericTestRunner(OpTest)
+    runner.run_and_exit()
+
+
+if __name__ == "__main__":
+    main()

test/infinicore/ops/conv2d.py

+import sys
+import os
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
+
+import torch
+import infinicore
+from framework.base import BaseOperatorTest, TensorSpec, TestCase
+from framework.runner import GenericTestRunner
+from framework.utils import is_broadcast
+
+# Test cases: (in_shape, in_strides_or_None, weight_shape, bias_shape_or_None, stride, padding, dilation, groups)
+
+_TEST_CASES_DATA = [
+    ((2, 4, 16, 16), None, (8, 4, 3, 3), None, (1, 1), (0, 0), (1, 1), 1),
+    ((1, 6, 15, 17), (1530, 255, 17, 1), (4, 6, 5, 3), (4,), (2, 2), (2, 1), (1, 1), 1),
+    ((2, 8, 32, 32), None, (8, 8, 1, 1), None, (1, 1), (0, 0), (1, 2), 1),
+    ((3, 3, 7, 9), (189, 63, 9, 1), (6, 3, 3, 3), None, 1, (1, 1), (1, 1), 1),
+    ((2, 2, 31, 29), None, (4, 2, 4, 3), (4,), (2, 1), (1, 0), (1, 1), 1),
+    ((1, 8, 9, 11), (792, 99, 11, 1), (8, 8, 3, 3), None, (1, 1), (1, 1), (1, 1), 1),
+]
+
+_TOLERANCE_MAP = {
+    infinicore.float16: {"atol": 1e-2, "rtol": 1e-2},
+    infinicore.float32: {"atol": 1e-4, "rtol": 1e-4},
+}
+
+_TENSOR_DTYPES = [infinicore.float16, infinicore.float32]
+
+
+def parse_test_cases():
+    tests = []
+    for (
+        in_shape,
+        in_strides,
+        w_shape,
+        b_shape,
+        stride,
+        padding,
+        dilation,
+        groups,
+    ) in _TEST_CASES_DATA:
+        for dtype in _TENSOR_DTYPES:
+            tol = _TOLERANCE_MAP.get(dtype, {"atol": 1e-5, "rtol": 1e-3})
+            in_spec = TensorSpec.from_tensor(in_shape, in_strides, dtype)
+            weight_spec = TensorSpec.from_tensor(w_shape, None, dtype)
+            if b_shape is not None:
+                bias_spec = TensorSpec.from_tensor(b_shape, None, dtype)
+            else:
+                bias_spec = None
+
+            kwargs = {
+                "stride": stride,
+                "padding": padding,
+                "dilation": dilation,
+                "groups": groups,
+            }
+            inputs = [in_spec, weight_spec]
+            if bias_spec is not None:
+                inputs.append(bias_spec)
+
+            tests.append(
+                TestCase(
+                    inputs=inputs,
+                    kwargs=kwargs,
+                    output_spec=None,
+                    comparison_target=None,
+                    tolerance=tol,
+                    description="Conv2d - OUT_OF_PLACE",
+                )
+            )
+
+    return tests
+
+
+class OpTest(BaseOperatorTest):
+    """Conv2d operator test with simplified implementation"""
+
+    def __init__(self):
+        super().__init__("Conv2d")
+
+    def get_test_cases(self):
+        return parse_test_cases()
+
+    def torch_operator(self, *args, **kwargs):
+        return torch.nn.functional.conv2d(*args, **kwargs)
+
+    # def infinicore_operator(self, *args, **kwargs):
+    #     """InfiniCore implementation (operator not yet available)."""
+    #     return infinicore.nn.functional.conv2d(*args, **kwargs)
+
+
+def main():
+    """Main entry point"""
+    runner = GenericTestRunner(OpTest)
+    runner.run_and_exit()
+
+
+if __name__ == "__main__":
+    main()

test/infinicore/ops/conv3d.py

+import sys
+import os
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
+
+import torch
+import infinicore
+from framework.base import BaseOperatorTest, TensorSpec, TestCase
+from framework.runner import GenericTestRunner
+from framework.utils import is_broadcast
+
+# Test cases: (in_shape, in_strides_or_None, weight_shape, bias_shape_or_None, stride, padding, dilation, groups)
+
+_TEST_CASES_DATA = [
+    ((1, 2, 8, 8, 8), None, (4, 2, 3, 3, 3), None, (1, 1, 1), (0, 0, 0), (1, 1, 1), 1),
+    (
+        (2, 3, 7, 9, 5),
+        (756, 252, 36, 4, 1),
+        (6, 3, 3, 3, 1),
+        (6,),
+        (2, 2, 1),
+        (1, 1, 0),
+        (1, 1, 1),
+        1,
+    ),
+    (
+        (1, 4, 16, 16, 6),
+        None,
+        (8, 4, 1, 1, 2),
+        None,
+        (1, 1, 2),
+        (0, 1, 0),
+        (1, 1, 1),
+        1,
+    ),
+    (
+        (2, 1, 9, 11, 7),
+        (693, 77, 77, 7, 1),
+        (6, 1, 3, 3, 3),
+        None,
+        1,
+        (1, 0, 1),
+        (1, 1, 1),
+        1,
+    ),
+    ((3, 2, 5, 6, 4), None, (4, 2, 2, 2, 2), (4,), (1, 1, 1), (0, 1, 0), (1, 1, 1), 1),
+    (
+        (2, 6, 10, 9, 8),
+        (4320, 720, 72, 8, 1),
+        (8, 6, 3, 3, 2),
+        None,
+        (2, 1, 2),
+        (1, 0, 1),
+        (1, 1, 1),
+        1,
+    ),
+]
+
+_TOLERANCE_MAP = {
+    infinicore.float16: {"atol": 1e-2, "rtol": 1e-2},
+    infinicore.float32: {"atol": 1e-4, "rtol": 1e-4},
+}
+
+_TENSOR_DTYPES = [infinicore.float16, infinicore.float32]
+
+
+def parse_test_cases():
+    tests = []
+    for (
+        in_shape,
+        in_strides,
+        w_shape,
+        b_shape,
+        stride,
+        padding,
+        dilation,
+        groups,
+    ) in _TEST_CASES_DATA:
+        for dtype in _TENSOR_DTYPES:
+            tol = _TOLERANCE_MAP.get(dtype, {"atol": 1e-5, "rtol": 1e-3})
+            in_spec = TensorSpec.from_tensor(in_shape, in_strides, dtype)
+            weight_spec = TensorSpec.from_tensor(w_shape, None, dtype)
+            if b_shape is not None:
+                bias_spec = TensorSpec.from_tensor(b_shape, None, dtype)
+            else:
+                bias_spec = None
+
+            kwargs = {
+                "stride": stride,
+                "padding": padding,
+                "dilation": dilation,
+                "groups": groups,
+            }
+            inputs = [in_spec, weight_spec]
+            if bias_spec is not None:
+                inputs.append(bias_spec)
+
+            tests.append(
+                TestCase(
+                    inputs=inputs,
+                    kwargs=kwargs,
+                    output_spec=None,
+                    comparison_target=None,
+                    tolerance=tol,
+                    description="Conv3d - OUT_OF_PLACE",
+                )
+            )
+
+    return tests
+
+
+class OpTest(BaseOperatorTest):
+    """Conv3d operator test with simplified implementation"""
+
+    def __init__(self):
+        super().__init__("Conv3d")
+
+    def get_test_cases(self):
+        return parse_test_cases()
+
+    def torch_operator(self, *args, **kwargs):
+        return torch.nn.functional.conv3d(*args, **kwargs)
+
+    # def infinicore_operator(self, *args, **kwargs):
+    #     """InfiniCore implementation (operator not yet available)."""
+    #     return infinicore.nn.functional.conv3d(*args, **kwargs)
+
+
+def main():
+    """Main entry point"""
+    runner = GenericTestRunner(OpTest)
+    runner.run_and_exit()
+
+
+if __name__ == "__main__":
+    main()

test/infinicore/ops/conv_transpose1d.py

+import sys
+import os
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
+
+import torch
+import infinicore
+from framework.base import BaseOperatorTest, TensorSpec, TestCase
+from framework.runner import GenericTestRunner
+from framework.utils import is_broadcast
+
+# Test cases: (in_shape, in_strides_or_None, weight_shape, bias_shape_or_None, stride, padding, output_padding, groups)
+
+_TEST_CASES_DATA = [
+    ((2, 4, 16), None, (4, 4, 3), None, 1, 0, 0, 1),
+    ((1, 6, 15), (90, 15, 1), (6, 6, 5), (6,), 2, 1, 1, 1),
+    ((2, 8, 32), None, (8, 4, 1), None, 1, 0, 0, 2),
+    ((3, 3, 7), (21, 7, 1), (3, 3, 3), None, 1, 0, 0, 1),
+    ((2, 2, 31), None, (2, 2, 4), (2,), 2, 1, 1, 1),
+    ((1, 8, 9), (72, 9, 1), (8, 4, 3), None, 1, 1, 0, 1),
+]
+
+_TOLERANCE_MAP = {
+    infinicore.float16: {"atol": 1e-2, "rtol": 1e-2},
+    infinicore.float32: {"atol": 1e-4, "rtol": 1e-4},
+}
+
+_TENSOR_DTYPES = [infinicore.float16, infinicore.float32]
+
+
+def parse_test_cases():
+    tests = []
+    for (
+        in_shape,
+        in_strides,
+        w_shape,
+        b_shape,
+        stride,
+        padding,
+        out_pad,
+        groups,
+    ) in _TEST_CASES_DATA:
+        for dtype in _TENSOR_DTYPES:
+            tol = _TOLERANCE_MAP.get(dtype, {"atol": 1e-5, "rtol": 1e-3})
+            in_spec = TensorSpec.from_tensor(in_shape, in_strides, dtype)
+            weight_spec = TensorSpec.from_tensor(w_shape, None, dtype)
+            if b_shape is not None:
+                bias_spec = TensorSpec.from_tensor(b_shape, None, dtype)
+            else:
+                bias_spec = None
+
+            kwargs = {
+                "stride": stride,
+                "padding": padding,
+                "output_padding": out_pad,
+                "groups": groups,
+            }
+            inputs = [in_spec, weight_spec]
+            if bias_spec is not None:
+                inputs.append(bias_spec)
+
+            tests.append(
+                TestCase(
+                    inputs=inputs,
+                    kwargs=kwargs,
+                    output_spec=None,
+                    comparison_target=None,
+                    tolerance=tol,
+                    description="ConvTranspose1d - OUT_OF_PLACE",
+                )
+            )
+
+    return tests
+
+
+class OpTest(BaseOperatorTest):
+    """ConvTranspose1d operator test with simplified implementation"""
+
+    def __init__(self):
+        super().__init__("ConvTranspose1d")
+
+    def get_test_cases(self):
+        return parse_test_cases()
+
+    def torch_operator(self, *args, **kwargs):
+        return torch.nn.functional.conv_transpose1d(*args, **kwargs)
+
+    # def infinicore_operator(self, *args, **kwargs):
+    #     """InfiniCore implementation (operator not yet available)."""
+    #     return infinicore.nn.functional.conv_transpose1d(*args, **kwargs)
+
+
+def main():
+    """Main entry point"""
+    runner = GenericTestRunner(OpTest)
+    runner.run_and_exit()
+
+
+if __name__ == "__main__":
+    main()

test/infinicore/ops/conv_transpose2d.py

+import sys
+import os
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
+
+import torch
+import infinicore
+from framework.base import BaseOperatorTest, TensorSpec, TestCase
+from framework.runner import GenericTestRunner
+from framework.utils import is_broadcast
+
+# Test cases: (in_shape, in_strides_or_None, weight_shape, bias_shape_or_None, stride, padding, output_padding, groups)
+
+_TEST_CASES_DATA = [
+    ((2, 4, 16, 16), None, (4, 4, 3, 3), None, (1, 1), (0, 0), (0, 0), 1),
+    ((1, 6, 15, 17), (1530, 255, 17, 1), (6, 6, 5, 3), (6,), (2, 2), (2, 1), (1, 1), 1),
+    ((2, 8, 32, 32), None, (8, 4, 1, 1), None, (1, 1), (0, 0), (0, 0), 1),
+    ((3, 3, 7, 9), (189, 63, 9, 1), (3, 3, 3, 3), None, 1, (1, 1), (0, 0), 1),
+    ((2, 2, 31, 29), None, (2, 2, 4, 3), (2,), (2, 1), (1, 0), (1, 0), 1),
+    ((1, 8, 9, 11), (792, 99, 11, 1), (8, 4, 3, 3), None, (1, 1), (1, 1), (0, 0), 1),
+]
+
+_TOLERANCE_MAP = {
+    infinicore.float16: {"atol": 1e-2, "rtol": 1e-2},
+    infinicore.float32: {"atol": 1e-4, "rtol": 1e-4},
+}
+
+_TENSOR_DTYPES = [infinicore.float16, infinicore.float32]
+
+
+def parse_test_cases():
+    tests = []
+    for (
+        in_shape,
+        in_strides,
+        w_shape,
+        b_shape,
+        stride,
+        padding,
+        out_pad,
+        groups,
+    ) in _TEST_CASES_DATA:
+        for dtype in _TENSOR_DTYPES:
+            tol = _TOLERANCE_MAP.get(dtype, {"atol": 1e-5, "rtol": 1e-3})
+            in_spec = TensorSpec.from_tensor(in_shape, in_strides, dtype)
+            weight_spec = TensorSpec.from_tensor(w_shape, None, dtype)
+            if b_shape is not None:
+                bias_spec = TensorSpec.from_tensor(b_shape, None, dtype)
+            else:
+                bias_spec = None
+
+            kwargs = {
+                "stride": stride,
+                "padding": padding,
+                "output_padding": out_pad,
+                "groups": groups,
+            }
+            inputs = [in_spec, weight_spec]
+            if bias_spec is not None:
+                inputs.append(bias_spec)
+
+            tests.append(
+                TestCase(
+                    inputs=inputs,
+                    kwargs=kwargs,
+                    output_spec=None,
+                    comparison_target=None,
+                    tolerance=tol,
+                    description="ConvTranspose2d - OUT_OF_PLACE",
+                )
+            )
+
+    return tests
+
+
+class OpTest(BaseOperatorTest):
+    """ConvTranspose2d operator test with simplified implementation"""
+
+    def __init__(self):
+        super().__init__("ConvTranspose2d")
+
+    def get_test_cases(self):
+        return parse_test_cases()
+
+    def torch_operator(self, *args, **kwargs):
+        return torch.nn.functional.conv_transpose2d(*args, **kwargs)
+
+    # def infinicore_operator(self, *args, **kwargs):
+    #     """InfiniCore implementation (operator not yet available)."""
+    #     return infinicore.nn.functional.conv_transpose2d(*args, **kwargs)
+
+
+def main():
+    """Main entry point"""
+    runner = GenericTestRunner(OpTest)
+    runner.run_and_exit()
+
+
+if __name__ == "__main__":
+    main()

test/infinicore/ops/conv_transpose3d.py

+import sys
+import os
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
+
+import torch
+import infinicore
+from framework.base import BaseOperatorTest, TensorSpec, TestCase
+from framework.runner import GenericTestRunner
+from framework.utils import is_broadcast
+
+# Test cases: (in_shape, in_strides_or_None, weight_shape, bias_shape_or_None, stride, padding, output_padding, groups)
+
+_TEST_CASES_DATA = [
+    ((1, 2, 8, 8, 8), None, (2, 2, 3, 3, 3), None, (1, 1, 1), (0, 0, 0), (0, 0, 0), 1),
+    (
+        (2, 3, 7, 9, 5),
+        (756, 252, 36, 4, 1),
+        (3, 3, 3, 3, 1),
+        (3,),
+        (2, 2, 1),
+        (1, 1, 0),
+        (0, 0, 0),
+        1,
+    ),
+    (
+        (1, 4, 16, 16, 6),
+        None,
+        (4, 2, 1, 1, 2),
+        None,
+        (1, 1, 2),
+        (0, 1, 0),
+        (0, 0, 0),
+        1,
+    ),
+    (
+        (2, 1, 9, 11, 7),
+        (693, 77, 77, 7, 1),
+        (1, 6, 3, 3, 3),
+        None,
+        1,
+        (1, 0, 1),
+        (0, 0, 0),
+        1,
+    ),
+    ((3, 2, 5, 6, 4), None, (2, 2, 2, 2, 2), (2,), (1, 1, 1), (0, 1, 0), (0, 0, 0), 1),
+    (
+        (2, 6, 10, 9, 8),
+        (4320, 720, 72, 8, 1),
+        (6, 8, 3, 3, 2),
+        None,
+        (2, 1, 2),
+        (1, 0, 1),
+        (1, 0, 1),
+        1,
+    ),
+]
+
+_TOLERANCE_MAP = {
+    infinicore.float16: {"atol": 1e-2, "rtol": 1e-2},
+    infinicore.float32: {"atol": 1e-4, "rtol": 1e-4},
+}
+
+_TENSOR_DTYPES = [infinicore.float16, infinicore.float32]
+
+
+def parse_test_cases():
+    tests = []
+    for (
+        in_shape,
+        in_strides,
+        w_shape,
+        b_shape,
+        stride,
+        padding,
+        out_pad,
+        groups,
+    ) in _TEST_CASES_DATA:
+        for dtype in _TENSOR_DTYPES:
+            tol = _TOLERANCE_MAP.get(dtype, {"atol": 1e-5, "rtol": 1e-3})
+            in_spec = TensorSpec.from_tensor(in_shape, in_strides, dtype)
+            weight_spec = TensorSpec.from_tensor(w_shape, None, dtype)
+            if b_shape is not None:
+                bias_spec = TensorSpec.from_tensor(b_shape, None, dtype)
+            else:
+                bias_spec = None
+
+            kwargs = {
+                "stride": stride,
+                "padding": padding,
+                "output_padding": out_pad,
+                "groups": groups,
+            }
+            inputs = [in_spec, weight_spec]
+            if bias_spec is not None:
+                inputs.append(bias_spec)
+
+            tests.append(
+                TestCase(
+                    inputs=inputs,
+                    kwargs=kwargs,
+                    output_spec=None,
+                    comparison_target=None,
+                    tolerance=tol,
+                    description="ConvTranspose3d - OUT_OF_PLACE",
+                )
+            )
+
+    return tests
+
+
+class OpTest(BaseOperatorTest):
+    """ConvTranspose3d operator test with simplified implementation"""
+
+    def __init__(self):
+        super().__init__("ConvTranspose3d")
+
+    def get_test_cases(self):
+        return parse_test_cases()
+
+    def torch_operator(self, *args, **kwargs):
+        return torch.nn.functional.conv_transpose3d(*args, **kwargs)
+
+    # def infinicore_operator(self, *args, **kwargs):
+    #     """InfiniCore implementation (operator not yet available)."""
+    #     return infinicore.nn.functional.conv_transpose3d(*args, **kwargs)
+
+
+def main():
+    """Main entry point"""
+    runner = GenericTestRunner(OpTest)
+    runner.run_and_exit()
+
+
+if __name__ == "__main__":
+    main()

test/infinicore/ops/corrcoef.py

+import sys
+import os
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
+
+import torch
+import infinicore
+from framework.base import BaseOperatorTest, TensorSpec, TestCase
+from framework.runner import GenericTestRunner
+
+# Test cases format: (input_shape, input_strides_or_None)
+# corrcoef accepts 1-D or 2-D inputs; we include both shapes.
+_TEST_CASES_DATA = [
+    ((5,), None),
+    ((3, 5), None),
+    ((4, 4), (16, 1)),
+    ((2, 8), None),
+    ((6, 6), None),
+    ((1, 7), None),
+]
+
+_TOLERANCE_MAP = {infinicore.float32: {"atol": 1e-5, "rtol": 1e-4}}
+
+_TENSOR_DTYPES = [infinicore.float32]
+
+
+def parse_test_cases():
+    test_cases = []
+    for data in _TEST_CASES_DATA:
+        shape, strides = data
+
+        for dtype in _TENSOR_DTYPES:
+            tol = _TOLERANCE_MAP.get(dtype)
+            input_spec = TensorSpec.from_tensor(shape, strides, dtype)
+            kwargs = {}
+
+            test_cases.append(
+                TestCase(
+                    inputs=[input_spec],
+                    kwargs=kwargs,
+                    output_spec=None,
+                    comparison_target=None,
+                    tolerance=tol,
+                    description=f"corrcoef - OUT_OF_PLACE",
+                )
+            )
+
+    return test_cases
+
+
+class OpTest(BaseOperatorTest):
+    """Corrcoef operator test with simplified implementation"""
+
+    def __init__(self):
+        super().__init__("Corrcoef")
+
+    def get_test_cases(self):
+        return parse_test_cases()
+
+    def torch_operator(self, *args, **kwargs):
+        return torch.corrcoef(*args, **kwargs)
+
+    # def infinicore_operator(self, *args, **kwargs):
+    #     """InfiniCore implementation (operator not yet available)."""
+    #     return infinicore.corrcoef(*args, **kwargs)
+
+
+def main():
+    """Main entry point"""
+    runner = GenericTestRunner(OpTest)
+    runner.run_and_exit()
+
+
+if __name__ == "__main__":
+    main()

test/infinicore/ops/cosh.py

+import sys
+import os
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
+
+import torch
+import infinicore
+from framework.base import BaseOperatorTest, TensorSpec, TestCase
+from framework.runner import GenericTestRunner
+from framework.utils import is_broadcast
+
+# =======================================================================
+# Test cases format: (shape, input_strides_or_None)
+# =======================================================================
+
+_TEST_CASES_DATA = [
+    ((13, 4), None),
+    ((13, 4), (10, 1)),
+    ((8, 16), None),
+    ((8, 16), (40, 1)),
+    ((2, 3, 4), None),
+    ((16, 5632), None),
+]
+
+_TOLERANCE_MAP = {
+    infinicore.float16: {"atol": 1e-3, "rtol": 1e-2},
+    infinicore.float32: {"atol": 1e-5, "rtol": 1e-4},
+    infinicore.bfloat16: {"atol": 1e-2, "rtol": 5e-2},
+}
+
+_TENSOR_DTYPES = [infinicore.float16, infinicore.bfloat16, infinicore.float32]
+
+
+def parse_test_cases():
+    test_cases = []
+    for data in _TEST_CASES_DATA:
+        shape = data[0]
+        in_strides = data[1] if len(data) > 1 else None
+
+        supports_inplace = not is_broadcast(in_strides)
+
+        for dtype in _TENSOR_DTYPES:
+            tol = _TOLERANCE_MAP.get(dtype, {"atol": 1e-5, "rtol": 1e-4})
+            input_spec = TensorSpec.from_tensor(shape, in_strides, dtype)
+            out_spec = TensorSpec.from_tensor(shape, None, dtype)
+
+            test_cases.append(
+                TestCase(
+                    inputs=[input_spec],
+                    kwargs={},
+                    output_spec=None,
+                    comparison_target=None,
+                    tolerance=tol,
+                    description="cosh - OUT_OF_PLACE",
+                )
+            )
+
+            test_cases.append(
+                TestCase(
+                    inputs=[input_spec],
+                    kwargs=None,
+                    output_spec=out_spec,
+                    comparison_target="out",
+                    tolerance=tol,
+                    description="cosh - INPLACE(out)",
+                )
+            )
+
+            if supports_inplace:
+                test_cases.append(
+                    TestCase(
+                        inputs=[input_spec],
+                        kwargs={"out": 0},
+                        output_spec=None,
+                        comparison_target=0,
+                        tolerance=tol,
+                        description="cosh - INPLACE(input)",
+                    )
+                )
+
+    return test_cases
+
+
+class OpTest(BaseOperatorTest):
+    """Cosh operator test with simplified implementation"""
+
+    def __init__(self):
+        super().__init__("Cosh")
+
+    def get_test_cases(self):
+        return parse_test_cases()
+
+    def torch_operator(self, *args, **kwargs):
+        return torch.cosh(*args, **kwargs)
+
+    # def infinicore_operator(self, *args, **kwargs):
+    #     """InfiniCore implementation (operator not yet available)."""
+    #     return infinicore.cosh(*args, **kwargs)
+
+
+def main():
+    """Main entry point"""
+    runner = GenericTestRunner(OpTest)
+    runner.run_and_exit()
+
+
+if __name__ == "__main__":
+    main()

test/infinicore/ops/cosine_embedding_loss.py

+import sys
+import os
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
+
+import torch
+import infinicore
+from framework.base import BaseOperatorTest, TensorSpec, TestCase
+from framework.runner import GenericTestRunner
+
+# Test cases format: (input1_shape, input2_shape, target_shape, input1_strides_or_None, input2_strides_or_None, target_strides_or_None, margin_or_None)
+# infinicore.nn.functional.cosine_embedding_loss(x1, x2, y, margin=0.0, reduction='mean')
+
+_TEST_CASES_DATA = [
+    ((4, 3), (4, 3), (4,), None, None, None, None),
+    ((8, 5), (8, 5), (8,), (40, 5), None, None, 0.5),
+    ((1, 10), (1, 10), (1,), None, None, None, 0.2),
+    ((16, 20), (16, 20), (16,), None, None, None, 0.0),
+    ((3, 7), (3, 7), (3,), None, (21, 7), None, None),
+    ((2, 2), (2, 2), (2,), None, None, None, 1.0),
+]
+
+_TOLERANCE_MAP = {
+    infinicore.float16: {"atol": 1e-3, "rtol": 1e-2},
+    infinicore.float32: {"atol": 1e-5, "rtol": 1e-4},
+    infinicore.bfloat16: {"atol": 1e-2, "rtol": 5e-2},
+}
+
+_TENSOR_DTYPES = [infinicore.float16, infinicore.bfloat16, infinicore.float32]
+
+
+def parse_test_cases():
+    test_cases = []
+    for s1, s2, st, st1, st2, stt, margin in _TEST_CASES_DATA:
+        for dtype in _TENSOR_DTYPES:
+            tol = _TOLERANCE_MAP.get(dtype, {"atol": 1e-5, "rtol": 1e-4})
+            a = TensorSpec.from_tensor(s1, st1, dtype)
+            b = TensorSpec.from_tensor(s2, st2, dtype)
+            y = TensorSpec.from_tensor(st, stt, dtype)
+
+            kwargs = {}
+            if margin is not None:
+                kwargs["margin"] = margin
+
+            test_cases.append(
+                TestCase(
+                    inputs=[a, b, y],
+                    kwargs=kwargs,
+                    output_spec=None,
+                    comparison_target=None,
+                    tolerance=tol,
+                    description="cosine_embedding_loss - OUT_OF_PLACE",
+                )
+            )
+
+    return test_cases
+
+
+class OpTest(BaseOperatorTest):
+    """cosine_embedding_loss operator test with simplified implementation"""
+
+    def __init__(self):
+        super().__init__("cosine_embedding_loss")
+
+    def get_test_cases(self):
+        return parse_test_cases()
+
+    def torch_operator(self, *args, **kwargs):
+        return torch.nn.functional.cosine_embedding_loss(*args, **kwargs)
+
+    # def infinicore_operator(self, *args, **kwargs):
+    #     """InfiniCore implementation (operator not yet available)."""
+    #     return infinicore.nn.functional.cosine_embedding_loss(*args, **kwargs)
+
+
+def main():
+    """Main entry point"""
+    runner = GenericTestRunner(OpTest)
+    runner.run_and_exit()
+
+
+if __name__ == "__main__":
+    main()

test/infinicore/ops/cosine_similarity.py

+import sys
+import os
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
+
+import torch
+import infinicore
+from framework.base import BaseOperatorTest, TensorSpec, TestCase
+from framework.runner import GenericTestRunner
+from framework.utils import is_broadcast
+
+# Test cases format: (shape, dim, eps, a_strides_or_None, b_strides_or_None)
+# infinicore.nn.functional.cosine_similarity(x1, x2, dim=1, eps=1e-8)
+
+_TEST_CASES_DATA = [
+    ((8, 16), 1, 1e-8, None, None),
+    ((8, 16), 1, 1e-6, (128, 1), (128, 1)),
+    ((2, 3, 4), 2, 1e-8, None, None),
+    ((16, 64), 1, 1e-8, None, None),
+    ((4, 5, 6), 0, 1e-8, None, None),
+    ((3, 4, 5), 1, 1e-7, (60, 20, 4), None),
+]
+
+_TOLERANCE_MAP = {
+    infinicore.float16: {"atol": 1e-2, "rtol": 1e-2},
+    infinicore.float32: {"atol": 1e-5, "rtol": 1e-4},
+    infinicore.bfloat16: {"atol": 1e-2, "rtol": 5e-2},
+}
+
+_TENSOR_DTYPES = [infinicore.float16, infinicore.bfloat16, infinicore.float32]
+
+
+def parse_test_cases():
+    test_cases = []
+    for data in _TEST_CASES_DATA:
+        shape, dim, eps = data[0], data[1], data[2]
+        a_strides = data[3] if len(data) > 3 else None
+        b_strides = data[4] if len(data) > 4 else None
+
+        a_supports_inplace = not is_broadcast(a_strides)
+        b_supports_inplace = not is_broadcast(b_strides)
+
+        for dtype in _TENSOR_DTYPES:
+            tol = _TOLERANCE_MAP.get(dtype, {"atol": 1e-5, "rtol": 1e-4})
+            a_spec = TensorSpec.from_tensor(shape, a_strides, dtype)
+            b_spec = TensorSpec.from_tensor(shape, b_strides, dtype)
+
+            kwargs = {"dim": dim, "eps": eps}
+
+            # Out-of-place
+            test_cases.append(
+                TestCase(
+                    inputs=[a_spec, b_spec],
+                    kwargs=kwargs,
+                    output_spec=None,
+                    comparison_target=None,
+                    tolerance=tol,
+                    description=f"cosine_similarity - OUT_OF_PLACE",
+                )
+            )
+
+            # PyTorch cosine_similarity does not support explicit out param; skip out tests
+
+    return test_cases
+
+
+class OpTest(BaseOperatorTest):
+    """CosineSimilarity operator test with simplified implementation"""
+
+    def __init__(self):
+        super().__init__("CosineSimilarity")
+
+    def get_test_cases(self):
+        return parse_test_cases()
+
+    def torch_operator(self, *args, **kwargs):
+        return torch.nn.functional.cosine_similarity(*args, **kwargs)
+
+    # def infinicore_operator(self, *args, **kwargs):
+    #     """InfiniCore implementation (operator not yet available)."""
+    #     return infinicore.nn.functional.cosine_similarity(*args, **kwargs)
+
+
+def main():
+    """Main entry point"""
+    runner = GenericTestRunner(OpTest)
+    runner.run_and_exit()
+
+
+if __name__ == "__main__":
+    main()