add torch cosine similarity (#6045)

ivy/functional/frontends/torch/nn/functional/distance_functions.py

@@ -0,0 +1,26 @@
+import ivy
+import ivy.functional.frontends.torch as torch_frontend
+from ivy.functional.frontends.torch.func_wrapper import to_ivy_arrays_and_back
+
+
+@to_ivy_arrays_and_back
+def cosine_similarity(x1, x2, *, dim=1, eps=1e-08):
+    x1, x2 = torch_frontend.promote_types_of_torch_inputs(x1, x2)
+
+    if len(x1.shape) == len(x2.shape) and len(x2.shape) >= 2:
+        numerator = ivy.sum(x1 * x2, axis=dim)
+        x1_squared_norm = ivy.sum(ivy.square(x1), axis=dim)
+        x2_squared_norm = ivy.sum(ivy.square(x2), axis=dim)
+    else:
+        numerator = ivy.sum(x1 * x2)
+        x1_squared_norm = ivy.sum(ivy.square(x1))
+        x2_squared_norm = ivy.sum(ivy.square(x2))
+
+    x1_norm = ivy.sqrt(x1_squared_norm)
+    x2_norm = ivy.sqrt(x2_squared_norm)
+    norm_mm = x1_norm * x2_norm
+    norm_mm, eps = torch_frontend.promote_types_of_torch_inputs(norm_mm, eps)
+    denominator = ivy.maximum(norm_mm, eps)
+
+    cosine = numerator / denominator
+    return cosine

ivy_tests/test_ivy/test_frontends/test_torch/test_distance_functions.py

@@ -0,0 +1,50 @@
+from hypothesis import assume, given, strategies as st
+
+# local
+import ivy_tests.test_ivy.helpers as helpers
+from ivy_tests.test_ivy.helpers import handle_cmd_line_args
+
+
+def _filter_dtypes(input_dtype):
+    assume(("bfloat16" not in input_dtype) and ("float16" not in input_dtype))
+
+
+# Cosine Similarity
+@handle_cmd_line_args
+@given(
+    d_type_and_x=helpers.dtype_and_values(
+        available_dtypes=helpers.get_dtypes("float", full=True),
+        min_value=2,
+        max_value=5,
+        min_dim_size=2,
+        shared_dtype=True,
+        num_arrays=2,
+    ),
+    dim=st.integers(min_value=-1, max_value=0),
+    num_positional_args=helpers.num_positional_args(
+        fn_name="ivy.functional.frontends.torch.nn.functional.cosine_similarity"
+    ),
+)
+def test_torch_cosine_similarity(
+    d_type_and_x,
+    dim,
+    with_out,
+    as_variable,
+    num_positional_args,
+    native_array,
+):
+    dtype, x = d_type_and_x
+    _filter_dtypes(dtype)
+    helpers.test_frontend_function(
+        input_dtypes=dtype,
+        as_variable_flags=as_variable,
+        with_out=with_out,
+        num_positional_args=num_positional_args,
+        native_array_flags=native_array,
+        frontend="torch",
+        fn_tree="nn.functional.cosine_similarity",
+        rtol=1e-01,
+        x1=x[0],
+        x2=x[1],
+        dim=dim,
+    )