[OpenVINO backend] support repeat

Author: Mohamed-Ashraf273

keras/src/backend/openvino/excluded_concrete_tests.txt

@@ -45,7 +45,6 @@ NumpyDtypeTest::test_multiply
 NumpyDtypeTest::test_power
 NumpyDtypeTest::test_prod
 NumpyDtypeTest::test_quantile
-NumpyDtypeTest::test_repeat
 NumpyDtypeTest::test_roll
 NumpyDtypeTest::test_round
 NumpyDtypeTest::test_searchsorted
@@ -106,7 +105,6 @@ NumpyOneInputOpsCorrectnessTest::test_pad_uint8_constant_2
 NumpyOneInputOpsCorrectnessTest::test_pad_int32_constant_2
 NumpyOneInputOpsCorrectnessTest::test_prod
 NumpyOneInputOpsCorrectnessTest::test_real
-NumpyOneInputOpsCorrectnessTest::test_repeat
 NumpyOneInputOpsCorrectnessTest::test_reshape
 NumpyOneInputOpsCorrectnessTest::test_roll
 NumpyOneInputOpsCorrectnessTest::test_round

keras/src/backend/openvino/numpy.py

@@ -1291,7 +1291,96 @@ def reciprocal(x):
 
 
 def repeat(x, repeats, axis=None):
-    raise NotImplementedError("`repeat` is not supported with openvino backend")
+    x = get_ov_output(x)
+
+    if axis is not None and axis < 0:
+        axis += len(x.get_partial_shape())
+
+    if axis is None:
+        x = ov_opset.reshape(
+            x, ov_opset.constant([-1], Type.i32), special_zero=False
+        ).output(0)
+        axis = 0
+
+    if isinstance(repeats, (int, np.integer)) or (
+        isinstance(repeats, np.ndarray)
+        and repeats.ndim == 1
+        and repeats.size == 1
+    ):
+        repeats_val = (
+            int(repeats) if isinstance(repeats, np.ndarray) else repeats
+        )
+        input_shape = ov_opset.shape_of(x, Type.i32).output(0)
+        dim_len = ov_opset.gather(
+            input_shape,
+            ov_opset.constant([axis], Type.i32),
+            ov_opset.constant(0, Type.i32),
+        ).output(0)
+        dim_len = ov_opset.squeeze(
+            dim_len, ov_opset.constant([0], Type.i32)
+        ).output(0)
+        idx_range = ov_opset.range(
+            ov_opset.constant(0, Type.i32),
+            dim_len,
+            ov_opset.constant(1, Type.i32),
+            output_type=Type.i32,
+        ).output(0)
+        idx_range = ov_opset.unsqueeze(
+            idx_range, ov_opset.constant([1], Type.i32)
+        ).output(0)
+        tiled = ov_opset.tile(
+            idx_range, ov_opset.constant([1, repeats_val], Type.i32)
+        ).output(0)
+        idx = ov_opset.reshape(
+            tiled, ov_opset.constant([-1], Type.i32), special_zero=False
+        ).output(0)
+        result = ov_opset.gather(
+            x, idx, ov_opset.constant(axis, Type.i32)
+        ).output(0)
+        return OpenVINOKerasTensor(result)
+
+    repeats_tensor = get_ov_output(repeats)
+    input_shape = ov_opset.shape_of(x, Type.i32)
+    axis_len = ov_opset.gather(
+        input_shape,
+        ov_opset.constant([axis], Type.i32),
+        ov_opset.constant(0, Type.i32),
+    )
+    axis_len = ov_opset.squeeze(axis_len, ov_opset.constant([0], Type.i32))
+
+    # cumsum and total output length
+    cumsum = ov_opset.cumsum(repeats_tensor, ov_opset.constant(0, Type.i32))
+    total = ov_opset.reduce_sum(
+        repeats_tensor, ov_opset.constant([0], Type.i32), keep_dims=False
+    )
+    total = ov_opset.convert(total, Type.i32)
+
+    # Build output indices [0, 1, ..., total-1]
+    out_indices = ov_opset.range(
+        ov_opset.constant(0, Type.i32),
+        total,
+        ov_opset.constant(1, Type.i32),
+        output_type=Type.i32,
+    )
+
+    # For each out_index, find which interval it falls in (searchsorted)
+    # Equivalent to: sum(out_indices >= cumsum) for each out_index
+    cumsum_unsq = ov_opset.unsqueeze(cumsum, ov_opset.constant([0], Type.i32))
+    out_indices_unsq = ov_opset.unsqueeze(
+        out_indices, ov_opset.constant([1], Type.i32)
+    )
+    cumsum_unsq = ov_opset.convert(cumsum_unsq, Type.i32)
+    mask = ov_opset.greater_equal(out_indices_unsq, cumsum_unsq)
+    gather_indices = ov_opset.reduce_sum(
+        ov_opset.convert(mask, Type.i32),
+        ov_opset.constant([1], Type.i32),
+        keep_dims=False,
+    )
+
+    result = ov_opset.gather(
+        x, gather_indices, ov_opset.constant(axis, Type.i32)
+    ).output(0)
+    return OpenVINOKerasTensor(result)
 
 
 def reshape(x, newshape):