def test_triu_square(self):
node = onnx.helper.make_node(
'Trilu',
inputs=['x'],
outputs=['y'],
domain="com.microsoft",
)
x = np.random.randn(3, 5, 5).astype(np.float32)
y = triu_reference_implementation(x)
expect(node, inputs=[x], outputs=[y], name='test_triu_square')
def test_triu(self):
node = onnx.helper.make_node(
"Trilu",
inputs=["x"],
outputs=["y"],
domain="com.microsoft",
)
x = np.random.randn(3, 4, 5).astype(np.float32)
y = triu_reference_implementation(x)
expect(node, inputs=[x], outputs=[y], name="test_triu")
def test_torch_embedding(self):
node = onnx.helper.make_node(
'TorchEmbedding',
inputs=['w', 'x'],
outputs=['y'],
domain="com.microsoft",
)
x = np.random.randn(2, 4).astype(np.int64)
w = np.random.randn(10, 3).astype(np.float32)
y = torch_embedding_reference_implementation(w, x)
expect(node, inputs=[w, x], outputs=[y], name='test_torch_embedding')
def test_triu_square_neg(self):
node = onnx.helper.make_node(
'Trilu',
inputs=['x', 'k'],
outputs=['y'],
domain="com.microsoft",
)
x = np.random.randn(3, 5, 5).astype(np.float32)
k = np.array([-1]).astype(np.int64)
y = triu_reference_implementation(x, k)
expect(node, inputs=[x, k], outputs=[y], name='test_triu_square_neg')
def test_triu_out_pos(self):
node = onnx.helper.make_node(
"Trilu",
inputs=["x", "k"],
outputs=["y"],
domain="com.microsoft",
)
x = np.random.randn(3, 4, 5).astype(np.float32)
k = np.array([6]).astype(np.int64)
y = triu_reference_implementation(x, k)
expect(node, inputs=[x, k], outputs=[y], name="test_triu_out_pos")
def test_tril_one_row_neg(self):
node = onnx.helper.make_node(
"Trilu",
inputs=["x", "k"],
outputs=["y"],
upper=0,
domain="com.microsoft",
)
x = np.random.randn(3, 1, 5).astype(np.float32)
k = np.array([-7]).astype(np.int64)
y = tril_reference_implementation(x, k)
expect(node, inputs=[x, k], outputs=[y], name="test_tril_one_row_neg")
def test_torch_embedding_zero_dim(self):
node = onnx.helper.make_node(
"TorchEmbedding",
inputs=["w", "x"],
outputs=["y"],
domain="com.microsoft",
)
x = np.random.randn(0, 4).astype(np.int64)
w = np.random.randn(10, 3).astype(np.float32)
y = torch_embedding_reference_implementation(w, x)
expect(node,
inputs=[w, x],
outputs=[y],
name="test_torch_embedding_zero_dim")
def test_torch_embedding_scale_grad_by_freq(self):
node = onnx.helper.make_node(
'TorchEmbedding',
inputs=['w', 'x', 'padding_idx', 'scale'],
outputs=['y'],
domain="com.microsoft",
)
x = np.random.randn(3, 4).astype(np.int64)
w = np.random.randn(10, 3).astype(np.float32)
padding_idx = np.random.randint(3, size=1).astype(np.int64)
scale = np.array([1]).astype(np.bool)
y = torch_embedding_reference_implementation(w, x, padding_idx, scale)
expect(node,
inputs=[w, x, padding_idx, scale],
outputs=[y],
name='test_torch_embedding_scale_grad_by_freq')
def test_torch_embedding_padding_idx(self):
node = onnx.helper.make_node(
"TorchEmbedding",
inputs=["w", "x", "padding_idx"],
outputs=["y"],
domain="com.microsoft",
)
x = np.random.randn(3, 4).astype(np.int64)
w = np.random.randn(10, 3).astype(np.float32)
padding_idx = np.random.randint(3, size=1).astype(np.int64)
y = torch_embedding_reference_implementation(w, x, padding_idx)
expect(
node,
inputs=[w, x, padding_idx],
outputs=[y],
name="test_torch_embedding_padding_idx",
)
