import unittest
import random
import torch
import torch_npu
import hypothesis
from torch_npu.testing.testcase import TestCase, run_tests
from torch_npu.testing.common_utils import SupportedDevices
class TestForeachNorm(TestCase):
torch_dtypes = {
"float16": torch.float16,
"float32": torch.float32,
"bfloat16": torch.bfloat16
}
def assert_equal(self, cpu_outs, npu_outs):
for cpu_out, npu_out in zip(cpu_outs, npu_outs):
if cpu_out.dtype != npu_out.dtype:
self.fail("dtype error!")
if (cpu_out.shape != npu_out.shape):
self.fail("shape error!")
result = torch.allclose(cpu_out, npu_out.cpu(), rtol=0.001, atol=0.001)
if not result:
self.fail("result error!")
return True
def create_tensors(self, dtype, shapes):
cpu_tensors = []
npu_tensors = []
for shape in shapes:
t = torch.randn((shape[0], shape[1]), dtype=self.torch_dtypes.get(dtype))
if dtype == "float16" or dtype == "bfloat16":
cpu_tensors.append(t.float())
else:
cpu_tensors.append(t)
npu_tensors.append(t.npu())
return tuple(cpu_tensors), tuple(npu_tensors)
def create_input_tensors(self, tensor_num, dtype):
input_nums = 1
cpu_inputs = []
npu_inputs = []
shapes = []
for i in range(tensor_num):
m = random.randint(1, 100)
n = random.randint(1, 100)
shapes.append([m, n])
for i in range(input_nums):
cpu_tensors, npu_tensors = self.create_tensors(dtype, shapes)
cpu_inputs.append(cpu_tensors)
npu_inputs.append(npu_tensors)
return cpu_inputs, npu_inputs
def test_foreach_norm_out_float32_shape_tensor_num(self):
tensor_num_list = [20, 50]
for tensor_num in tensor_num_list:
cpu_tensors, npu_tensors = self.create_input_tensors(tensor_num, "float32")
cpu_output = torch._foreach_norm(cpu_tensors[0], ord=2, dtype=torch.float32)
npu_output = torch._foreach_norm(npu_tensors[0], ord=2, dtype=torch.float32)
self.assert_equal(cpu_output, npu_output)
def test_foreach_norm_out_float16_shape_tensor_num(self):
tensor_num_list = [20, 50]
for tensor_num in tensor_num_list:
cpu_tensors, npu_tensors = self.create_input_tensors(tensor_num, "float16")
cpu_output = tuple([out for out in torch._foreach_norm(cpu_tensors[0], ord=1, dtype=torch.float32)])
npu_output = torch._foreach_norm(npu_tensors[0], ord=1, dtype=torch.float32)
self.assert_equal(cpu_output, npu_output)
@SupportedDevices(['Ascend910B'])
def test_foreach_norm_out_bfloat16_shape_tensor_num(self):
tensor_num_list = [20, 50]
for tensor_num in tensor_num_list:
cpu_tensors, npu_tensors = self.create_input_tensors(tensor_num, "bfloat16")
cpu_output = tuple([out for out in torch._foreach_norm(cpu_tensors[0], ord=2, dtype=torch.float32)])
npu_output = torch._foreach_norm(npu_tensors[0], ord=2, dtype=torch.float32)
self.assert_equal(cpu_output, npu_output)
if __name__ == "__main__":
run_tests()