File size: 1,977 Bytes
96e64e9 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 |
# Copyright (c) 2024 NVIDIA CORPORATION.
# Licensed under the MIT license.
import os
import sys
# to import modules from parent_dir
parent_dir = os.path.abspath(os.path.join(os.path.dirname(__file__), ".."))
sys.path.append(parent_dir)
import torch
from alias_free_activation.cuda import activation1d
from activations import SnakeBeta
def test_load_fused_kernels():
try:
print("[Success] load_fused_kernels")
except ImportError as e:
print("[Fail] load_fused_kernels")
raise e
def test_anti_alias_activation():
data = torch.rand((10, 10, 200), device="cuda")
# Check activations, Snake CUDA vs. Torch
fused_anti_alias_activation = activation1d.Activation1d(
activation=SnakeBeta(10), fused=True
).cuda()
fused_activation_output = fused_anti_alias_activation(data)
torch_anti_alias_activation = activation1d.Activation1d(
activation=SnakeBeta(10), fused=False
).cuda()
torch_activation_output = torch_anti_alias_activation(data)
test_result = (fused_activation_output - torch_activation_output).abs()
while test_result.dim() != 1:
test_result = test_result.mean(dim=-1)
diff = test_result.mean(dim=-1)
if diff <= 1e-3:
print(
f"\n[Success] test_fused_anti_alias_activation"
f"\n > mean_difference={diff}"
f"\n > fused_values={fused_activation_output[-1][-1][:].tolist()}"
f"\n > torch_values={torch_activation_output[-1][-1][:].tolist()}"
)
else:
print(
f"\n[Fail] test_fused_anti_alias_activation"
f"\n > mean_difference={diff}, "
f"\n > fused_values={fused_activation_output[-1][-1][:].tolist()}, "
f"\n > torch_values={torch_activation_output[-1][-1][:].tolist()}"
)
if __name__ == "__main__":
from alias_free_activation.cuda import load
load.load()
test_load_fused_kernels()
test_anti_alias_activation()
|