Spaces:
Running
on
Zero
Running
on
Zero
File size: 6,228 Bytes
a49cc2f |
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 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 |
# coding=utf-8
# Copyright 2024 HuggingFace Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import unittest
import torch
from diffusers import AutoencoderKLHunyuanVideo
from diffusers.utils.testing_utils import (
enable_full_determinism,
floats_tensor,
torch_device,
)
from ..test_modeling_common import ModelTesterMixin, UNetTesterMixin
enable_full_determinism()
class AutoencoderKLHunyuanVideoTests(ModelTesterMixin, UNetTesterMixin, unittest.TestCase):
model_class = AutoencoderKLHunyuanVideo
main_input_name = "sample"
base_precision = 1e-2
def get_autoencoder_kl_hunyuan_video_config(self):
return {
"in_channels": 3,
"out_channels": 3,
"latent_channels": 4,
"down_block_types": (
"HunyuanVideoDownBlock3D",
"HunyuanVideoDownBlock3D",
"HunyuanVideoDownBlock3D",
"HunyuanVideoDownBlock3D",
),
"up_block_types": (
"HunyuanVideoUpBlock3D",
"HunyuanVideoUpBlock3D",
"HunyuanVideoUpBlock3D",
"HunyuanVideoUpBlock3D",
),
"block_out_channels": (8, 8, 8, 8),
"layers_per_block": 1,
"act_fn": "silu",
"norm_num_groups": 4,
"scaling_factor": 0.476986,
"spatial_compression_ratio": 8,
"temporal_compression_ratio": 4,
"mid_block_add_attention": True,
}
@property
def dummy_input(self):
batch_size = 2
num_frames = 9
num_channels = 3
sizes = (16, 16)
image = floats_tensor((batch_size, num_channels, num_frames) + sizes).to(torch_device)
return {"sample": image}
@property
def input_shape(self):
return (3, 9, 16, 16)
@property
def output_shape(self):
return (3, 9, 16, 16)
def prepare_init_args_and_inputs_for_common(self):
init_dict = self.get_autoencoder_kl_hunyuan_video_config()
inputs_dict = self.dummy_input
return init_dict, inputs_dict
def test_enable_disable_tiling(self):
init_dict, inputs_dict = self.prepare_init_args_and_inputs_for_common()
torch.manual_seed(0)
model = self.model_class(**init_dict).to(torch_device)
inputs_dict.update({"return_dict": False})
torch.manual_seed(0)
output_without_tiling = model(**inputs_dict, generator=torch.manual_seed(0))[0]
torch.manual_seed(0)
model.enable_tiling()
output_with_tiling = model(**inputs_dict, generator=torch.manual_seed(0))[0]
self.assertLess(
(output_without_tiling.detach().cpu().numpy() - output_with_tiling.detach().cpu().numpy()).max(),
0.5,
"VAE tiling should not affect the inference results",
)
torch.manual_seed(0)
model.disable_tiling()
output_without_tiling_2 = model(**inputs_dict, generator=torch.manual_seed(0))[0]
self.assertEqual(
output_without_tiling.detach().cpu().numpy().all(),
output_without_tiling_2.detach().cpu().numpy().all(),
"Without tiling outputs should match with the outputs when tiling is manually disabled.",
)
def test_enable_disable_slicing(self):
init_dict, inputs_dict = self.prepare_init_args_and_inputs_for_common()
torch.manual_seed(0)
model = self.model_class(**init_dict).to(torch_device)
inputs_dict.update({"return_dict": False})
torch.manual_seed(0)
output_without_slicing = model(**inputs_dict, generator=torch.manual_seed(0))[0]
torch.manual_seed(0)
model.enable_slicing()
output_with_slicing = model(**inputs_dict, generator=torch.manual_seed(0))[0]
self.assertLess(
(output_without_slicing.detach().cpu().numpy() - output_with_slicing.detach().cpu().numpy()).max(),
0.5,
"VAE slicing should not affect the inference results",
)
torch.manual_seed(0)
model.disable_slicing()
output_without_slicing_2 = model(**inputs_dict, generator=torch.manual_seed(0))[0]
self.assertEqual(
output_without_slicing.detach().cpu().numpy().all(),
output_without_slicing_2.detach().cpu().numpy().all(),
"Without slicing outputs should match with the outputs when slicing is manually disabled.",
)
def test_gradient_checkpointing_is_applied(self):
expected_set = {
"HunyuanVideoDecoder3D",
"HunyuanVideoDownBlock3D",
"HunyuanVideoEncoder3D",
"HunyuanVideoMidBlock3D",
"HunyuanVideoUpBlock3D",
}
super().test_gradient_checkpointing_is_applied(expected_set=expected_set)
# We need to overwrite this test because the base test does not account length of down_block_types
def test_forward_with_norm_groups(self):
init_dict, inputs_dict = self.prepare_init_args_and_inputs_for_common()
init_dict["norm_num_groups"] = 16
init_dict["block_out_channels"] = (16, 16, 16, 16)
model = self.model_class(**init_dict)
model.to(torch_device)
model.eval()
with torch.no_grad():
output = model(**inputs_dict)
if isinstance(output, dict):
output = output.to_tuple()[0]
self.assertIsNotNone(output)
expected_shape = inputs_dict["sample"].shape
self.assertEqual(output.shape, expected_shape, "Input and output shapes do not match")
@unittest.skip("Unsupported test.")
def test_outputs_equivalence(self):
pass
|