ultralytics-thop 2.0.14__py3-none-any.whl → 2.0.15__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- thop/__init__.py +1 -1
- thop/vision/basic_hooks.py +2 -1
- ultralytics_thop-2.0.15.dist-info/METADATA +258 -0
- ultralytics_thop-2.0.15.dist-info/RECORD +13 -0
- {ultralytics_thop-2.0.14.dist-info → ultralytics_thop-2.0.15.dist-info}/WHEEL +1 -1
- ultralytics_thop-2.0.14.dist-info/METADATA +0 -199
- ultralytics_thop-2.0.14.dist-info/RECORD +0 -13
- {ultralytics_thop-2.0.14.dist-info → ultralytics_thop-2.0.15.dist-info/licenses}/LICENSE +0 -0
- {ultralytics_thop-2.0.14.dist-info → ultralytics_thop-2.0.15.dist-info}/top_level.txt +0 -0
thop/__init__.py
CHANGED
thop/vision/basic_hooks.py
CHANGED
|
@@ -0,0 +1,258 @@
|
|
|
1
|
+
Metadata-Version: 2.4
|
|
2
|
+
Name: ultralytics-thop
|
|
3
|
+
Version: 2.0.15
|
|
4
|
+
Summary: Ultralytics THOP package for fast computation of PyTorch model FLOPs and parameters.
|
|
5
|
+
Author-email: Ligeng Zhu <ligeng.zhu+github@gmail.com>
|
|
6
|
+
Maintainer-email: Ultralytics <hello@ultralytics.com>
|
|
7
|
+
License: AGPL-3.0
|
|
8
|
+
Project-URL: Homepage, https://ultralytics.com
|
|
9
|
+
Project-URL: Source, https://github.com/ultralytics/thop
|
|
10
|
+
Project-URL: Documentation, https://docs.ultralytics.com
|
|
11
|
+
Project-URL: Bug Reports, https://github.com/ultralytics/thop/issues
|
|
12
|
+
Project-URL: Changelog, https://github.com/ultralytics/thop/releases
|
|
13
|
+
Keywords: FLOPs,PyTorch,Model Analysis
|
|
14
|
+
Classifier: Development Status :: 4 - Beta
|
|
15
|
+
Classifier: Intended Audience :: Developers
|
|
16
|
+
Classifier: Intended Audience :: Education
|
|
17
|
+
Classifier: Intended Audience :: Science/Research
|
|
18
|
+
Classifier: License :: OSI Approved :: GNU Affero General Public License v3 or later (AGPLv3+)
|
|
19
|
+
Classifier: Programming Language :: Python :: 3
|
|
20
|
+
Classifier: Programming Language :: Python :: 3.8
|
|
21
|
+
Classifier: Programming Language :: Python :: 3.9
|
|
22
|
+
Classifier: Programming Language :: Python :: 3.10
|
|
23
|
+
Classifier: Programming Language :: Python :: 3.11
|
|
24
|
+
Classifier: Programming Language :: Python :: 3.12
|
|
25
|
+
Classifier: Programming Language :: Python :: 3.13
|
|
26
|
+
Classifier: Topic :: Software Development
|
|
27
|
+
Classifier: Topic :: Scientific/Engineering
|
|
28
|
+
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
|
|
29
|
+
Classifier: Topic :: Scientific/Engineering :: Image Recognition
|
|
30
|
+
Classifier: Operating System :: POSIX :: Linux
|
|
31
|
+
Classifier: Operating System :: MacOS
|
|
32
|
+
Classifier: Operating System :: Microsoft :: Windows
|
|
33
|
+
Requires-Python: >=3.8
|
|
34
|
+
Description-Content-Type: text/markdown
|
|
35
|
+
License-File: LICENSE
|
|
36
|
+
Requires-Dist: numpy
|
|
37
|
+
Requires-Dist: torch
|
|
38
|
+
Dynamic: license-file
|
|
39
|
+
|
|
40
|
+
<a href="https://www.ultralytics.com/"><img src="https://raw.githubusercontent.com/ultralytics/assets/main/logo/Ultralytics_Logotype_Original.svg" width="320" alt="Ultralytics logo"></a>
|
|
41
|
+
|
|
42
|
+
# 🚀 THOP: PyTorch-OpCounter
|
|
43
|
+
|
|
44
|
+
Welcome to the [THOP](https://github.com/ultralytics/thop) repository, your comprehensive solution for profiling [PyTorch](https://pytorch.org/) models by computing the number of Multiply-Accumulate Operations (MACs) and parameters. Developed by Ultralytics, this tool is essential for [deep learning](https://www.ultralytics.com/glossary/deep-learning-dl) practitioners aiming to evaluate model efficiency and performance, crucial aspects discussed in our [model training tips guide](https://docs.ultralytics.com/guides/model-training-tips/).
|
|
45
|
+
|
|
46
|
+
[](https://github.com/ultralytics/thop/actions/workflows/format.yml)
|
|
47
|
+
[](https://discord.com/invite/ultralytics)
|
|
48
|
+
[](https://community.ultralytics.com/)
|
|
49
|
+
[](https://reddit.com/r/ultralytics)
|
|
50
|
+
|
|
51
|
+
## 📄 Description
|
|
52
|
+
|
|
53
|
+
THOP offers an intuitive API designed to profile PyTorch models by calculating the total number of MACs and parameters. This functionality is vital for assessing the computational efficiency and memory footprint of deep learning models, helping developers optimize performance for deployment, especially on [edge devices](https://www.ultralytics.com/glossary/edge-ai). Understanding these metrics is key to selecting the right model architecture, a topic explored in our [model comparison pages](https://docs.ultralytics.com/compare/).
|
|
54
|
+
|
|
55
|
+
## 📦 Installation
|
|
56
|
+
|
|
57
|
+
Get started with THOP quickly by installing it via pip:
|
|
58
|
+
|
|
59
|
+
[](https://pypi.org/project/ultralytics-thop/) [](https://www.pepy.tech/projects/ultralytics-thop) [](https://pypi.org/project/ultralytics-thop/)
|
|
60
|
+
|
|
61
|
+
```bash
|
|
62
|
+
pip install ultralytics-thop
|
|
63
|
+
```
|
|
64
|
+
|
|
65
|
+
Alternatively, for the latest features and updates, install directly from the GitHub repository:
|
|
66
|
+
|
|
67
|
+
```bash
|
|
68
|
+
pip install --upgrade git+https://github.com/ultralytics/thop.git
|
|
69
|
+
```
|
|
70
|
+
|
|
71
|
+
This ensures you have the most recent version, incorporating the latest improvements and bug fixes.
|
|
72
|
+
|
|
73
|
+
## 🛠️ How to Use
|
|
74
|
+
|
|
75
|
+
### Basic Usage
|
|
76
|
+
|
|
77
|
+
Profiling a standard PyTorch model like [ResNet50](https://pytorch.org/vision/main/models/generated/torchvision.models.resnet50.html) is straightforward. Import the necessary libraries, load your model and a sample input tensor, then use the `profile` function:
|
|
78
|
+
|
|
79
|
+
```python
|
|
80
|
+
import torch
|
|
81
|
+
from torchvision.models import resnet50 # Example model
|
|
82
|
+
|
|
83
|
+
from thop import profile # Import the profile function from THOP
|
|
84
|
+
|
|
85
|
+
# Load a pre-trained model (e.g., ResNet50)
|
|
86
|
+
model = resnet50()
|
|
87
|
+
|
|
88
|
+
# Create a dummy input tensor matching the model's expected input shape
|
|
89
|
+
dummy_input = torch.randn(1, 3, 224, 224)
|
|
90
|
+
|
|
91
|
+
# Profile the model
|
|
92
|
+
macs, params = profile(model, inputs=(dummy_input,))
|
|
93
|
+
|
|
94
|
+
print(f"MACs: {macs}, Parameters: {params}")
|
|
95
|
+
# Expected output: MACs: 4139975680.0, Parameters: 25557032.0
|
|
96
|
+
```
|
|
97
|
+
|
|
98
|
+
### Define Custom Rules for Third-Party Modules
|
|
99
|
+
|
|
100
|
+
If your model includes custom or third-party modules not natively supported by THOP, you can define custom profiling rules using the `custom_ops` argument. This allows for accurate profiling even with complex or non-standard architectures, which is useful when working with models like those found in the [Ultralytics models section](https://docs.ultralytics.com/models/).
|
|
101
|
+
|
|
102
|
+
```python
|
|
103
|
+
import torch
|
|
104
|
+
import torch.nn as nn
|
|
105
|
+
|
|
106
|
+
from thop import profile
|
|
107
|
+
|
|
108
|
+
|
|
109
|
+
# Define your custom module
|
|
110
|
+
class YourCustomModule(nn.Module):
|
|
111
|
+
def __init__(self):
|
|
112
|
+
super().__init__()
|
|
113
|
+
# Define layers, e.g., a convolution
|
|
114
|
+
self.conv = nn.Conv2d(3, 64, kernel_size=3, stride=1, padding=1)
|
|
115
|
+
|
|
116
|
+
def forward(self, x):
|
|
117
|
+
return self.conv(x)
|
|
118
|
+
|
|
119
|
+
|
|
120
|
+
# Define a custom counting function for your module
|
|
121
|
+
# This function should calculate and return the MACs for the module's operations
|
|
122
|
+
def count_your_custom_module(module, x, y):
|
|
123
|
+
# Example: Calculate MACs for the conv layer
|
|
124
|
+
# Note: This is a simplified example. Real calculations depend on the module's specifics.
|
|
125
|
+
# MACs = output_height * output_width * kernel_height * kernel_width * in_channels * out_channels
|
|
126
|
+
# For simplicity, we'll just assign a placeholder value or use a helper if available
|
|
127
|
+
# In a real scenario, you'd implement the precise MAC calculation here.
|
|
128
|
+
# For nn.Conv2d, THOP usually handles it, but this demonstrates the concept.
|
|
129
|
+
macs = 0 # Placeholder: Implement actual MAC calculation based on module logic
|
|
130
|
+
# You might need access to module properties like kernel_size, stride, padding, channels etc.
|
|
131
|
+
# Example for a Conv2d layer (simplified):
|
|
132
|
+
if isinstance(module, nn.Conv2d):
|
|
133
|
+
_, _, H, W = y.shape # Output shape
|
|
134
|
+
k_h, k_w = module.kernel_size
|
|
135
|
+
in_c = module.in_channels
|
|
136
|
+
out_c = module.out_channels
|
|
137
|
+
groups = module.groups
|
|
138
|
+
macs = (k_h * k_w * in_c * out_c * H * W) / groups
|
|
139
|
+
module.total_ops += torch.DoubleTensor([macs]) # Accumulate MACs
|
|
140
|
+
|
|
141
|
+
|
|
142
|
+
# Instantiate a model containing your custom module
|
|
143
|
+
model = YourCustomModule() # Or a larger model incorporating this module
|
|
144
|
+
|
|
145
|
+
# Create a dummy input
|
|
146
|
+
dummy_input = torch.randn(1, 3, 224, 224)
|
|
147
|
+
|
|
148
|
+
# Profile the model, providing the custom operation mapping
|
|
149
|
+
macs, params = profile(model, inputs=(dummy_input,), custom_ops={YourCustomModule: count_your_custom_module})
|
|
150
|
+
|
|
151
|
+
print(f"Custom MACs: {macs}, Parameters: {params}")
|
|
152
|
+
# Expected output: Custom MACs: 87457792.0, Parameters: 1792.0
|
|
153
|
+
```
|
|
154
|
+
|
|
155
|
+
### Improve Output Readability
|
|
156
|
+
|
|
157
|
+
For clearer and more interpretable results, use the `thop.clever_format` function. This formats the raw MACs and parameter counts into human-readable strings (e.g., GigaMACs, MegaParams). This formatting helps in quickly understanding the scale of computational resources required, similar to the metrics provided in our [Ultralytics YOLOv8 documentation](https://docs.ultralytics.com/models/yolov8/).
|
|
158
|
+
|
|
159
|
+
```python
|
|
160
|
+
import torch
|
|
161
|
+
from torchvision.models import resnet50
|
|
162
|
+
|
|
163
|
+
from thop import clever_format, profile
|
|
164
|
+
|
|
165
|
+
model = resnet50()
|
|
166
|
+
dummy_input = torch.randn(1, 3, 224, 224)
|
|
167
|
+
macs, params = profile(model, inputs=(dummy_input,))
|
|
168
|
+
|
|
169
|
+
# Format the numbers into a readable format (e.g., 4.14 GMac, 25.56 MParams)
|
|
170
|
+
macs_readable, params_readable = clever_format([macs, params], "%.3f")
|
|
171
|
+
|
|
172
|
+
print(f"Formatted MACs: {macs_readable}, Formatted Parameters: {params_readable}")
|
|
173
|
+
# Expected output: Formatted MACs: 4.140G, Formatted Parameters: 25.557M
|
|
174
|
+
```
|
|
175
|
+
|
|
176
|
+
## 📊 Results of Recent Models
|
|
177
|
+
|
|
178
|
+
The table below showcases the parameters and MACs for several popular [computer vision](https://www.ultralytics.com/glossary/computer-vision-cv) models, profiled using THOP. These benchmarks provide a comparative overview of model complexity and computational cost. You can reproduce these results by running the script located at `benchmark/evaluate_famous_models.py` in this repository. Comparing these metrics is essential for tasks like selecting models for [object detection](https://www.ultralytics.com/glossary/object-detection) or [image classification](https://www.ultralytics.com/glossary/image-classification). For more comparisons, see our [model comparison section](https://docs.ultralytics.com/compare/).
|
|
179
|
+
|
|
180
|
+
<table align="center">
|
|
181
|
+
<tr>
|
|
182
|
+
<td>
|
|
183
|
+
|
|
184
|
+
| Model | Params(M) | MACs(G) |
|
|
185
|
+
| ---------------- | --------- | ------- |
|
|
186
|
+
| alexnet | 61.10 | 0.77 |
|
|
187
|
+
| vgg11 | 132.86 | 7.74 |
|
|
188
|
+
| vgg11_bn | 132.87 | 7.77 |
|
|
189
|
+
| vgg13 | 133.05 | 11.44 |
|
|
190
|
+
| vgg13_bn | 133.05 | 11.49 |
|
|
191
|
+
| vgg16 | 138.36 | 15.61 |
|
|
192
|
+
| vgg16_bn | 138.37 | 15.66 |
|
|
193
|
+
| vgg19 | 143.67 | 19.77 |
|
|
194
|
+
| vgg19_bn | 143.68 | 19.83 |
|
|
195
|
+
| resnet18 | 11.69 | 1.82 |
|
|
196
|
+
| resnet34 | 21.80 | 3.68 |
|
|
197
|
+
| resnet50 | 25.56 | 4.14 |
|
|
198
|
+
| resnet101 | 44.55 | 7.87 |
|
|
199
|
+
| resnet152 | 60.19 | 11.61 |
|
|
200
|
+
| wide_resnet101_2 | 126.89 | 22.84 |
|
|
201
|
+
| wide_resnet50_2 | 68.88 | 11.46 |
|
|
202
|
+
|
|
203
|
+
</td>
|
|
204
|
+
<td>
|
|
205
|
+
|
|
206
|
+
| Model | Params(M) | MACs(G) |
|
|
207
|
+
| ------------------ | --------- | ------- |
|
|
208
|
+
| resnext50_32x4d | 25.03 | 4.29 |
|
|
209
|
+
| resnext101_32x8d | 88.79 | 16.54 |
|
|
210
|
+
| densenet121 | 7.98 | 2.90 |
|
|
211
|
+
| densenet161 | 28.68 | 7.85 |
|
|
212
|
+
| densenet169 | 14.15 | 3.44 |
|
|
213
|
+
| densenet201 | 20.01 | 4.39 |
|
|
214
|
+
| squeezenet1_0 | 1.25 | 0.82 |
|
|
215
|
+
| squeezenet1_1 | 1.24 | 0.35 |
|
|
216
|
+
| mnasnet0_5 | 2.22 | 0.14 |
|
|
217
|
+
| mnasnet0_75 | 3.17 | 0.24 |
|
|
218
|
+
| mnasnet1_0 | 4.38 | 0.34 |
|
|
219
|
+
| mnasnet1_3 | 6.28 | 0.53 |
|
|
220
|
+
| mobilenet_v2 | 3.50 | 0.33 |
|
|
221
|
+
| shufflenet_v2_x0_5 | 1.37 | 0.05 |
|
|
222
|
+
| shufflenet_v2_x1_0 | 2.28 | 0.15 |
|
|
223
|
+
| shufflenet_v2_x1_5 | 3.50 | 0.31 |
|
|
224
|
+
| shufflenet_v2_x2_0 | 7.39 | 0.60 |
|
|
225
|
+
| inception_v3 | 27.16 | 5.75 |
|
|
226
|
+
|
|
227
|
+
</td>
|
|
228
|
+
</tr>
|
|
229
|
+
</table>
|
|
230
|
+
|
|
231
|
+
## 🙌 Contribute
|
|
232
|
+
|
|
233
|
+
We actively welcome and encourage community contributions to make THOP even better! Whether it's adding support for new [PyTorch layers](https://pytorch.org/docs/stable/nn.html), improving existing calculations, enhancing documentation, or fixing bugs, your input is valuable. Please see our [Contributing Guide](https://docs.ultralytics.com/help/contributing/) for detailed instructions on how to participate. Together, we can ensure THOP remains a state-of-the-art tool for the [machine learning](https://www.ultralytics.com/glossary/machine-learning-ml) community. Don't hesitate to share your feedback and suggestions!
|
|
234
|
+
|
|
235
|
+
## 📜 License
|
|
236
|
+
|
|
237
|
+
THOP is distributed under the [AGPL-3.0 License](https://www.gnu.org/licenses/agpl-3.0.en.html). This license promotes open collaboration and sharing of improvements. For complete details, please refer to the [LICENSE](https://github.com/ultralytics/thop/blob/main/LICENSE) file included in the repository. Understanding the license is important before integrating THOP into your projects, especially for commercial applications which may require an [Enterprise License](https://www.ultralytics.com/license).
|
|
238
|
+
|
|
239
|
+
## 📧 Contact
|
|
240
|
+
|
|
241
|
+
Encountered a bug or have a feature request? Please submit an issue through our [GitHub Issues](https://github.com/ultralytics/thop/issues) page. For general discussions, questions, and community support, join the vibrant Ultralytics community on our [Discord server](https://discord.com/invite/ultralytics). We look forward to hearing from you and collaborating!
|
|
242
|
+
|
|
243
|
+
<br>
|
|
244
|
+
<div align="center">
|
|
245
|
+
<a href="https://github.com/ultralytics"><img src="https://github.com/ultralytics/assets/raw/main/social/logo-social-github.png" width="3%" alt="Ultralytics GitHub"></a>
|
|
246
|
+
<img src="https://github.com/ultralytics/assets/raw/main/social/logo-transparent.png" width="3%" alt="space">
|
|
247
|
+
<a href="https://www.linkedin.com/company/ultralytics/"><img src="https://github.com/ultralytics/assets/raw/main/social/logo-social-linkedin.png" width="3%" alt="Ultralytics LinkedIn"></a>
|
|
248
|
+
<img src="https://github.com/ultralytics/assets/raw/main/social/logo-transparent.png" width="3%" alt="space">
|
|
249
|
+
<a href="https://twitter.com/ultralytics"><img src="https://github.com/ultralytics/assets/raw/main/social/logo-social-twitter.png" width="3%" alt="Ultralytics Twitter"></a>
|
|
250
|
+
<img src="https://github.com/ultralytics/assets/raw/main/social/logo-transparent.png" width="3%" alt="space">
|
|
251
|
+
<a href="https://youtube.com/ultralytics?sub_confirmation=1"><img src="https://github.com/ultralytics/assets/raw/main/social/logo-social-youtube.png" width="3%" alt="Ultralytics YouTube"></a>
|
|
252
|
+
<img src="https://github.com/ultralytics/assets/raw/main/social/logo-transparent.png" width="3%" alt="space">
|
|
253
|
+
<a href="https://www.tiktok.com/@ultralytics"><img src="https://github.com/ultralytics/assets/raw/main/social/logo-social-tiktok.png" width="3%" alt="Ultralytics TikTok"></a>
|
|
254
|
+
<img src="https://github.com/ultralytics/assets/raw/main/social/logo-transparent.png" width="3%" alt="space">
|
|
255
|
+
<a href="https://ultralytics.com/bilibili"><img src="https://github.com/ultralytics/assets/raw/main/social/logo-social-bilibili.png" width="3%" alt="Ultralytics BiliBili"></a>
|
|
256
|
+
<img src="https://github.com/ultralytics/assets/raw/main/social/logo-transparent.png" width="3%" alt="space">
|
|
257
|
+
<a href="https://discord.com/invite/ultralytics"><img src="https://github.com/ultralytics/assets/raw/main/social/logo-social-discord.png" width="3%" alt="Ultralytics Discord"></a>
|
|
258
|
+
</div>
|
|
@@ -0,0 +1,13 @@
|
|
|
1
|
+
thop/__init__.py,sha256=mEWL6Ykse9frz8z9HIb5QWmXY9QpDL41UP2yStOsVaM,219
|
|
2
|
+
thop/fx_profile.py,sha256=ACI2RLyDYBLB7Cru9y2IAx_YCFX_uhYEArDR0np_tFc,8231
|
|
3
|
+
thop/profile.py,sha256=VaCR0K2oWTHp0y-W3SbuA4TdnyxTP6n6VSURltJGzrE,7980
|
|
4
|
+
thop/rnn_hooks.py,sha256=JKZ2eSCvIKvhvCDqM4oWPZjmBkdyJ4R2Q7XSn63lsX0,6503
|
|
5
|
+
thop/utils.py,sha256=IwFJQ1v-SLyhm-313Li535R6fhtomkm8Fem1Kfe6G_U,1484
|
|
6
|
+
thop/vision/__init__.py,sha256=lm6MckFYCPTbqIoX7w0s_daxdjNeBeKW6DXppv1-QUM,70
|
|
7
|
+
thop/vision/basic_hooks.py,sha256=UKMHbH-dP2wx0LCaWVMHtS9o5hJsClmFQrJrPFm0uxY,4714
|
|
8
|
+
thop/vision/calc_func.py,sha256=s-_tRmimNBl4PYY_WQHb3C1H4nLy-oEW2kvsZDVjvdI,4362
|
|
9
|
+
ultralytics_thop-2.0.15.dist-info/licenses/LICENSE,sha256=hIahDEOTzuHCU5J2nd07LWwkLW7Hko4UFO__ffsvB-8,34523
|
|
10
|
+
ultralytics_thop-2.0.15.dist-info/METADATA,sha256=8pRLCm325cTHHSpygLu0DlmP8XR9rQR-qwytLJ4PYTw,14922
|
|
11
|
+
ultralytics_thop-2.0.15.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
|
|
12
|
+
ultralytics_thop-2.0.15.dist-info/top_level.txt,sha256=HQ7D0gSvDJ31CNR-f0EuXNVve05RYBmwyIkHQKiEhU8,5
|
|
13
|
+
ultralytics_thop-2.0.15.dist-info/RECORD,,
|
|
@@ -1,199 +0,0 @@
|
|
|
1
|
-
Metadata-Version: 2.2
|
|
2
|
-
Name: ultralytics-thop
|
|
3
|
-
Version: 2.0.14
|
|
4
|
-
Summary: Ultralytics THOP package for fast computation of PyTorch model FLOPs and parameters.
|
|
5
|
-
Author-email: Ligeng Zhu <ligeng.zhu+github@gmail.com>
|
|
6
|
-
Maintainer-email: Ultralytics <hello@ultralytics.com>
|
|
7
|
-
License: AGPL-3.0
|
|
8
|
-
Project-URL: Homepage, https://ultralytics.com
|
|
9
|
-
Project-URL: Source, https://github.com/ultralytics/thop
|
|
10
|
-
Project-URL: Documentation, https://docs.ultralytics.com
|
|
11
|
-
Project-URL: Bug Reports, https://github.com/ultralytics/thop/issues
|
|
12
|
-
Project-URL: Changelog, https://github.com/ultralytics/thop/releases
|
|
13
|
-
Keywords: FLOPs,PyTorch,Model Analysis
|
|
14
|
-
Classifier: Development Status :: 4 - Beta
|
|
15
|
-
Classifier: Intended Audience :: Developers
|
|
16
|
-
Classifier: Intended Audience :: Education
|
|
17
|
-
Classifier: Intended Audience :: Science/Research
|
|
18
|
-
Classifier: License :: OSI Approved :: GNU Affero General Public License v3 or later (AGPLv3+)
|
|
19
|
-
Classifier: Programming Language :: Python :: 3
|
|
20
|
-
Classifier: Programming Language :: Python :: 3.8
|
|
21
|
-
Classifier: Programming Language :: Python :: 3.9
|
|
22
|
-
Classifier: Programming Language :: Python :: 3.10
|
|
23
|
-
Classifier: Programming Language :: Python :: 3.11
|
|
24
|
-
Classifier: Programming Language :: Python :: 3.12
|
|
25
|
-
Classifier: Programming Language :: Python :: 3.13
|
|
26
|
-
Classifier: Topic :: Software Development
|
|
27
|
-
Classifier: Topic :: Scientific/Engineering
|
|
28
|
-
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
|
|
29
|
-
Classifier: Topic :: Scientific/Engineering :: Image Recognition
|
|
30
|
-
Classifier: Operating System :: POSIX :: Linux
|
|
31
|
-
Classifier: Operating System :: MacOS
|
|
32
|
-
Classifier: Operating System :: Microsoft :: Windows
|
|
33
|
-
Requires-Python: >=3.8
|
|
34
|
-
Description-Content-Type: text/markdown
|
|
35
|
-
License-File: LICENSE
|
|
36
|
-
Requires-Dist: numpy
|
|
37
|
-
Requires-Dist: torch
|
|
38
|
-
|
|
39
|
-
<br>
|
|
40
|
-
<a href="https://www.ultralytics.com/" target="_blank"><img src="https://raw.githubusercontent.com/ultralytics/assets/main/logo/Ultralytics_Logotype_Original.svg" width="320" alt="Ultralytics logo"></a>
|
|
41
|
-
|
|
42
|
-
# 🚀 THOP: PyTorch-OpCounter
|
|
43
|
-
|
|
44
|
-
Welcome to the [THOP](https://github.com/ultralytics/thop) repository, your comprehensive solution for profiling PyTorch models by computing the number of Multiply-Accumulate Operations (MACs) and parameters. This tool is essential for deep learning practitioners to evaluate model efficiency and performance.
|
|
45
|
-
|
|
46
|
-
[](https://github.com/ultralytics/thop/actions/workflows/main.yml) <a href="https://discord.com/invite/ultralytics"><img alt="Discord" src="https://img.shields.io/discord/1089800235347353640?logo=discord&logoColor=white&label=Discord&color=blue"></a> <a href="https://community.ultralytics.com/"><img alt="Ultralytics Forums" src="https://img.shields.io/discourse/users?server=https%3A%2F%2Fcommunity.ultralytics.com&logo=discourse&label=Forums&color=blue"></a> <a href="https://reddit.com/r/ultralytics"><img alt="Ultralytics Reddit" src="https://img.shields.io/reddit/subreddit-subscribers/ultralytics?style=flat&logo=reddit&logoColor=white&label=Reddit&color=blue"></a>
|
|
47
|
-
|
|
48
|
-
## 📄 Description
|
|
49
|
-
|
|
50
|
-
THOP offers an intuitive API to profile PyTorch models by calculating the number of MACs and parameters. This functionality is crucial for assessing the computational efficiency and memory footprint of deep learning models.
|
|
51
|
-
|
|
52
|
-
## 📦 Installation
|
|
53
|
-
|
|
54
|
-
You can install THOP via pip:
|
|
55
|
-
|
|
56
|
-
[](https://pypi.org/project/ultralytics-thop/) [](https://www.pepy.tech/projects/ultralytics-thop) [](https://pypi.org/project/ultralytics-thop/)
|
|
57
|
-
|
|
58
|
-
```bash
|
|
59
|
-
pip install ultralytics-thop
|
|
60
|
-
```
|
|
61
|
-
|
|
62
|
-
Alternatively, install the latest version directly from GitHub:
|
|
63
|
-
|
|
64
|
-
```bash
|
|
65
|
-
pip install --upgrade git+https://github.com/ultralytics/thop.git
|
|
66
|
-
```
|
|
67
|
-
|
|
68
|
-
## 🛠 How to Use
|
|
69
|
-
|
|
70
|
-
### Basic Usage
|
|
71
|
-
|
|
72
|
-
To profile a model, you can use the following example:
|
|
73
|
-
|
|
74
|
-
```python
|
|
75
|
-
import torch
|
|
76
|
-
from torchvision.models import resnet50
|
|
77
|
-
|
|
78
|
-
from thop import profile
|
|
79
|
-
|
|
80
|
-
model = resnet50()
|
|
81
|
-
input = torch.randn(1, 3, 224, 224)
|
|
82
|
-
macs, params = profile(model, inputs=(input,))
|
|
83
|
-
```
|
|
84
|
-
|
|
85
|
-
### Define Custom Rules for Third-Party Modules
|
|
86
|
-
|
|
87
|
-
You can define custom rules for unsupported modules:
|
|
88
|
-
|
|
89
|
-
```python
|
|
90
|
-
import torch.nn as nn
|
|
91
|
-
|
|
92
|
-
|
|
93
|
-
class YourModule(nn.Module):
|
|
94
|
-
# your definition
|
|
95
|
-
pass
|
|
96
|
-
|
|
97
|
-
|
|
98
|
-
def count_your_model(model, x, y):
|
|
99
|
-
# your rule here
|
|
100
|
-
pass
|
|
101
|
-
|
|
102
|
-
|
|
103
|
-
input = torch.randn(1, 3, 224, 224)
|
|
104
|
-
macs, params = profile(model, inputs=(input,), custom_ops={YourModule: count_your_model})
|
|
105
|
-
```
|
|
106
|
-
|
|
107
|
-
### Improve Output Readability
|
|
108
|
-
|
|
109
|
-
Use `thop.clever_format` for a more readable output:
|
|
110
|
-
|
|
111
|
-
```python
|
|
112
|
-
from thop import clever_format
|
|
113
|
-
|
|
114
|
-
macs, params = clever_format([macs, params], "%.3f")
|
|
115
|
-
```
|
|
116
|
-
|
|
117
|
-
## 📊 Results of Recent Models
|
|
118
|
-
|
|
119
|
-
The following table presents the parameters and MACs for popular models. These results can be reproduced using the script `benchmark/evaluate_famous_models.py`.
|
|
120
|
-
|
|
121
|
-
<table align="center">
|
|
122
|
-
<tr>
|
|
123
|
-
<td>
|
|
124
|
-
|
|
125
|
-
| Model | Params(M) | MACs(G) |
|
|
126
|
-
| ---------------- | --------- | ------- |
|
|
127
|
-
| alexnet | 61.10 | 0.77 |
|
|
128
|
-
| vgg11 | 132.86 | 7.74 |
|
|
129
|
-
| vgg11_bn | 132.87 | 7.77 |
|
|
130
|
-
| vgg13 | 133.05 | 11.44 |
|
|
131
|
-
| vgg13_bn | 133.05 | 11.49 |
|
|
132
|
-
| vgg16 | 138.36 | 15.61 |
|
|
133
|
-
| vgg16_bn | 138.37 | 15.66 |
|
|
134
|
-
| vgg19 | 143.67 | 19.77 |
|
|
135
|
-
| vgg19_bn | 143.68 | 19.83 |
|
|
136
|
-
| resnet18 | 11.69 | 1.82 |
|
|
137
|
-
| resnet34 | 21.80 | 3.68 |
|
|
138
|
-
| resnet50 | 25.56 | 4.14 |
|
|
139
|
-
| resnet101 | 44.55 | 7.87 |
|
|
140
|
-
| resnet152 | 60.19 | 11.61 |
|
|
141
|
-
| wide_resnet101_2 | 126.89 | 22.84 |
|
|
142
|
-
| wide_resnet50_2 | 68.88 | 11.46 |
|
|
143
|
-
|
|
144
|
-
</td>
|
|
145
|
-
<td>
|
|
146
|
-
|
|
147
|
-
| Model | Params(M) | MACs(G) |
|
|
148
|
-
| ------------------ | --------- | ------- |
|
|
149
|
-
| resnext50_32x4d | 25.03 | 4.29 |
|
|
150
|
-
| resnext101_32x8d | 88.79 | 16.54 |
|
|
151
|
-
| densenet121 | 7.98 | 2.90 |
|
|
152
|
-
| densenet161 | 28.68 | 7.85 |
|
|
153
|
-
| densenet169 | 14.15 | 3.44 |
|
|
154
|
-
| densenet201 | 20.01 | 4.39 |
|
|
155
|
-
| squeezenet1_0 | 1.25 | 0.82 |
|
|
156
|
-
| squeezenet1_1 | 1.24 | 0.35 |
|
|
157
|
-
| mnasnet0_5 | 2.22 | 0.14 |
|
|
158
|
-
| mnasnet0_75 | 3.17 | 0.24 |
|
|
159
|
-
| mnasnet1_0 | 4.38 | 0.34 |
|
|
160
|
-
| mnasnet1_3 | 6.28 | 0.53 |
|
|
161
|
-
| mobilenet_v2 | 3.50 | 0.33 |
|
|
162
|
-
| shufflenet_v2_x0_5 | 1.37 | 0.05 |
|
|
163
|
-
| shufflenet_v2_x1_0 | 2.28 | 0.15 |
|
|
164
|
-
| shufflenet_v2_x1_5 | 3.50 | 0.31 |
|
|
165
|
-
| shufflenet_v2_x2_0 | 7.39 | 0.60 |
|
|
166
|
-
| inception_v3 | 27.16 | 5.75 |
|
|
167
|
-
|
|
168
|
-
</td>
|
|
169
|
-
</tr>
|
|
170
|
-
</table>
|
|
171
|
-
|
|
172
|
-
## 💡 Contribute
|
|
173
|
-
|
|
174
|
-
We welcome community contributions to enhance THOP. Please check our [Contributing Guide](https://docs.ultralytics.com/help/contributing/) for more details. Your feedback and suggestions are highly appreciated!
|
|
175
|
-
|
|
176
|
-
## 📄 License
|
|
177
|
-
|
|
178
|
-
THOP is licensed under the AGPL-3.0 License. For more information, see the [LICENSE](https://github.com/ultralytics/thop/blob/main/LICENSE) file.
|
|
179
|
-
|
|
180
|
-
## 📮 Contact
|
|
181
|
-
|
|
182
|
-
For bugs or feature requests, please open an issue on [GitHub Issues](https://github.com/ultralytics/thop/pulls). Join our community on [Discord](https://discord.com/invite/ultralytics) for discussions and support.
|
|
183
|
-
|
|
184
|
-
<br>
|
|
185
|
-
<div align="center">
|
|
186
|
-
<a href="https://github.com/ultralytics"><img src="https://github.com/ultralytics/assets/raw/main/social/logo-social-github.png" width="3%" alt="Ultralytics GitHub"></a>
|
|
187
|
-
<img src="https://github.com/ultralytics/assets/raw/main/social/logo-transparent.png" width="3%" alt="space">
|
|
188
|
-
<a href="https://www.linkedin.com/company/ultralytics/"><img src="https://github.com/ultralytics/assets/raw/main/social/logo-social-linkedin.png" width="3%" alt="Ultralytics LinkedIn"></a>
|
|
189
|
-
<img src="https://github.com/ultralytics/assets/raw/main/social/logo-transparent.png" width="3%" alt="space">
|
|
190
|
-
<a href="https://twitter.com/ultralytics"><img src="https://github.com/ultralytics/assets/raw/main/social/logo-social-twitter.png" width="3%" alt="Ultralytics Twitter"></a>
|
|
191
|
-
<img src="https://github.com/ultralytics/assets/raw/main/social/logo-transparent.png" width="3%" alt="space">
|
|
192
|
-
<a href="https://youtube.com/ultralytics?sub_confirmation=1"><img src="https://github.com/ultralytics/assets/raw/main/social/logo-social-youtube.png" width="3%" alt="Ultralytics YouTube"></a>
|
|
193
|
-
<img src="https://github.com/ultralytics/assets/raw/main/social/logo-transparent.png" width="3%" alt="space">
|
|
194
|
-
<a href="https://www.tiktok.com/@ultralytics"><img src="https://github.com/ultralytics/assets/raw/main/social/logo-social-tiktok.png" width="3%" alt="Ultralytics TikTok"></a>
|
|
195
|
-
<img src="https://github.com/ultralytics/assets/raw/main/social/logo-transparent.png" width="3%" alt="space">
|
|
196
|
-
<a href="https://ultralytics.com/bilibili"><img src="https://github.com/ultralytics/assets/raw/main/social/logo-social-bilibili.png" width="3%" alt="Ultralytics BiliBili"></a>
|
|
197
|
-
<img src="https://github.com/ultralytics/assets/raw/main/social/logo-transparent.png" width="3%" alt="space">
|
|
198
|
-
<a href="https://discord.com/invite/ultralytics"><img src="https://github.com/ultralytics/assets/raw/main/social/logo-social-discord.png" width="3%" alt="Ultralytics Discord"></a>
|
|
199
|
-
</div>
|
|
@@ -1,13 +0,0 @@
|
|
|
1
|
-
thop/__init__.py,sha256=OSUVoANnWspwoPV8jMIKVV29T_vwfq0SwK-zB63EwMU,219
|
|
2
|
-
thop/fx_profile.py,sha256=ACI2RLyDYBLB7Cru9y2IAx_YCFX_uhYEArDR0np_tFc,8231
|
|
3
|
-
thop/profile.py,sha256=VaCR0K2oWTHp0y-W3SbuA4TdnyxTP6n6VSURltJGzrE,7980
|
|
4
|
-
thop/rnn_hooks.py,sha256=JKZ2eSCvIKvhvCDqM4oWPZjmBkdyJ4R2Q7XSn63lsX0,6503
|
|
5
|
-
thop/utils.py,sha256=IwFJQ1v-SLyhm-313Li535R6fhtomkm8Fem1Kfe6G_U,1484
|
|
6
|
-
thop/vision/__init__.py,sha256=lm6MckFYCPTbqIoX7w0s_daxdjNeBeKW6DXppv1-QUM,70
|
|
7
|
-
thop/vision/basic_hooks.py,sha256=PlhEIJv4I1NOtcK3pomZxbut8oZ-JhHwJ7NtOLO1rbQ,4708
|
|
8
|
-
thop/vision/calc_func.py,sha256=s-_tRmimNBl4PYY_WQHb3C1H4nLy-oEW2kvsZDVjvdI,4362
|
|
9
|
-
ultralytics_thop-2.0.14.dist-info/LICENSE,sha256=hIahDEOTzuHCU5J2nd07LWwkLW7Hko4UFO__ffsvB-8,34523
|
|
10
|
-
ultralytics_thop-2.0.14.dist-info/METADATA,sha256=Kvy_nKqnpl5s7ABwfVTA1SBbeYhovolYQWy3sOlBXr0,9362
|
|
11
|
-
ultralytics_thop-2.0.14.dist-info/WHEEL,sha256=In9FTNxeP60KnTkGw7wk6mJPYd_dQSjEZmXdBdMCI-8,91
|
|
12
|
-
ultralytics_thop-2.0.14.dist-info/top_level.txt,sha256=HQ7D0gSvDJ31CNR-f0EuXNVve05RYBmwyIkHQKiEhU8,5
|
|
13
|
-
ultralytics_thop-2.0.14.dist-info/RECORD,,
|
|
File without changes
|
|
File without changes
|