nnprof
Introduction
nnprof is a profile tool for pytorch neural networks.
Features
- multi profile mode: nnprof support 4 profile mode: Layer level, Operation level, Mixed level, Layer Tree level. Please check below for detail usage.
- time and memory profile: nnprof support both time and memory profile now. But since memory profile is first supported in pytorch 1.6, please use torch version >= 1.6 for memory profile.
- support sorted by given key and show profile percent: user could print table with percentage and sorted profile info using a given key, which is really helpful for optimiziing neural network.
Requirements
- Python >= 3.6
- PyTorch
- Numpy
Get Started
install nnprof
- pip install:
pip install nnprof
- from source:
python -m pip install 'git+https://github.com/FateScript/nnprof.git'
# or install after clone this repo
git clone https://github.com/FateScript/nnprof.git
pip install -e nnprof
use nnprf
from nnprof import profile, ProfileMode
import torch
import torchvision
model = torchvision.models.alexnet(pretrained=False)
x = torch.rand([1, 3, 224, 224])
# mode could be anyone in LAYER, OP, MIXED, LAYER_TREE
mode = ProfileMode.LAYER
with profile(model, mode=mode) as prof:
y = model(x)
print(prof.table(average=False, sorted_by="cpu_time"))
# table could be sorted by presented header.
Part of presented table looks like table below, Note that they are sorted by cpu_time.
╒══════════════════════╤═══════════════════╤═══════════════════╤════════╕
│ name │ self_cpu_time │ cpu_time │ hits │
╞══════════════════════╪═══════════════════╪═══════════════════╪════════╡
│ AlexNet.features.0 │ 19.114ms (34.77%) │ 76.383ms (45.65%) │ 1 │
├──────────────────────┼───────────────────┼───────────────────┼────────┤
│ AlexNet.features.3 │ 5.148ms (9.37%) │ 20.576ms (12.30%) │ 1 │
├──────────────────────┼───────────────────┼───────────────────┼────────┤
│ AlexNet.features.8 │ 4.839ms (8.80%) │ 19.336ms (11.56%) │ 1 │
├──────────────────────┼───────────────────┼───────────────────┼────────┤
│ AlexNet.features.6 │ 4.162ms (7.57%) │ 16.632ms (9.94%) │ 1 │
├──────────────────────┼───────────────────┼───────────────────┼────────┤
│ AlexNet.features.10 │ 2.705ms (4.92%) │ 10.713ms (6.40%) │ 1 │
├──────────────────────┼───────────────────┼───────────────────┼────────┤
You are welcomed to try diffierent profile mode and more table format.
Contribution
Any issues and pull requests are welcomed.
Acknowledgement
Some thoughts of nnprof are inspired by torchprof and torch.autograd.profile . Many thanks to the authors.
Metadata
Release files for nnprof 0.1.1
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| nnprof-0.1.1.tar.gz | 7.2 kB | Without provenance | Details for nnprof-0.1.1.tar.gz |
Release files / nnprof-0.1.1.tar.gz
| Download URL | nnprof-0.1.1.tar.gz |
|---|---|
| Size | 7.2 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
dca701810c075ca01cd553d4ccff6a9cea62a1c7d9371933ec4ccaf1ed3eb85b
|
|
BLAKE2b-256 checksum How to use checksums |
9db3a978d2b1185c6d26decb62d13d04903f23e6c25adecfd5b20ff78e96fd17
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/3.3.0 pkginfo/1.6.1 requests/2.23.0 setuptools/41.2.0 requests-toolbelt/0.9.1 tqdm/4.36.1 CPython/3.6.8
|
Security for nnprof 0.1.1
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.