nvidia / runx Goto Github PK
View Code? Open in Web Editor NEWDeep Learning Experiment Management
License: BSD 3-Clause "New" or "Revised" License
Deep Learning Experiment Management
License: BSD 3-Clause "New" or "Revised" License
Here is a simple test script:
main.py
import argparse
parser = argparse.ArgumentParser()
parser.add_argument('--text')
args = parser.parse_args()
print(args.text)
sweep.yml
cmd: 'python main.py'
hparams:
text: ['job1', 'job2']
The simple example in the readme is not working.
I got this error:
$ python -m runx.runx sweep.yml
...
raise('can\'t find file ./.runx or ~/.config/runx.yml config files')
not a direct implementation for tensorboard add_scalars not add_scalar
How can I use runx without farm settings? Or how can I set the farm in ubuntu? I am confused about the 'submit_job' refer.
@ajtao Can I use this with custom deep learning models in PyTorch and Tensorflow+Keras 2.0?
Thanks for your great work. Would you support Logger Objects in the future?
When i run this command: python -m runx.runx sweep.yml, there is a problem below, why?
"""Copying codebase to ./logs/sweep/witty-fennec_2020.05.27_21.05 ...
Submitting job sweep_witty-fennec_2020.05.27_21.05
/bin/sh: 1: submit_job: not found """
Such as
LOGROOT: ~/.cache/runx
can be expanded as $HOME/.cache/runx
? Thanks.
Hey,
first thanks for this nice, light-weight tool! Very helpful.
Just one thing, I can't get sumx running:
For python -m runx.sumx config_simple
, I get the following error:
No valid experiments found for /Users/svengiegerich/runx/config_simple
using version 0.0.10
.
However, the path is correct and there are two successful runs (folders) in it; each subfolder contains a metrics.csv
looking like this,
start,start/step,0,timestamp,1621378323.1818151
val,loss,0.013856839059957424,epoch,1,timestamp,1621378337.119943
val,loss,0.00520349506242475,epoch,2,timestamp,1621378344.595231
metrics are added by the following code lines,
metrics_val = {'loss': epoch_loss}
logx.metric(phase='val', metrics=metrics_val, epoch=epoch_i + 1)
Every other logging works smoothly, e.g. logx.add_scalar()
for tensorboard.
-> Any idea what's wrong here?
My .runx
,
LOGROOT: /Users/svengiegerich/runx
CODE_IGNORE_PATTERNS: '*.git,data/raw*,.*,results*'
FARM: bigfarm
# Farm resource needs
bigfarm:
SUBMIT_CMD: 'submit_job'
RESOURCES:
image: mydocker-image-big:1.0
gpu: 8
cpu: 64
mem: 450
and the config_simple.yml
,
CMD: 'python train.py'
HPARAMS: [
{
logdir: LOGDIR,
epochs: [1,2],
RUNX.TAG: 'transformer',
arch: 'transformer',
}
]
Is runx going to support this feature given that major conferences such as NeurIPS now requires submitting mean+std for the experiments?
In runx.logx
line 110
if not os.path.isdir(self.logdir):
change to this:
if not os.path.isdir(self.logdir) and self.rank0:
if not, it will make a lot of dirs
Hi,
could I ask for an example to configure .runx, if I would run with just one GPU? I don't have farm at all. Thanks a lot.
pip install runx installs v.0.0.11.
This breaks the log functionality somehow, and I'm seeing:
Copying codebase to ../logs/sweep/spectral-okapi_2021.07.12_15.16 ...
Running job sweep_spectral-okapi_2021.07.12_15.16
cd ../logs/sweep/spectral-okapi_2021.07.12_15.16/code; PYTHONPATH= python testrunx.py --lr 0.01 --solver first --logdir ..//logs
/bin/sh: 1: cd: can't cd to ../logs/sweep/spectral-okapi_2021.07.12_15.16/code
python: can't open file 'testrunx.py': [Errno 2] No such file or directory
Copying codebase to ../logs/sweep/teal-trogon_2021.07.12_15.16 ...
Running job sweep_teal-trogon_2021.07.12_15.16
cd ../logs/sweep/teal-trogon_2021.07.12_15.16/code; PYTHONPATH= python testrunx.py --lr 0.01 --solver second --logdir ..//logs
/bin/sh: 1: cd: can't cd to ../logs/sweep/teal-trogon_2021.07.12_15.16/code
python: can't open file 'testrunx.py': [Errno 2] No such file or directory
Copying codebase to ../logs/sweep/garrulous-wombat_2021.07.12_15.16 ...
Running job sweep_garrulous-wombat_2021.07.12_15.16
cd ../logs/sweep/garrulous-wombat_2021.07.12_15.16/code; PYTHONPATH= python testrunx.py --lr 0.02 --solver first --logdir ..//logs
/bin/sh: 1: cd: can't cd to ../logs/sweep/garrulous-wombat_2021.07.12_15.16/code
python: can't open file 'testrunx.py': [Errno 2] No such file or directory
Copying codebase to ../logs/sweep/lavender-prawn_2021.07.12_15.16 ...
Running job sweep_lavender-prawn_2021.07.12_15.16
cd ../logs/sweep/lavender-prawn_2021.07.12_15.16/code; PYTHONPATH= python testrunx.py --lr 0.02 --solver second --logdir ..//logs
/bin/sh: 1: cd: can't cd to ../logs/sweep/lavender-prawn_2021.07.12_15.16/code
python: can't open file 'testrunx.py': [Errno 2] No such file or directory
Looking at ../logs, I'm seeing recursive paths like this one:
../logs/sweep/logs/sweep/logs/sweep/logs/sweep/complex-mongrel_2021.07.12_15.06/submit_cmd.sh
Here is my python script (testrunx.py)
import sys
print(sys.argv)
And here is my sweep.yml:
CMD: 'python testrunx.py'
HPARAMS:
lr: [0.01, 0.02]
solver: ['first', 'second']
logdir: '..//logs'
Installing version 0.0.5 fixes the problem.
Hi:
I've been used runx for months with your origin version, but the latest NGC version seems doesn't work well with farm.
A declarative, efficient, and flexible JavaScript library for building user interfaces.
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. ๐๐๐
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google โค๏ธ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.