watchmen for GPU scheduling
Project description
watchmen
A simple and easy-to-use toolkit for GPU scheduling.
Dependencies
- Python >= 3.6
- requests >= 2.24.0
- pydantic >= 1.7.1
- gpustat >= 0.6.0
- flask >= 1.1.2
- apscheduler >= 3.6.3
Installation
- Install dependencies.
$ pip install -r requirements.txt
- Install watchmen.
Install from source code:
$ pip install -e .
Or you can install the stable version package from pypi.
$ pip install gpu-watchmen -i https://pypi.org/simple
Quick Start
- Start the server
The default port of the server is 62333
$ python -m watchmen.server
If you want the server to be running backend, try:
$ nohup python -m watchmen.server &
There are some configurations for the server
usage: server.py [-h] [--host HOST] [--port PORT]
[--queue_timeout QUEUE_TIMEOUT]
[--request_interval REQUEST_INTERVAL]
[--status_queue_keep_time STATUS_QUEUE_KEEP_TIME]
optional arguments:
-h, --help show this help message and exit
--host HOST host address for api server
--port PORT port for api server
--queue_timeout QUEUE_TIMEOUT
timeout for queue waiting (seconds)
--request_interval REQUEST_INTERVAL
interval for gpu status requesting (seconds)
--status_queue_keep_time STATUS_QUEUE_KEEP_TIME
hours for keeping the client status
- Modify the source code in your project:
client = Client(id="short description of this running", gpus=[1],
server_host="127.0.0.1", server_port=62333)
client.wait()
When the program goes on after client.wait()
, you are in the queue.
You can check an example in example/single_card_mnist.py
- Check the queue in browser.
Open the following link to your browser: http://<server ip address>:<server port>
, for example: http://192.168.126.143:62333
.
And you can get a result like the demo below. Please be aware that the page is not going to change dynamically, so you can refresh the page manually to check the latest status.
UPDATE
- v0.1.1: fix html package data
TODO
- add reminders
- add webui html support
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for gpu_watchmen-0.1.1-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | d084c0ad47a61e9e47ef41e10fc04cd9cb27cf4b2045226b15843e218437a551 |
|
MD5 | dc872a49f5e7d8c017292c60c52f8ace |
|
BLAKE2b-256 | 014ffcc3a723a104c05cdaed96e86c8e3f5ef970366aa35f5131dda54ba0f965 |