JLclient

Interacting with Jarvislabs.ai for creating GPU/CPU powered instances on top of A100, A6000, RTX 5000 and RTX6000Ada.

Installation

pip install git+https://github.com/jarvislabsai/jlclient.git

Imports and configure

from jlclient import jarvisclient

from jlclient.jarvisclient import *



jarvisclient.token = '**************************duWRbO68IiMTkQKWi48'

Generate a token from here.

Managing GPU/CPU powered instances on Jarvislabs.ai

Create

Parameter	Type	Description/Values	Default Value
instance_type	str	Choose between GPU or CPU.	GPU
num_gpus / num_cpus	int	Choose between 1 to 8 for GPU instance.	1
gpu_type	str	Choose from A100, RTX6000Ada, A5000, A6000, RTX5000.	RTX5000
template	str	Use `User.get_templates()` to get all templates.	pytorch
script_id	str	Use `User.get_scripts()` to get all script ids and pass it.	None
is_reserved	bool	True refers to an on-demand instance. False refers to a spot instance.	True
duration	str	Choose hour, week, and month. The pricing changes based on the duration..	hour
http_ports	str	As per your requirement, you can specify the ports.	None
storage	int	Choose between 20GB to 2TB.	20

# CPU Instance Example

instance: Instance = Instance.create('CPU',
                            num_cpus=1,
                            storage=25,
                            template='pytorch',
                            name='cpu instance')


# GPU Instance Example

instance: Instance = Instance.create('GPU',
                            gpu_type='RTX6000Ada',
                            num_gpus=1,
                            storage=50,
                            template='pytorch',
                            name='gpu instance')

This should return the Instance object, which includes the following attributes

gpu_type
num_gpus
num_cpus
storage
name
machine_id
script_id
is_reserved
duration
script_args
http_ports
template
url
endpoints
ssh_str
status

If the Instance object isn't returned, an error dictionary will be provided.

Note: Please contact us if you encounter any errors while launching the instance.

Pause

instance.pause()

Pause existing Instance

    # Get the running Instance
    instance: Instance = User.get_instance(instance_id=12345)

    instance.pause()

You can call pause() on any Instance object.

Resume

#Example 1:

instance.resume()



#Example 2:

instance.resume(num_gpus=1,
                gpu_type='RTX5000',
                storage=100)


#Switching GPU to CPU Instance, pass the num_cpus parameter

instance.resume(num_cpus=1,
                storage=25)

#Switching CPU to GPU Instance, pass the num_gpus & gpu_type

instance.resume(gpu_type='RTX6000Ada',
                num_gpus=1,
                storage=25)

Resume existing Instance

   # Get the paused Instance
   instance: Instance = User.get_instance(instance_id=12345)

   # Resuming the old instance
   new_instance = instance.resume()

You can modify an existing instance by changing the below resume parameters.

num_gpus
gpu_type
storage

or just call resume to start with the same configuration.

Destroy

instance.destroy()

Destroy the existing Instance

    # Get the paused or running Instance
    instance: Instance = User.get_instance(instance_id=12345)

    instance.destroy()

Invoking the destroy method on any instance object will permanently delete the instance and it cannot be retrieve.

User management.

The User class comes with the below key functionalities.

User.get_templates() : Returns the list of templates.
User.get_instances() : Returns a list of Instance objects representing instances in your account.
User.get_instance() : Returns the Instance object.
User.get_balance() : Return the balance of the user.
User.get_scripts() : Return the list of scripts of the user

Get the scripts of the User

scripts = User.get_scripts()

'''

scripts = [{'script_id': 123, 'script_name': 'script1.sh'},
           {'script_id': 124, 'script_name': 'script2.sh'}]

'''

Invoking User.get_scripts() method to retrieve the scripts associated with the User.

Issues/Feature request

Do you like to see any new features, we are all ears. You can drop us an email to [email protected] or chat with us for any new features or issues.

License

This project is licensed under the terms of the MIT license.

atom-101 / jlclient Goto Github PK

jlclient's Introduction