nunet-py is a Python package for running jobs on NuNet!
Table of Contents
nunet-py is a Python client for submitting jobs to NuNet. It is intended to be a
direct interface to the DMS.
This is considered an alpha project. The NuNet network is in active development, so it is possible this package will break as changes are made to the DMS. Currently, everything happens on the Cardano preprod testnet.
Right now, the setup is fairly involved unless you're already a NuNet alpha tester. I highly recommend joining the NuNet Discord for assistance. I (Elder Millenial) tend to hang out there as well as a lot of other highly knowledgeable people, and the team is very responsive.
The basic steps to setup your computer to use nunet-py are as follows:
- Join the NuNet Alpha Testing Program. Follow the directions to become both a resource provider and service provider.
- Create a Blockfrost account, and create a project id for the predprod testnet.
- (Optional) Save
sample.envas.env, and replace the values. For the seed phrase, use the seed phrase used when setting up your system for the Alpha Testing Program. Also put in your Blockfrost project id.
This example will submit one of the example ML jobs provided for the NuNet Alpha Testers program. This will break up the entire example into sections, and the complete example can be found at the end and in the examples.
During development of this tool a bug was found in pycardano, a dependency of this project. The bug was fixed and merged into the project, but the fix has not been released so the project cannot be pushed to PyPI. The best way to install and get started is to install directly from this repo:
pip install git+https://github.com/theeldermillenial/nunet-pyAssuming you made the .env with your preprod wallet seed phrase as described in under
Getting Started, we start by loading the the environment file,
loading the seed phrase from the environment, and initializing the adapter with the seed
phrase. The reason why we take this approach is that you should never put your seed
phrase into code, and even though this is a preprod wallet, we want to use best
practices.
import os
from dotenv import load_dotenv
from nunet import NuNetAdapter
# Load the environment file
load_dotenv()
# Get your seed phrase from the environment
seed = os.environ["SEED"]
# Create a NuNet adapter
adapter = NuNetAdapter(seed)Now we can get a list of peers and print off the information we get about them.
import pprint
peer_list = adapter.peer_list()
pprint.pprint(peer_list.model_dump(), indent=2)There are a number of details needed to configure your job to ensure your job request
is matched with an appropriate resource provider. However, this can also be burdensome
so some convenience objects have been created to make it easier to get started. In this
example, we just use a very low resource request (CONSTRAINTS_LOW).
from nunet import CONSTRAINTS_LOW
pprint.pprint(CONSTRAINTS_LOW.model_dump(), indent=2)Next we need to configure some additional information about our job, including whether we need a GPU or if a CPU will suffice. In addition to that, we need to indicate what piece of code we want to run and configure the package dependencies. For this, we just use the simple ml-test from the NuNet Test Program examples, which has no package requirements.
from nunet import JobParams, ImageId, MachineType
params = JobParams(
machine_type=MachineType.GPU,
image_id=ImageId.ML_ON_CPU_REGISTRY,
model_url="https://gitlab.com/nunet/ml-on-gpu/ml-on-cpu-service/-/raw/develop/examples/cpu-ml-test-scikit-learn.py",
packages=[],
)Next we need to compile this data together and indicate the maximum NTX we are willing to pay and our address.
from nunet import JobRequest, ServiceType
job_config = JobRequest(
address_user=adapter.address.encode(),
max_ntx=10,
service_type=ServiceType.CPU,
params=params,
constraints=CONSTRAINTS_LOW,
)Finally, we need to request a service provider, pay the contract to execute the job, then submit the job to the network and listen for the output.
job_request = adapter.request_service(job_config=job_config)
txid = adapter.pay(job_request)
for log in adapter.job(txid):
print(log)When you execute the last piece of code, it will likely hang a bit as the order is submitted, then wait for the job to begin executed. If you don't see activity after a few minutes, use ctrl+c to exit the code. You will see logs until the code completes, and the code will exit when it completes.
Once this happens, you've successfully run your first job on NuNet!
import os
import pprint
from dotenv import load_dotenv
from nunet import (
CONSTRAINTS_LOW,
ImageId,
JobParams,
JobRequest,
MachineType,
NuNetAdapter,
ServiceType,
)
# Load the environment file
load_dotenv()
# Get your seed phrase from the environment
seed = os.environ["SEED"]
# Create a NuNet adapter
adapter = NuNetAdapter(seed)
# Get a list of peers
print("Peers:")
peer_list = adapter.peer_list()
pprint.pprint(peer_list.model_dump(), indent=2)
print()
print("CONSTRAINTS_LOW:")
pprint.pprint(CONSTRAINTS_LOW.model_dump(), indent=2)
print()
params = JobParams(
machine_type=MachineType.GPU,
image_id=ImageId.ML_ON_CPU_REGISTRY,
model_url="https://gitlab.com/nunet/ml-on-gpu/ml-on-cpu-service/-/raw/develop/examples/cpu-ml-test-scikit-learn.py",
packages=[],
)
job_request: JobRequest = JobRequest(
address_user=adapter.address.encode(),
max_ntx=10,
service_type=ServiceType.CPU,
params=params,
constraints=CONSTRAINTS_LOW,
)
job_request = adapter.request_service(job_request=job_request)
txid = adapter.pay(job_request)
for msg_type, msg in adapter.job(txid):
print(msg, flush=True)I can always use volunteers to take on specific chunks of the project. I work on this in my free time, along with some other Cardano projects. You can help by reaching out on Twitter or Discord. Alternatively, sending tips is also helpful to cover the costs of production. Tips can be sent to:
addr1q9hw8fuex09vr3rqwtn4fzh9qxjlzjzh8aww684ln0rv0cfu3f0de6qkmh7c7yysfz808978wwe6ll30wu8l3cgvgdjqa7egnl