nimblebox 0.9.16

ML Inference 🥶

🏖️ Nbox

nbox is NimbleBox.ai's official SDK.

The entire purpose of this package is to make using ML 🥶.

pip install nbox

🔥 Usage

nbox provides first class support API for all NimbleBox.ai infrastructure (NBX-Build, Jobs, Deploy) and services (NBX-Workspaces) components. Write jobs using nbox.Operators:

from nbox import Operator
from nbox.nbxlib.ops import Magic

# define a class object
weekly_trainer: Operator = Magic()

# call your operators
weekly_trainer(
  pass_values = "directly",
)

# confident? deploy it to your cloud
weekly_trainer.deploy(
  job_id_or_name = "magic_jobs",
  schedule = Schedule(4, 30, ['fri']) # schedule like humans
)

Deploy your machine learning or statistical models:

from nbox import Model
from transformers import AutoModelForSequenceClassification, AutoTokenizer

# define your pre and post processing functions
def pre(x: Dict):
  return AutoTokenizer(**x)

# load your classifier with functions
model = AutoModelForSequenceClassification.from_pretrained("distill-bert")
classifier = Model(model, pre = pre)

# call your model
classifier(f"Is this a good picture?")

# get full control on exporting it
spec = classifier.torch_to_onnx(
  TorchToOnnx(...)
)

# confident? deploy it your cloud
url, key = classifier.deploy(
  spec, deployment_id_or_name = "classification"
)

# use it anywhere
pred = requests.post(
  url,
  json = {
    "text": f"Is this a good picture?"
  },
  header = {"Authorization": f"Bearer {key}"}
).json()

🧩 License

The code in thist repo is licensed as Apache License 2.0. Please check for individual repositories for licenses.