A Python package to manage delphai machine learning operations.
Project description
delphai-ml-utils
Installation
pip install delphai-ml-utils
Usage
- Upload to Azure Blob
The delphai-hybrid
cluster allows model training with gpu. This feature allows uploading the trained model from inside the cluster to Azure blobs.
This works by adding a config file to your project config/ml-config.yml
.
With this yaml file you can configure to which storage account you want to upload your trained model.
cluster: delphai-hybrid
training_dir: model-gpu
model_name: test-model
dest:
storage_account_secret: azure-storage/connection-string
training_dir
: is the output directory of your trained model (model directory)
model_name
: Name your model and with it name the new created azure container to save the model into it (Note if the container name already exists will throw an error)
storage_account_secret
: Here add the kubernetes secret name that contains the connection string to the storage account. example azure-storage/conenction-string
How to use with python:
from ml_utils import upload
# Train Model
model.train_model(train_df, use_cuda=True)
# Upload to Azure blob with delphai-ml-utils
upload.upload_to_azure_blob()
- register gpu
If your model will serve with gpu, we need to monitor every request that comes to the model so kubernetes removes the deployment once it's not being used.
NOTE: if you don't register the model for gpu, it will be destroyed once deployed
from ml_utils import upload, gpu
# serving function
def foo():
gpu.monitor()
#
# Some Logic
#
return bar
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 delphai_ml_utils-1.0.5-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 4fb3fa2369fae37703eef0c735a4b744f7157729227ed997c72983e0a7a506dc |
|
MD5 | a03097567feb145cb5615b15f45a9006 |
|
BLAKE2b-256 | 3a3f975758a78709f3bbf081c4c3f4c127b8496533bbbf34693e25b274db6b9e |