Determined is an open-source machine learning platform that simplifies distributed training, hyperparameter tuning, experiment tracking, and resource management. Works with PyTorch and TensorFlow.
APACHE-2.0 License
Determined is an all-in-one deep learning platform, compatible with PyTorch and TensorFlow.
It takes care of:
The main components of Determined are the Python library, the command line interface (CLI), and the Web UI.
Use the Python library to make your existing PyTorch or Tensorflow code compatible with Determined.
You can do this by organizing your code into one of the class-based APIs:
from determined.pytorch import PyTorchTrial
class YourExperiment(PyTorchTrial):
def __init__(self, context):
...
Or by using just the functions you want, via the Core API:
import determined as det
with det.core.init() as core_context:
...
You can use the CLI to:
det deploy local cluster-up
det deploy aws up
det experiment create gpt.yaml .
Configure everything from distributed training to hyperparameter tuning using YAML files:
resources:
slots_per_trial: 8
priority: 1
hyperparameters:
learning_rate:
type: double
minval: .0001
maxval: 1.0
searcher:
name: adaptive_asha
metric: validation_loss
smaller_is_better: true
Use the Web UI to view loss curves, hyperparameter plots, code and configuration snapshots, model registries, cluster utilization, debugging logs, performance profiling reports, and more.
To install the CLI:
pip install determined
Then use det deploy
to start the Determined cluster locally, or on cloud services like AWS and GCP.
For installation details, visit the the cluster deployment guide for your environment:
Get familiar with Determined by exploring the 30+ examples in the examples folder and the determined-examples repo.
If you need help, want to file a bug report, or just want to keep up-to-date with the latest news about Determined, please join the Determined community!
[email protected]
.