ʎzy is a platform for a hybrid execution of ML workflows that transparently integrates local and remote runtimes with the following properties:
- Python-native SDK
- Automatic env (pip/conda) sync
- K8s-native runtime
- Resources allocation on-demand
- Env-independent results storage
ʎzy allows running any python functions on a cluster by annotating them with @op
decorator:
@op(gpu_count=1, gpu_type=GpuType.V100.name)
def train(data_set: Bunch) -> CatBoostClassifier:
cb_model = CatBoostClassifier(iterations=1000, task_type="GPU", devices='0:1', train_dir='/tmp/catboost')
cb_model.fit(data_set.data, data_set.target, verbose=True)
return cb_model
# local python function call
model = train(data_set)
# remote call on a cluster
lzy = Lzy()
with lzy.workflow("training"):
model = train(data_set)
Please read the tutorial for details.
Check out our key concepts and architecture intro.
Join our chat on telegram!
Development guide.
Deployment guide.