hjerpe's picture
F006/F008: serve Qwen models + model switcher (vanilla-first)
656f91e verified
Raw
History Blame Contribute Delete
316 Bytes
"""Backend adapters for the training pipeline ports (see training/ports.py).
Each adapter implements a port for one provider. The compute backend itself is
selected by entrypoint, not by code: ``modal run training/modal_app.py`` today;
a k8s Job manifest or Vertex CustomJob launcher would be new modules here.
"""