Skip to content

The FSDL App

Tech Stack:

Tutorials:

Experiment and artefact Tracking

  • Weights and Biases stores
  • PyTorch lighting checkpoits that can be used to restart model training if interrupted. This enables used of chepaer preemptible cloud instance. These are essentially excess capacity that a cloud provider has which can be utilised, with the caveat that resources can be pre-empted (allocated away at the cloud provider's will)
  • deployed model files

Model Handoff to production

  • Pytorch lightning models are compiled down to a dialect of Torch called torchscript