flor
flor copied to clipboard
🌻 FlorFlow: Flor, now with Dataflow

Large files the developer doesn't want autosaved on every run should be gitignored
- Resume training on crashes. - Write valid logs when possible, even during failures (exception handling). - Reconstruct logs on sudden failures.
Statements written to stdout, and any statements with **expected visibility** like tensorboard losses should be intercepted (hard or soft link) for post-hoc retrieval.
The CLI flags configure the run and must be stored in `.replay.json`