A catalog of recent changes in the codebase.
- A continuous score for each request to enable ranking of requests in accordance with how bad the model thinks it is.
- If the amount of data that we are training from is exactly 16,000 requests - training will crash and no model will be mounted.
- Some 500 responses were still occasionally returned.
- 500 errors that are returned by the ingestion service.
- Explainability has been improved, and is now more readable.
- First distributable cut.