sbt-idp/cope2n-ai-fi/TODO.md
2023-12-13 16:01:31 +07:00

1.5 KiB

Bring abs path to relative

  • save_dir Kie_Invoice_AP/prediction_sap.py:18
  • detector Kie_Invoice_AP/AnyKey_Value/ocr-engine/settings.yml fixed
  • rotator_version Kie_Invoice_AP/AnyKey_Value/ocr-engine/settings.yml fixed
  • cfg Kie_Invoice_AP/prediction_fi.py
  • weight Kie_Invoice_AP/prediction_fi.py
  • save_dir Kie_Invoice_AP/prediction_fi.py:18

Bring abs path to .env

  • CELERY_BROKER:
  • SAP_KIE_MODEL: Kie_Invoice_AP/prediction_sap.py:20 NEED_REFACTOR
  • FI_KIE_MODEL: Kie_Invoice_AP/prediction_fi.py:20 NEED_REFACTOR

Possible logic confict

Refactor

  • Each model should be loaded in a docker container and serve as a service
  • Some files (weights, ...) should be mounted in container in a format for endurability
  • Kie_Invoice_AP/prediction_fi.py and Kie_Invoice_AP/prediction_fi.py should be merged into a single file as it shared resources with different logic
  • Kie_Invoice_AP/prediction.py seems to be the base function, this should act as a proxy which import all other predict_{anything else} functions
  • There should be a unique folder to keep all models with different versions then mount as /models in container. Currently, fi is loading from /models/Kie_invoice_fi while sap is loading from Kie_Invoice_AP/AnyKey_Value/experiments/key_value_understanding-20231003-171748. Another model weight is at sdsvtd/hub for unknown reason
  • Env variables should have its description in README