TseEnSun / python-ml-server

python-ml-server

How to run the server?

docker compose build
docker compose up

There is a pre-start that will create the first user by the name and password in the .env file. During the pre-start, the DB migration will also trigger, then the User, Search, and Feedback tables will be created. After docker compose up, the server should run at 0.0.0.0 and 80 ports. http://0.0.0.0/sticker-search/docs will let you access the Swagger UI.

P.S. I found that the ImageSearch class consumes a large size of memory (15G~). But I still need to run it on my machine. Thus, I created a dummy search without any model, just an offset of the hash of the input text.

TODO

• Unit Test

  • API Endpoint
    • feedback
    • report
    • search
    • user
    • login
  • db
    • crud
      • feedback
      • search
      • user
  • services
    • image_search

• Deep dive into ImageSearch which cost too much mempry

  • Memory increasing when generating image_vectors (15G~)
    • Use no_grad() to save memory

• Refind the search result encode method storing in DB.