Skip to content

Get command/title embeddings locally #15

@forrestbao

Description

@forrestbao

We do not wanna package a sentence-BERT model into the SmartCommand plug-in. Nor do we wanna package the embeddings of commands into it too. Too many floating point numbers. They will make the plug-in too bulky. This is the last resort.

Three options left:

  1. Download Sentence-BERT model (or whatever Transformer-based model) and run a Python script to embed locally.
  2. Sending strings to a cloud server to embed. The cloud server can be ours or provided by a third party, such as OpenAI's Plugin API.
  3. Pre-embed the commands (and their labels/titles) and store in our cloud server. When SmartCommand is first installed, download such embeddings. We can expand the plug-ins supported gradually.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions