Fine-Tuning Worker

The fine_tuning_worker serves as a specialized worker responsible for executing tasks related to fine-tuning. Typically deployed on machines equipped with GPUs, these workers enable faster model retraining. It's important to note that, currently, our system is compatible exclusively with Nvidia GPUs. This ensures optimal performance and efficiency in the fine-tuning process.

After writing your plugin (fine-tuning method), you can start the Embedding Studio worker - fine_tuning_worker. To do this, you need to build an image with your plugin and start the worker:

Rebuild the fine_tuning_worker image with your plugin and start it:

docker compose build --no-cache fine_tuning_worker

and

docker compose up -d fine_tuning_worker

It will pick up your plugin and wait for fine-tuning tasks.

Tip

To avoid rebuilding the image every time your change it, you can mount the plugins directory inside the container. To do this, add a volume section to the docker-compose.yml:

services:
  ...
  fine_tuning_worker:
    ...
    volumes:
      - ./plugins:/embedding_studio/plugins
  ...

and start the container:

docker compose up -d fine_tuning_worker