Skip to main content

Fine-tuning

The Knowledgator MCP server provides tools to launch, inspect, list, and control custom model fine-tuning jobs directly from AI coding environments.


Available Fine-Tuning Tools​

  • Start Training Job — Launch a fine-tuning job for a base model or adapter (start_training_job).
  • Get Training Status — Retrieve metrics, logs, and current status of a training run (get_training_status).
  • List Trainings — Browse all training runs with status, model, and dataset filters (list_trainings).
  • Stop Training Job — Safely stop or cancel an active training job (stop_training_job).

Example Workflow​

Agents can automate model customization end-to-end:

  1. Launch: Initiate fine-tuning via start_training_job with chosen hyperparameters (learning_rate_mult, num_of_epochs, batch).
  2. Monitor: Periodically check progress, loss, and training milestones using get_training_status.
  3. Interrupt: Call stop_training_job if convergence targets are satisfied or loss diverges.