How to: Fine-tune Model with Foundry

:wave: Hello Community!

:rocket: Introducing Model Fine-tuning: Allows to tailor the model to custom use cases!

:dart: Problems Addressed:

The Model Fine tuning capability addresses the limitation of using foundational models for detecting objects beyond the scope of their existing feature set.

:hammer_and_wrench: How to Use:


  1. Go to Model , Experiment and click Create then select Model experiment.

  2. Import data rows with labels to the experiment for training.

  3. Configure the model experiment by giving it a name, ontology containing custom feature to train and define splits.

Then click Submit which initiates the “Send to Experiment” task.

  1. Once the model run gets populated in the newly created experiment, click on
    “Fine-tune model” button

  2. Choose a base model(YOLO is only model supported right now)

  3. Configure training parameters and kick off fine-tuning job.

  4. Use fine-tuned model(found in Model–>Custom) just as you would use a Foundry model for inference and auto-labeling

Model Fine-tuning empowers you to train models that are tailored precisely to your needs for using custom features.

Please try Model Fine-tuning feature and share your valuable feedback.

:green_book: For detailed steps: Fine-tune model

Cheers! :tada:


This is awesome, my team is going to try it out this week.

Will this type of fine tuning be made available for other data types / foundation models?

We do have a pipeline to get text added to the fine tuning capability to Foundry, I can’t provide a timeline yet, I will revise this message when I have more info.
What would be your requirements?

1 Like