Skip to content
Kitbay

finetuning

MITUpdated on Nov 10, 20261,636GitHub starsIncludes scripts

A Microsoft sub-skill of microsoft-foundry for fine-tuning models in Microsoft Foundry with supervised fine-tuning, direct preference optimisation or reinforcement fine-tuning with graders.

What the skill teaches

A Microsoft sub-skill of microsoft-foundry for fine-tuning models in Microsoft Foundry with supervised fine-tuning, direct preference optimisation or reinforcement fine-tuning with graders. It covers preparing and validating training data, generating distillation data, scoring dataset quality, uploading large files, submitting and monitoring training jobs, calibrating graders, deploying and evaluating the tuned model and cleaning up afterwards. Bundled Python scripts carry out each of these steps.

Capabilities

  • Prepare and validate fine-tuning datasets
  • Submit and monitor SFT, DPO or RFT training jobs
  • Calibrate graders for reinforcement fine-tuning
  • Deploy and evaluate fine-tuned models