MonsterTuner now supports checkpointing, allowing users to terminate a fine-tuning job before completion and still access the last saved checkpoint. Checkpoints are auto-saved periodically during training, providing a way to validate experiments in progress.

A Colab notebook is available for deploying the TinyLlama 1.1B model on MonsterAPI. This model, trained on 1T tokens, offers cost-effective chat application capabilities.

📘

Support for JSON Payload