sft

Supervised fine-tuning (SFT) with the Tinker API requires coupling your dataset with the appropriate renderer, selecting a learning rate, and applying the guided training configuration to execute the run from model selection to export.

Instruction tuning adapts models to follow specific commands, while chat fine-tuning optimizes for multi-turn conversational dynamics; both supervised learning tasks are supported by configuring the appropriate renderer and dataset.

Preparing instruction data for LLM fine-tuning involves formatting your dataset to match a compatible renderer, ensuring the input aligns with the Tinker API's training configuration requirements for the selected model base.

Yes, the Tinker API supports broader machine learning workflows and supervised learning tasks across different datasets and model bases, extending beyond just chat fine-tuning to general instruction tuning.

Configuring the learning rate for supervised fine-tuning is handled through the guided setup, which helps match the training configuration to your specific model base and dataset to ensure stable learning.