vision-sft

Produce a validated adapter configuration for supervised fine-tuning of a vision-language model on image-and-text data.

When to use it

Use when adapting a VLM to a visual domain or task, configuring frozen-vision-tower LoRA, or debugging training that does not improve the model.

Give it an image-and-text dataset and the selected VLM base model; it returns a validated configuration covering frozen components, LoRA targets, and image-resolution budgets.