Fine-tune vision-language models with Unsloth's FastVisionModel and LoRA.
npx skills add https://github.com/atrawog/overthink-plugins --skill vision-atrawog
Please help me install this Agent Skill. Skill: vision Source: https://github.com/atrawog/overthink-plugins/tree/main/overthink-jupyter/skills/vision Command: npx skills add https://github.com/atrawog/overthink-plugins --skill vision-atrawog
Enables efficient fine-tuning of vision-language models using Unsloth's FastVisionModel, reducing the boilerplate and setup time for multi-modal tasks.
Fine-tune a vision-language model by loading a vision model with FastVisionModel, preparing an image+text dataset, and applying vision-specific LoRA during training.
High-intent search queries and answers about installing and using this skill.
You can fine-tune vision-language models using Unsloth's FastVisionModel to apply vision-specific LoRA configuration on image+text datasets, enabling efficient adaptation for OCR, visual QA, and image captioning tasks.
Yes, LoRA configuration supports per-parameter training flags including finetune_vision_layers, finetune_language_layers, finetune_attention_modules, and finetune_mlp_modules to selectively train vision and language components.
Format datasets as image+text pairs with dataset_text_field empty and skip_prepare_dataset set, then use UnslothVisionDataCollator to properly prepare multimodal data for Pixtral and Ministral vision-language training.
Yes, 4-bit loading options are supported when loading vision models with Unsloth's FastVisionModel optimizations, reducing memory requirements during multimodal fine-tuning of vision-language models.
Setting skip_prepare_dataset prevents automated text preprocessing that would corrupt multimodal image+text formatting, ensuring UnslothVisionDataCollator receives the raw structured data needed for vision-language training.
ovachieverLoad, infer, and fine-tune pre-trained transformer models across modalities.
tianhao909Bridge frozen image encoders and large language models for vision-language tasks.
Monjyu1101Curate large multimodal datasets for LLM training with GPU-accelerated deduplication and filtering.
ldzhhxxFine-tune large language models with LoRA and QLoRA for memory efficiency.
