What problem does it solve? Building image-understanding features like visual question answering, image captioning, and multi-turn image chat typically requires expensive proprietary APIs. This Skill provides complete guidance for running the open-source LLaVA vision-language model locally, from installation and inference to custom fine-tuning. ## Core Features & Use Cases - Vision-Language Inference: Load LLaVA models (7B, 13B, 34B) with transformers and torch to answer questions about images, generate captions, and hold multi-turn image conversations. - Flexible Deployment: Run via Python API, CLI, or a Gradio web interface, with 4-bit and 8-bit quantization options to reduce VRAM requirements. - Custom Fine-Tuning: Train on your own instruction data using full fine-tuning with DeepSpeed or memory-efficient LoRA, following the two-stage pretrain/finetune pipeline. - Use Case: Build a document-understanding chatbot that answers questions about scanned forms by loading llava-v1.5-7b with 4-bit quantization on a single GPU. ## Quick Start Ask the AI to load the liuhaotian/llava-v1.5-7b model and describe what is in a given image file.
