Qwen2.5-VL 7B image-and-text instruction tuning
A LoRA training configuration for a Qwen vision-language model using image and conversation pairs.
Published by Axolotl
How to use it
The recipe loads Qwen2.5-VL-7B-Instruct with an automatic processor and uses HuggingFaceH4's LLaVA instruction mix. It demonstrates the multimodal preprocessing path and adapter-based training.
Check that your sample image resolves correctly and that the formatted conversation places its image token where intended. Image sizes affect memory and throughput, so measure a small representative run first.
- Framework
- Axolotl
- Dataset
- HuggingFaceH4/llava-instruct-mix-vsft
- Hardware
- Minimum VRAM unspecified.
- Access
- Public publisher configuration; model and dataset terms apply.