Lavender: Diffusion Instruction Tuning Boosting SoTA vision-language model with Stable Diffusion https://astrazeneca.github.io/vlm/