Accelerating vision-language models with LFM2.5-VL-DSpark

Hugging Face has released an experimental DSpark draft model for their LFM2.5-VL-3B vision-language model. This new model, LFM2.5-VL-DSpark, incorporates a speculative decoding path to accelerate performance. It achieves a significant speedup with only a minimal increase in memory footprint, while maintaining the original output quality. This advancement aims to accelerate vision-language models on edge devices and beyond.