Skip to content
Hugging Face Blog

Liquid AI adds a 280M speculative decoding drafter to its 3B vision model, hitting 3.13× decode speedup on-device

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

Liquid AI released LFM2.5-VL-DSpark, an experimental speculative decoding drafter for its LFM2.5-VL-3B vision-language model. The drafter adds only 280M parameters (8.9% of the 3B target), leaves output quality unchanged, and delivers up to 3.13× decode speedup on-device and 2.66× on an H100; end-to-end gains reach 2.62× and 2.27×. It taps hidden states from intermediate layers of the target model to draft candidate tokens—image patches and text tokens are projected into a shared representation beforehand, so the inference algorithm stays identical to the text-only version. Day-one integrations include llama.cpp, MLX-VLM, and SGLang. The post does not disclose training data size, absolute latency numbers, or speedup variation across batch sizes.

Why it matters: Liquid AI pairs its 3B vision model LFM2.5-VL with a roughly 280M-parameter draft model, using speculative decoding to speed up on-device decoding 3.13x while adding only 8.9% to deployed parameters.

Read the original ↗Export Markdown