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

LiquidAI released LFM2.5-VL-DSpark, a 279.5M-parameter speculative decoding drafter for its LFM2.5-VL-3B vision-language model that adds 8.9% to the target model's parameter count while delivering decode speedups up to 3.13x on device and 2.66x on an H100, with end-to-end gains up to 2.62x and 2.27x respectively. The drafter ships with day-one support for llama.cpp, MLX-VLM, and SGLang, and uses a 4-layer attention-only architecture with a recommended block size of 8 or 9. LiquidAI noted speculative decoding accelerates only decode, not vision encoding or prefill, which limits gains on edge devices where prefill dominates latency.

Image-Text-to-Text • 3B • Updated • 24.7k • 211 https://huggingface.co/LiquidAI/LFM2.5-VL-3B Accelerating vision-language models with LFM2.5-VL-DSpark Team Article https://huggingface.co/blog DSpark draft model for our vision-language model VLM LFM2.5-VL-3B https://huggingface.co/LiquidAI/LFM2.5-VL-3B . As with our recently released LFM2.5-DSpark drafter models https://huggingface.co/blog/LiquidAI/lfm25-dspark , it adds a speculative decoding path that trades a minimal increase in memory footprint for a larger speedup without changing output quality. - Faster inference: decode speedups up to 3.13x on device and 2.66x on an H100, with end-to-end gains up to 2.62x and 2.27x. - Small memory cost: the drafter adds 280M parameters, 8.9% on top of the 3B target - Day-one support: LFM-compatible DSpark integrations for llama.cpp, MLX-VLM, and SGLang How does speculative decoding work for VLMs The vision drafter uses the same architecture as our text LFM2.5-DSpark drafters: it captures the target model's hidden states at a fixed set of tapped layers and conditions on them to draft a block of k candidate tokens. Image patches and text tokens are projected into a shared representation before those layers, so the drafter operates on hidden-state vectors of identical dimensionality regardless of input modality. The inference algorithm is therefore unchanged from the text models. Training and Architecture We follow the DSpark recipe with a mixture of vision-language SFT data, weighted toward the workloads we expect the model to serve. Based on ablations across 3, 4, and 5 layers, the draft model is a simplified attention-only drafter with 4 layers and a block size of 9. We ran 10 epochs on the final mixture and measured acceptance after each, which improved with additional training tokens before reaching diminishing returns. At inference time, we recommend a block size of 8 or 9 depending on the hardware. The resulting drafter has approximately 280M parameters and increases the deployed model’s parameter count by just 8.9%. | Component | LFM2.5-VL-3B | |---|---| | Decoder stack 4 layers | 193.0M | | Hidden-state projection | 21.0M | | Markov head | 65.5M | | Norms + confidence head | 6.4k | | Total | 279.5M | Inference Speedup on CPU and GPU The DSpark draft model for LFM2.5-VL-3B ships with day-one support for llama.cpp https://github.com/ggml-org/llama.cpp , MLX-VLM https://github.com/Blaizzy/mlx-vlm , and SGLang https://github.com/sgl-project/sglang . We measure both on-device inference and GPU inference. Both configurations use a DSpark block size of 8 and are evaluated on six diverse vision-based tasks, including general VQA, text VQA, image captioning, chart VQA, complex reasoning, and multi-turn conversation, following the MMSpec benchmark https://huggingface.co/papers/2603.14989 . On-device inference. With MLX on an M5 Max, decoding runs 2.30x to 3.13x faster by task. End-to-end latency improves by 1.56x to 2.62x. With llama.cpp on an M3 Ultra, decoding improves by 1.57x to 2.14x and end-to-end by 1.30x to 1.77x. GPU inference. On H100, the same drafter delivers 20.4x to 2.66x faster decoding, with end-to-end improvements of 1.64x to 2.27x. Limitations of speculation for vision workloads In LLMs, prefill is mostly compute-bound, and its cost grows sub quadratically with prompt length. VLMs add to this because the image first passes through a vision encoder, then the language backbone processes hundreds of visual tokens along with the text prompt. Edge devices have far less compute than datacenter GPUs, so prefill takes up more of the end-to-end latency, as time-to-first-token and decode measurements on Apple silicon and H100 show. The M5's per-core GPU neural accelerators narrow this gap . Speculative decoding speeds up only decode, not vision encoding or prefill. When those stages already take up much of the wall time, even a large decode speedup gives only a modest end-to-end gain. This is Amdahl's law, where the overall speedup is capped by the part of the workload that isn't accelerated. How to use LFM2.5-VL-DSpark Running the DSpark draft models with SGLang requires an SGLang build with DSpark support for LFM2 targets PR 40651 https://github.com/sgl-project/sglang/pull/40651 . Launch the target with the draft attached: python -m sglang.launch server \ --model-path LiquidAI/LFM2.5-VL-3B \ --speculative-algorithm DSPARK \ --speculative-draft-model-path LiquidAI/LFM2.5-VL-3B-DSpark \ --speculative-draft-attention-backend flashinfer \ --speculative-dspark-block-size 9 \ --disable-radix-cache Then query the OpenAI-compatible endpoint at http://localhost:30000/v1 . The block size is read from the draft's config.json ; the baseline is the same command without the three --speculative- flags. Running them with llama.cpp requires the respective llama.cpp build PR 29339 https://github.com/ggml-org/llama.cpp/pull/29339 . llama-server -m models/LFM2.5-VL-3B-F16.gguf \ --mmproj models/mmproj-LFM2.5-VL-3B-F16.gguf \ -md LFM2.5-2.6B-DSpark-F16.gguf \ --spec-type draft-dspark --spec-draft-n-max 8 --spec-draft-n-min 0 \ -fa on -ngl 99 -c 8192 Running them with MLX-VLM requires the respective build PR 2280 https://github.com/Blaizzy/mlx-vlm/pull/2280 . mlx vlm.server --model LiquidAI/LFM2.5-VL-3B --draft-model LiquidAI/LFM2.5-VL-3B-DSpark The block size is read from the sidecar metadata n-max is clamped to it . Speculative decoding is exact : the target verifies every proposed token, so greedy output equals the target alone; per-response timings report draft n / draft n accepted . Get Started Our vision DSpark draft model is available on Hugging Face in Safetensors https://huggingface.co/LiquidAI/LFM2.5-VL-3B-DSpark and GGUF formats https://huggingface.co/LiquidAI/LFM2.5-VL-3B-DSpark-GGUF . With LFM2.5, we're delivering on our vision of AI that runs anywhere. These models are: - Open-weight — Download, fine-tune, and deploy without restrictions. - Fast from day one — Day-one support for llama.cpp, MLX, and SGLang. - A complete family — From base models for customization to specialized audio and vision variants, one architecture covers diverse use cases We can’t wait to see what you build. Citation For citations, please use the following reference or BibTeX: Liquid AI, "LFM2.5-VL-DSpark: Accelerating vision-language models on edge and beyond", Liquid AI Blog, Sep 2026. @article{liquidAI2026vldspark, author = {Liquid AI}, title = {LFM2.5-VL-DSpark: Accelerating vision-language models on edge and beyond}, journal = {Liquid AI Blog}, year = {2026}, note = {www.liquid.ai/blog/lfm2-5-vl-dspark}, }