Liquid AI Releases LFM2.5-VL-DSpark to Accelerate Vision-Language Model Inference

Liquid AI Releases LFM2.5-VL-DSpark to Accelerate Vision-Language Model Inference Liquid AI has released an experimental DSpark draft model for its LFM2.5-VL-3B vision-language model, providing speculative decoding to boost inference speeds on device and GPU hardware with minimal parameter overhead. Adds a 279.5M-parameter attention-only draft model, increasing the 3B target model's parameter count by 8.9%. Delivers decoding speedups up to 3.13x on an M5 Max with MLX and up to 2.66x on an H100 GPU. Provides day-one integration across llama.cpp, MLX-VLM, and SGLang using Safetensors and GGUF formats. Overall end-to-end acceleration is bounded by Amdahl's law, as speculative decoding speeds up decoding but does not accelerate visual encoding or prefill. Architecture and Speculative Mechanism The vision drafter adopts the same architecture as Liquid AI's text LFM2.5-DSpark models. Because image patches and text tokens are projected into a shared dimensionality prior to the tapped layers, the drafter operates directly on target hidden states regardless of modality. The drafter features 4 attention-only layers, a hidden-state projection, a Markov head, and norms with a confidence head, totaling roughly 280M parameters. It was trained over 10 epochs on vision-language SFT data, with recommended inference block sizes of 8 or 9. Hardware Benchmark Performance Evaluated across six MMSpec tasks, MLX on an Apple M5 Max saw decoding run 2.30x to 3.13x faster, with end-to-end latency improving between 1.56x and 2.62x. Running llama.cpp on an M3 Ultra improved decoding by 1.57x to 2.14x and end-to-end latency by 1.30x to 1.77x. On an H100 GPU, the draft model yielded decode speedups up to 2.66x and end-to-end latency reductions of 1.64x to 2.27x. Workload Limitations and Amdahl's Law VLM inference involves passing images through an encoder and processing hundreds of visual tokens alongside text prompts, making prefill compute-heavy. Because speculative decoding accelerates only the token decode stage and leaves vision encoding and prefill unaccelerated, Amdahl's law restricts total end-to-end speedups, particularly on edge hardware where prefill constitutes a substantial fraction of overall runtime.

Liquid AI推出LFM2.5-VL-DSpark草稿模型：为3B视觉语言模型带来显著推理加速 Liquid AI发布了针对视觉语言模型LFM2.5-VL-3B的实验性DSpark投机解码草稿模型。该模型仅增加约2.8亿参数（8.9%的显存开销），便可在不改变输出质量的前提下，实现端到端最高2.62倍的推理加速，并上线即支持llama.cpp、MLX-VLM和SGLang等主流框架。 DSpark草稿模型拥有约279.5M参数，采用4层纯注意力简化架构，建议块大小为8或9。 在M5 Max设备上解码速度提升达2.30至3.13倍（端到端提升1.56至2.62倍），H100上端到端加速达1.64至2.27倍。 投机解码仅优化生成解码阶段，不加速视觉编码与Prefill阶段，整体加速比受阿姆达尔定律制约。 模型已开源权重，并支持Safetensors与GGUF格式，在SGLang、llama.cpp及MLX-VLM中实现开箱即用。 架构设计与工作机制 LFM2.5-VL-DSpark沿用了文本模型的投机解码架构。图像块与文本词元在进入抽取的网络层前已被投影至共享表征，草稿模型直接在相同维度的隐藏状态向量上进行预测并生成k个候选词元。经过层数消融实验后，草稿模型确定为4层结构，由193.0M的解码器堆栈、21.0M隐藏状态投影和65.5M的Markov头等部分组成，总计约279.5M参数。 端侧与GPU推理加速表现 在MMSpec基准下的六项视觉任务测试中，该模型表现出稳定的加速效果。在端侧平台，搭载MLX的M5 Max解码提速达2.30至3.13倍，搭载llama.cpp的M3 Ultra端到端提速达1.30至1.77倍；在数据中心H100 GPU上，解码速度与端到端延迟也分别实现了显著优化。由于目标模型会逐一验证草稿词元，其输出结果与单独运行目标模型严格一致。 视觉任务中的加速局限性 视觉语言模型相较于纯文本模型在推理初期存在额外的计算瓶颈。图像必须先经过视觉编码器处理，并与文本提示一同产生数百个视觉词元，使得Prefill耗时占据了端到端延迟的较大比例。投机解码技术仅能缩短生成解码时间，无法压缩视觉编码与Prefill耗时，因此遵循阿姆达尔定律，端到端的总体提速幅度会受到未被加速阶段的限制。

Today, we release an experimental DSpark draft model for our vision-language model (VLM) LFM2.5-VL-3B . As with our recently released LFM2.5-DSpark drafter models , 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 , MLX-VLM , and 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 . 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 ). 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 ). 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 ). 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 and GGUF formats . 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}, }