LFM2.5-VL-3B for Better and Faster Vision Capabilities for the Edge

LFM2.5-VL-3B：为边缘计算提供更优、更快的视觉能力

LFM2.5-VL-3B is our most capable vision-language model you can run on your own hardware. It understands documents and screens alike, grounds objects, and can call tools. It answers directly instead of reasoning, so responses stay fast in real-time and on-device apps. LFM2.5-VL-3B 是我们目前功能最强大的视觉语言模型，您可以直接在自己的硬件上运行它。它能够理解文档和屏幕内容、进行物体定位，并具备调用工具的能力。该模型采用直接回答而非推理的模式，确保了实时应用和端侧应用中的响应速度。

LFM2.5-VL-3B extends the vision-language capabilities of our previous releases with four major improvements: LFM2.5-VL-3B 在我们之前版本的基础上，通过以下四项重大改进扩展了视觉语言能力：

Screen/UI understanding: Strong understanding of digital screens across different devices.
屏幕/UI 理解： 对不同设备的数字屏幕具有强大的理解能力。
Grounding: Improved grounding and object detection with natural language queries.
定位（Grounding）： 通过自然语言查询改进了物体定位和检测能力。
Multi-image input: Improved reasoning across multiple images.
多图像输入： 提升了跨多张图像的推理能力。
Function calling: Significantly stronger at function calling, in text-only and vision-text situations.
函数调用： 在纯文本和视觉-文本场景下，函数调用能力显著增强。

How we trained our most capable vision-language model

我们如何训练出这款最强大的视觉语言模型

LFM2.5-VL-3B pairs a SigLIP2 400M NaFlex vision encoder with the same pre-trained backbone as our LFM2.5-2.6B text model. It is pre-trained on about 34T tokens, with 4x more vision data than before, drawn from curated and synthetic image-caption, OCR, grounding, and instruction-following sets. To support non-Latin scripts, we doubled the vocabulary to 128K by extending the tokenizer in place rather than retraining from scratch. Post-training runs in two stages: First is supervised fine-tuning (SFT), with knowledge distillation from a larger teacher and Antidoom training. Second is multi-reward reinforcement learning (RL). LFM2.5-VL-3B 将 SigLIP2 400M NaFlex 视觉编码器与我们 LFM2.5-2.6B 文本模型相同的预训练主干网络相结合。它在约 34T tokens 的数据上进行了预训练，视觉数据量比之前增加了 4 倍，这些数据来自精选和合成的图像描述、OCR、定位以及指令遵循数据集。为了支持非拉丁语系脚本，我们将词汇表扩展至 128K，通过原地扩展分词器（tokenizer）而非从头开始重新训练来实现。训练后的阶段分为两步：首先是监督微调（SFT），利用大型教师模型进行知识蒸馏和 Antidoom 训练；其次是多奖励强化学习（RL）。

Benchmark results

基准测试结果

We evaluated LFM2.5-VL-3B across both vision and text benchmarks. The vision benchmarks cover multilingual visual comprehension, instruction following, visual math and scientific reasoning, document understanding, object detection, multi-image understanding, and screen understanding. LFM2.5-VL-3B leads its size class on real-world image tasks, while also reading digital content well, from documents and charts to on-screen UI elements. 我们对 LFM2.5-VL-3B 进行了视觉和文本基准测试。视觉基准测试涵盖了多语言视觉理解、指令遵循、视觉数学与科学推理、文档理解、物体检测、多图像理解以及屏幕理解。LFM2.5-VL-3B 在同等规模模型中，在现实世界图像任务中处于领先地位，同时在处理从文档、图表到屏幕 UI 元素的数字内容阅读方面表现出色。

(Note: Due to the length of the provided table, please refer to the original article for the full data breakdown.) （注：由于原文表格篇幅较长，完整数据请参考原文。）

These results demonstrate that LFM2.5-VL-3B is a strong, general-purpose vision-language model. It covers everyday tasks (captioning, visual question answering, document understanding) and is especially good at grounding objects, reading screens and documents, and calling tools. 这些结果表明，LFM2.5-VL-3B 是一款强大的通用视觉语言模型。它涵盖了日常任务（图像描述、视觉问答、文档理解），并且在物体定位、屏幕与文档阅读以及工具调用方面表现尤为突出。

Inference speed on CPU and GPU

CPU 和 GPU 上的推理速度

LFM2.5-VL-3B ships with day-one support across the inference ecosystem, including llama.cpp, MLX, vLLM, SGLang, and ONNX. LFM2.5-VL-3B 在发布首日即支持整个推理生态系统，包括 llama.cpp、MLX、vLLM、SGLang 和 ONNX。

On-device inference: LFM2.5-VL-3B decodes 228 tokens/s on an M5 Max and 116 tokens/s on a Ryzen AI Max+ 395, and fits in about 3 GB of memory. It even reaches 20 tokens/s on a Galaxy S26 Ultra, so you can run it fully on-device.
端侧推理： LFM2.5-VL-3B 在 M5 Max 上可达到 228 tokens/s 的解码速度，在 Ryzen AI Max+ 395 上为 116 tokens/s，且仅占用约 3 GB 内存。它在 Galaxy S26 Ultra 上甚至能达到 20 tokens/s，因此您可以完全在设备上运行它。
GPU inference: LFM2.5-VL-3B keeps latency consistently low and is the fastest on multi-frame inputs. LFM2.5-VL-3B is also the fastest on output throughput out of all models we tested, reaching about 11K tokens per second at high concurrency. That is roughly 2× the larger 4B-class models and ahead of even the smaller 2B-class models, which adds up to nearly 1B output tokens per day on a single H100.
GPU 推理： LFM2.5-VL-3B 保持了持续的低延迟，并且在多帧输入处理上速度最快。在所有测试模型中，LFM2.5-VL-3B 的输出吞吐量也是最高的，在高并发下可达到每秒约 11K tokens。这大约是大型 4B 类模型的 2 倍，甚至领先于更小的 2B 类模型，这意味着单张 H100 显卡每天可输出近 10 亿个 tokens。

How to use LFM2.5-VL-3B

如何使用 LFM2.5-VL-3B

Reach for LFM2.5-VL-3B when you need on-device intelligence for high-volume workloads. 当您需要为高负载工作流提供端侧智能时，请选择 LFM2.5-VL-3B。