OpenBMB 发布 MiniCPM-V 4.7，3B 激活参数可在 iPhone 上运行
OpenBMB 放出了 MiniCPM-V 4.7，3B 激活参数就能在 iPhone 上跑到 13 tokens/s，性能还能压过参数多 10 倍的模型，本地跑多模态的朋友可以试试。
OpenBMB 开源了 MiniCPM-V 4.7 系列，包含 35B-A3B 和 1B 两个多模态模型，激活参数仅 3B。它能在 iPad Pro M4 上跑到 17+ tokens/s，在 iPhone 17 Pro Max 上跑到 13+ tokens/s，内存占用只有 3GB。核心是 LES（Lookahead Expert Scheduling）技术，通过预加载专家权重实现 5 倍推理加速，输出与原模型逐字节一致。视觉侧采用 4× 和 16× 混合 token 压缩，16× 压缩下 OCRBench 仍拿到 90.4 分，比 Qwen3.6 35B-A3B 高 5.2 分。模型覆盖 34 个基准测试，并支持 Apple、Intel、AMD、Arm、Rockchip 的 Day-0 适配。
Local AI is evolving REALLY fast Qwen 3.8 27B was released just 2 months ago and was smarter than Opus 4.6 But you needed powerful/expensive hardware to run it MiniCPM-V 4.7 is as powerful and: - only 3B active parameters (!) - it runs even on an iphone - better than models w/ 10x more active parameters So you can run a model on a phone that is better than the most powerful one available 6 months ago. The future is local. OpenBMB @OpenBMB 🚀 Introducing MiniCPM-V 4.7, our new open-source MLLM series! Advancing the Performance–Efficiency Pareto Frontier. It brings strong multimodal capabilities to iOS and Android devices with extreme efficiency. Two new models: 🔹 MiniCPM-V 4.7 35B-A3B — Bring SOTA performance to local devices with newly introduced LES (Lookahead Expert Scheduling) technique, a new milestone of on-device multimodal AI. 🔹 MiniCPM-V 4.7 1B — Lightweight, efficient multimodal AI. Easier to deploy. ⚡ What's new? SOTA performance: Competitive results across 34 benchmarks spanning multimodal understanding, STEM reasoning, OCR and long-video comprehension. Fast on-device inference: LES predicts and preloads expert weights to reduce inference stalls, boosting 5× speedup on iPad Pro M4 (16GB) with byte-for-byte identical outputs. It enables MiniCPM-V 4.7 35B-A3B to achieve 17+ tokens/s on iPad Pro M4 (16GB) and 13+ tokens/s on iPhone 17 Pro Max, while costing only 3GB memory 🤯 Efficient visual processing: 4× and 16× hybrid visual token compression. Even at 16× compression, 35B-A3B scores 90.4 on OCRBench, 5.2 points ahead of Qwen3.6 35B-A3B. Easier customization: We introduce Self-Replay Fine-Tuning, a simple way to customize new target capabilities at low cost while maintaining general performance. Broad hardware support: Day-0 adaptations for Apple, Intel, AMD, Arm and Rockchip, with many more platforms supported through FlagOS! Download, deploy, and build something we haven't imagined yet. 👀👇 🔗 G github.com/OpenBMB/MiniCP… m2QlrvX 🤗 Huggin huggingface.co/collections/op… 5QKGhymw 🔗 View Quoted Tweet 💬 5 🔄 2 ❤️ 10 👀 688 📊 5 ⚡
