MiniCPM-V-4.7-1B 模型发布
1B 多模态模型强化视觉编码与端侧推理效率。
为什么重要
为端侧和本地多模态应用提供更高效的 1B 级模型选择。
阅读原文huggingface.co
AI 解读（根据原文生成）
openbmb 在 Hugging Face 发布 MiniCPM-V-4.7-1B，任务类型为 image-text-to-text，基于 transformers，许可证为 apache-2.0。模型是 MiniCPM-V 4.7 系列的 1B 版本，延续单图、多图和视频理解能力，并采用 LLaVA-UHD v4 的 intra-ViT early compression 技术，在视觉编码器内提前压缩视觉 token，支持 4x/16x 混合视觉 token 压缩。模型权重已在 Hugging Face 提供，另有 Demo、数据集和相关代码链接。
要点（AI 据原文整理，数字均已与原文核对）
- MiniCPM-V 4.7 is the latest model in the MiniCPM-V series, available in a 1B variant.
- It builds on the family's single-image, multi-image, and video understanding capabilities.
- MiniCPM-V 4.7 supports iOS, Android, and HarmonyOS, together with open-source on-device deployment support.
- It also provides day-0 support for chip platforms from Rockchip, Houmo, Intel, AMD, and Arm.
- It provides GGUF, BNB, AWQ, and GPTQ quantized variants and adds FP8 quantization support.
- We also release Self-Replay Fine-Tuning, a fine-tuning approach that helps developers extend model capabilities at low cost while reducing forgetting of existing capabilities.
