OpenBMB · AI news

Even small 2B-scale retrievers can double mixed-corpus retrieval quality once modality bias is addressed in training, which matters for anyone building RAG over documents with screenshots, tables or charts. The paper, code and project page are open, so teams can adopt the Multi-Positive View InfoNCE recipe or swap Trident-Qwen3VL-2B into their existing embedding stack.

#OpenBMB #Trident #Qwen3VL #RAG

OpenBMB@OpenBMBSource

Models Builders11:37

Chinese contentMiniCPM-V 4.7 登上 Hugging Face 热榜第二

OpenBMB 的 MiniCPM-V 4.7 冲上 Hugging Face 热榜第二，35B-A3B MoE 手机端 3GB 内存就能跑，想玩端侧多模态的可以试试。

#MiniCPM-V #OpenBMB #Hugging Face #多模态

OpenBMB@OpenBMBSource

Products Builders10:34

Chinese contentOpenBMB 转发用户实测：MiniCPM-V 4.7 1B 在手机端运行

OpenBMB 官方转发的端侧实测，1B 参数的 MiniCPM-V 4.7 直接跑在手机上，想玩本地部署的可以看看这个演示。

#MiniCPM #OpenBMB #端侧模型 #多模态

OpenBMB@OpenBMBSource

Products BuildersOfficial09:59

Chinese content清华、面壁智能与 OpenBMB 开源 PilotDeck 2.0 智能体操作系统

清华和面壁开源了个智能体操作系统 PilotDeck 2.0，记忆能直接看能改，简单任务自动交给小模型跑，长任务开销省了 83%。

#PilotDeck #ModelBest #OpenBMB #智能体

官方一手pandaily@[email protected] (Pandaily)Source

Products US vs ChinaSelected21:42

PilotDeck 2.0 Open-Source Agent Harness Beats Claude Code on Workbuddy Bench

An open-source agent harness beating Claude Code on a named benchmark challenges the assumption that agent platforms must lock users into proprietary models. Developers running local or Chinese open-weight models like Qwen get a transparent alternative with verifiable file outputs rather than drafts. Availability is immediate via GitHub and desktop installer, with no subscription cost.

#PilotDeck #OpenBMB #Qwen #agents

OpenBMB@OpenBMBSource

Models BuildersSelected11:39

OpenBMB releases MiniCPM-V 4.7 multimodal model for phones

The core shift is system-level: LES moves the bottleneck from memory capacity to flash bandwidth and expert-prediction hit rate, letting a 35B-class model live on consumer phones without a cloud round-trip. Developers building on-device apps can fine-tune with SRFT's 4:1 task-to-replay data mix while keeping general abilities largely intact. Weights and code are on Hugging Face and GitHub, so teams can experiment today.

#MiniCPM-V #OpenBMB #open-source models #multimodal

shao__meng@shao__mengSource

Models Builders21:25

Chinese contentMiniCPM-V 4.7 发布：35B 多模态 MoE 模型仅用 3GB 内存跑在手机上

OpenBMB 把 35B 的多模态模型塞进了 3GB 内存，手机能离线读文档、认路牌，还给了 1B 小版本和微调方法，端侧玩法直接变多了。

#MiniCPM-V #OpenBMB #MoE #多模态

OpenBMB@OpenBMBSource

Models BuildersMulti-source18:26

Chinese contentOpenBMB 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。

#MiniCPM-V #OpenBMB #Qwen #多模态

Event

Paul Couvert@itsPaulAi8 sourcesSource

Industry Builders17:15

Chinese contentOpenBMB：社区推动本地多模态 AI 落地手机与边缘设备

OpenBMB 转发致谢，聊的是本地多模态 AI 跑手机和边缘设备这事，社区热情挺高，可以顺手关注一下。

#OpenBMB #多模态 #边缘设备

OpenBMB@OpenBMBSource

Models Builders13:43

Chinese contentOpenBMB 推出 MiniCPM-V-4.7-35B-A3B 20GB 量化版，保留约 95% 性能

MiniCPM-V-4.7-35B-A3B 出了 20GB 量化版，性能只掉 5% 左右，消费级显卡就能本地跑视觉模型了。

#MiniCPM-V #OpenBMB #量化 #多模态

OpenBMB@OpenBMBSource

Models Builders11:37

Chinese contentOpenBMB 展示 MiniCPM-V-4.7-35B-A3B 在真实智能体工作流中的表现

OpenBMB 的 MiniCPM-V-4.7（35B 总参、3B 激活）跑进了真实 agent 工作流，重点考验工具调用和长上下文，关注高效模型跑 agent 的可以看看。

#MiniCPM #OpenBMB #智能体 #开源模型

OpenBMB@OpenBMBSource

Models Builders05:59

Chinese contentAtomic Chat 发布去审查版 MiniCPM-V 4.7 35B-A3B

OpenBMB 的 MiniCPM-V 4.7 35B-A3B 被第三方去掉了审查，拒绝率从 96.3% 降到 0，Hugging Face 上能直接下载本地跑。

#MiniCPM-V 4.7 #OpenBMB #开源模型 #多模态

@atomic_chat_hq@atomic_chat_hqSource

Models US vs China05:18

Chinese contentMiniCPM-V 4.7 在 Pagoda 测试中击败 Qwen 3.6 35B-3A

OpenBMB 的 MiniCPM-V 4.7 在 3D 生成上赢了 Qwen 3.6 35B-3A，还省 10% token，本地跑视觉模型可以关注一下。

#MiniCPM-V #OpenBMB #Qwen #多模态

@atomic_chat_hq@atomic_chat_hqSource

Models BuildersMulti-source00:24

Chinese contentOpenBMB 发布开源多模态模型 MiniCPM-V 4.7 系列

OpenBMB 的 MiniCPM-V 4.7 能在 iPhone 上跑 35B 多模态模型，3GB 内存 13 tokens/s，OCR 还压过 Qwen3.6，端侧玩家可以试试。

#MiniCPM-V #OpenBMB #多模态 #端侧部署

Event

OpenBMB@OpenBMB7 sourcesSource

Papers Builders71°21:09

Chinese contentOpenBMB推出扩散奖励模型DRM

OpenBMB发DRM模型，不再把人类偏好压缩成单一分数，能保留分歧和不确定性，还能提升奖励决策可靠性。

#DRM #OpenBMB #奖励模型 #RLHF

OpenBMB@OpenBMBSource

Models BuildersMulti-sourceSelected73°21:10

MiniCPM5-2B 4-bit EXL3 quantization shrinks weights to 1.61 GB

A 1.61 GB weight file that still hits ~70 tokens/s on a Tesla T4 shows how far small open models can be squeezed onto aging or low-power GPUs, no datacenter needed. Practitioners running local inference on modest hardware can pair it with ExLlamaV3 or TabbyAPI to serve on-device chat or agent backends. Compared with running the full-precision MiniCPM5-2B, this trade-off suits edge and consumer setups over maximum accuracy.

#MiniCPM #OpenBMB #quantization #open-source models

Event

OpenBMB@OpenBMB11 sourcesSource

Models Builders02:24

Chinese contentOpenBMB发布MiniCPM模型及Meshy训练框架

OpenBMB开源了MiniCPM模型和Meshy训练框架，小型开源模型现在表现很不错。

#MiniCPM #OpenBMB #Meshy #开源模型

Paul Couvert@itsPaulAiSource

Models BuildersSelected73°15:02

MiniCPM5-2B model achieves top benchmark score

MiniCPM5-2B offers superior performance for edge deployment with its 2B parameters, outperforming larger models while being accessible for developers in the region.

#MiniCPM5-2B #面壁智能 #OpenBMB #open-source models

IT之家Source