openbmb/MiniCPM-V-4.7-35B-A3B-gguf — new model trending #30 on Hugging Face
OpenBMB released GGUF weights for MiniCPM-V 4.7 35B-A3B, an on-device multimodal MoE model.
Hugging Face is hosting a GGUF (llama.cpp) quantization of OpenBMB's MiniCPM-V 4.7 35B-A3B, the first mixture-of-experts model in the MiniCPM-V vision-language series. OpenBMB says Lookahead Expert Scheduling, layer-wise quantization, and speculative sampling enable smooth on-device decoding at more than 13 tokens per second using 3 GB on an iPhone 17 Pro Max. In Instruct mode, 35B-A3B is claimed to beat Qwen3.6 35B-A3B overall while using less inference time, with early visual-token compression and mixed 4x/16x compression. Weights support SGLang, vLLM, llama.cpp, and Ollama, plus GGUF, BNB, AWQ, GPTQ, and FP8 formats and Self-Replay Fine-Tuning.
64