MiniCPM-V 4.7 (35B-A3B)
Vision-language MoE, 35B total / 3B active, Apache-2.0. OpenBMB's llama.cpp-omni fork adds Lookahead Expert Scheduling to stream experts on low-memory devices (vendor demos on iPhone/iPad).
Recommended quants
- Official Q4_K_M (21.2 GB)
openbmb/MiniCPM-V-4.7-35B-A3B-gguf - Official Q4_0 (19.7 GB)
openbmb/MiniCPM-V-4.7-35B-A3B-gguf - Official Q8_0 (36.9 GB)
openbmb/MiniCPM-V-4.7-35B-A3B-gguf - DSpark draft model
openbmb/MiniCPM-V-4.7-35B-A3B-DSparkSpeculative decoding.
Chat template Works
Thinking toggled with enable_thinking.
Fix: For images in Transformers, pass downsample_mode to BOTH apply_chat_template and generate.
Sampling settings
| Instruct (official) | temp 0.7, top_p 0.95, top_k 100, min_p 0, presence_penalty 0, repetition_penalty 1.03 |
|---|
Runtimes
| vLLM | Works | Listed by OpenBMB. |
|---|---|---|
| llama.cpp | Works | Listed; low-memory expert streaming needs the OpenBMB llama.cpp-omni fork. |
| Ollama | Works | openbmb/minicpm-v4.7 on Ollama. |
| LM Studio | Unverified | |
| MLX | Unverified |
Tool calling in agent tools
| Claude Code (local endpoint) | Unverified | |
|---|---|---|
| OpenCode | Unverified | |
| Cline | Unverified | |
| Aider | Unverified |
Known issues
Sources checked
- https://huggingface.co/openbmb/MiniCPM-V-4.7-35B-A3B
- https://huggingface.co/openbmb/MiniCPM-V-4.7-35B-A3B-gguf
- https://ollama.com/openbmb/minicpm-v4.7
Spotted something wrong or a new fix? Corrections are reviewed before publishing. (Submission form coming soon.)