MiniCPM-V-4.6-GGUF
GGUF quantizations of openbmb/MiniCPM-V-4.6 with its vision projector, for llama.cpp and Ollama.
| File | What |
|---|---|
MiniCPM-V-4.6-Q4_K_M.gguf |
Language model, Q4_K_M (recommended) |
MiniCPM-V-4.6-Q8_0.gguf |
Language model, Q8_0 |
mmproj-MiniCPM-V-4.6-F16.gguf |
Vision projector (f16), required for images |
Ollama
ollama run richardyoung/minicpm-v-4.6
Tags: Q4_K_M (default, latest), Q8_0. Image input works in the CLI (drag in a file path) and via the API images field.
llama.cpp
llama-mtmd-cli -m MiniCPM-V-4.6-Q4_K_M.gguf --mmproj mmproj-MiniCPM-V-4.6-F16.gguf --image page.png -p "Transcribe all the text in this image."
Verified
Tested before upload through Ollama with a synthetic invoice and a shapes image (Q4_K_M shown):
- Invoice transcription: 'INVOICE 4471 / Date: 2026-09-28 / Widgets x3 .... $42.00 / TOTAL: $86.20'
- Shapes: 'The image shows two shapes: a red circle and a blue square. The circle is red, and the square is blue. The shapes are placed against a light gray background.'
- Converted with
--no-mtp: the upstream config declares a multi-token-prediction layer whose weights are not in the release. Converted with llama.cppconvert_hf_to_gguf.py(text and--mmproj), quantized withllama-quantize.