README.md

---
base_model: openbmb/MiniCPM-V-4
model_creator: openbmb
model_name: MiniCPM-V-4
quantized_by: Second State Inc.
pipeline_tag: visual-question-answering
language:
- en
- zh
---

<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<img src="https://github.com/LlamaEdge/LlamaEdge/raw/dev/assets/logo.svg" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</div>
<hr style="margin-top: 1.0em; margin-bottom: 1.0em;">
<!-- header end -->

# MiniCPM-V-4-GGUF

## Original Model

[openbmb/MiniCPM-V-4](https://huggingface.co/openbmb/MiniCPM-V-4)

## Run with LlamaEdge

- LlamaEdge version: [v0.25.1](https://github.com/LlamaEdge/LlamaEdge/releases/tag/0.25.1) and above

- Prompt template

  - Prompt type: `minicpmv`

  - Prompt string

    ```text
    <|system|>
    {system_message}<|end|>
    <|user|>
    {user_message_1}<|end|>
    <|assistant|>
    {assistant_message_1}<|end|>
    <|user|>
    {user_message_2}<|end|>
    <|assistant|>
    ```

    The `{user_message_n}` has the format: `{image_base64_encoding_string}\n{user_question}`.

- Context size: `128000`

- Run as LlamaEdge service

  ```bash
  wasmedge --dir .:. \
    --nn-preload default:GGML:AUTO:MiniCPM-V-4-Q5_K_M.gguf \
    llama-api-server.wasm \
    --prompt-template minicpmv \
    --ctx-size 128000 \
    --llava-mmproj MiniCPM-V-4-mmproj-f16.gguf \
    --model-name minicpmv-26
  ```

## Quantized GGUF Models

| Name | Quant method | Bits | Size | Use case |
| ---- | ---- | ---- | ---- | ----- |
| [MiniCPM-V-4-Q2_K.gguf](https://huggingface.co/second-state/MiniCPM-V-4-GGUF/blob/main/MiniCPM-V-4-Q2_K.gguf)     | Q2_K   | 2 | 1.40 GB| smallest, significant quality loss - not recommended for most purposes |
| [MiniCPM-V-4-Q3_K_L.gguf](https://huggingface.co/second-state/MiniCPM-V-4-GGUF/blob/main/MiniCPM-V-4-Q3_K_L.gguf) | Q3_K_L | 3 | 1.93 GB| small, substantial quality loss |
| [MiniCPM-V-4-Q3_K_M.gguf](https://huggingface.co/second-state/MiniCPM-V-4-GGUF/blob/main/MiniCPM-V-4-Q3_K_M.gguf) | Q3_K_M | 3 | 1.79 GB| very small, high quality loss |
| [MiniCPM-V-4-Q3_K_S.gguf](https://huggingface.co/second-state/MiniCPM-V-4-GGUF/blob/main/MiniCPM-V-4-Q3_K_S.gguf) | Q3_K_S | 3 | 1.63 GB| very small, high quality loss |
| [MiniCPM-V-4-Q4_0.gguf](https://huggingface.co/second-state/MiniCPM-V-4-GGUF/blob/main/MiniCPM-V-4-Q4_0.gguf)     | Q4_0   | 4 | 2.08 GB| legacy; small, very high quality loss - prefer using Q3_K_M |
| [MiniCPM-V-4-Q4_K_M.gguf](https://huggingface.co/second-state/MiniCPM-V-4-GGUF/blob/main/MiniCPM-V-4-Q4_K_M.gguf) | Q4_K_M | 4 | 2.19 GB| medium, balanced quality - recommended |
| [MiniCPM-V-4-Q4_K_S.gguf](https://huggingface.co/second-state/MiniCPM-V-4-GGUF/blob/main/MiniCPM-V-4-Q4_K_S.gguf) | Q4_K_S | 4 | 2.09 GB| small, greater quality loss |
| [MiniCPM-V-4-Q5_0.gguf](https://huggingface.co/second-state/MiniCPM-V-4-GGUF/blob/main/MiniCPM-V-4-Q5_0.gguf)     | Q5_0   | 5 | 2.51 GB| legacy; medium, balanced quality - prefer using Q4_K_M |
| [MiniCPM-V-4-Q5_K_M.gguf](https://huggingface.co/second-state/MiniCPM-V-4-GGUF/blob/main/MiniCPM-V-4-Q5_K_M.gguf) | Q5_K_M | 5 | 2.56 GB| large, very low quality loss - recommended |
| [MiniCPM-V-4-Q5_K_S.gguf](https://huggingface.co/second-state/MiniCPM-V-4-GGUF/blob/main/MiniCPM-V-4-Q5_K_S.gguf) | Q5_K_S | 5 | 2.51 GB| large, low quality loss - recommended |
| [MiniCPM-V-4-Q6_K.gguf](https://huggingface.co/second-state/MiniCPM-V-4-GGUF/blob/main/MiniCPM-V-4-Q6_K.gguf)     | Q6_K   | 6 | 2.96 GB| very large, extremely low quality loss |
| [MiniCPM-V-4-Q8_0.gguf](https://huggingface.co/second-state/MiniCPM-V-4-GGUF/blob/main/MiniCPM-V-4-Q8_0.gguf)     | Q8_0   | 8 | 3.83 GB| very large, extremely low quality loss - not recommended |
| [MiniCPM-V-4-f16.gguf](https://huggingface.co/second-state/MiniCPM-V-4-GGUF/blob/main/MiniCPM-V-4-f16.gguf)       | f16   | 16 | 7.21 GB|  |
| [MiniCPM-V-4-mmproj-f16.gguf](https://huggingface.co/second-state/MiniCPM-V-4-GGUF/blob/main/MiniCPM-V-4-mmproj-f16.gguf)       | f16   | 16 | 959 MB|  |

*Quantized with llama.cpp b6138.*