deepgeek/MiniCPM-o
2
0
Fork 0
mirror of https://github.com/OpenBMB/MiniCPM-V.git synced 2026-02-05 02:09:20 +08:00
Code Issues Packages Projects Releases Wiki Activity
Files
1d3c5f455e2a641694a10e88e40d3d84ae54d40c
MiniCPM-o/docs/minicpm_v2dot6_zh.md
yiranyyu 663d96c887 update readme
2025-08-26 18:31:16 +08:00

774 lines
24 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

## MiniCPM-V 2.6
> Archieve at: 2025-08-25
**MiniCPM-V 2.6** 是 MiniCPM-V 系列中最新、性能最佳的模型。该模型基于 SigLip-400M 和 Qwen2-7B 构建，共 8B 参数。与 MiniCPM-Llama3-V 2.5 相比，MiniCPM-V 2.6 性能提升显著，并引入了多图和视频理解的新功能。MiniCPM-V 2.6 的主要特点包括：
- 🔥 **领先的性能。**
MiniCPM-V 2.6 在最新版本 OpenCompass 榜单上（综合 8 个主流多模态评测基准）平均得分 65.2，**以8B量级的大小在单图理解方面超越了 GPT-4o mini、GPT-4V、Gemini 1.5 Pro 和 Claude 3.5 Sonnet 等主流商用闭源多模态大模型**。
- 🖼️ **多图理解和上下文学习。**
MiniCPM-V 2.6 还支持**多图对话和推理**。它在 Mantis-Eval、BLINK、Mathverse mv 和 Sciverse mv 等主流多图评测基准中取得了**最佳水平**，并展现出了优秀的上下文学习能力。
- 🎬 **视频理解。**
MiniCPM-V 2.6 还可以**接受视频输入**，进行对话和提供涵盖时序和空间信息的详细视频描述。模型在 有/无字幕 评测场景下的 Video-MME 表现均超过了 **GPT-4V、Claude 3.5 Sonnet 和 LLaVA-NeXT-Video-34B**等商用闭源模型。
- 💪 **强大的 OCR 能力及其他功能。**
MiniCPM-V 2.6 可以处理任意长宽比的图像，像素数可达 180 万（如 1344x1344）。在 OCRBench 上取得**最佳水平，超过 GPT-4o、GPT-4V 和 Gemini 1.5 Pro 等商用闭源模型**。基于最新的 [RLAIF-V](https://github.com/RLHF-V/RLAIF-V/) 和 [VisCPM](https://github.com/OpenBMB/VisCPM) 技术，其具备了**可信的多模态行为**，在 Object HalBench 上的幻觉率显著低于 GPT-4o 和 GPT-4V，并支持英语、中文、德语、法语、意大利语、韩语等**多种语言**。
- 🚀 **卓越的效率。**
除了对个人用户友好的模型大小，MiniCPM-V 2.6 还表现出**最先进的视觉 token 密度**（即每个视觉 token 编码的像素数量）。它**仅需 640 个 token 即可处理 180 万像素图像，比大多数模型少 75%**。这一特性优化了模型的推理速度、首 token 延迟、内存占用和功耗。因此，MiniCPM-V 2.6 可以支持 iPad 等终端设备上的高效**实时视频理解**。
- 💫 **易于使用。**
MiniCPM-V 2.6 可以通过多种方式轻松使用：(1) [llama.cpp](https://github.com/OpenBMB/llama.cpp/blob/minicpmv-main/examples/llava/README-minicpmv2.6.md) 和 [ollama](https://github.com/OpenBMB/ollama/blob/minicpm-v2.6/examples/minicpm-v2.6/README.md) 支持在本地设备上进行高效的 CPU 推理，(2) [int4](https://huggingface.co/openbmb/MiniCPM-V-2_6-int4) 和 [GGUF](https://huggingface.co/openbmb/MiniCPM-V-2_6-gguf) 格式的量化模型，有 16 种尺寸，(3) [vLLM](#vllm-部署-) 支持高吞吐量和内存高效的推理，(4) 针对新领域和任务进行微调，(5) 使用 [Gradio](#本地-webui-demo-) 快速设置本地 WebUI 演示，(6) 在线[demo](http://120.92.209.146:8887/)即可体验。
### 性能评估 <!-- omit in toc -->
<div align="center">
<img src=assets/radar_final.png width=90% />
</div>
<details>
<summary>点击查看 OpenCompass, MME, MMVet, OCRBench, MMMU, MathVista, MMB, AI2D, TextVQA, DocVQA, HallusionBench, Object HalBench 上的单图评测结果详情。 </summary>
<div align="center">
<table style="margin: 0px auto;">
<thead>
<tr>
<th align="left">Model</th>
<th>Size</th>
<th>Token Density<sup>+</sup></th>
<th>OpenCompass</th>
<th>MME</th>
<th>MMVet</th>
<th>OCRBench</th>
<th>MMMU val</th>
<th>MathVista mini</th>
<th>MMB1.1 test</th>
<th>AI2D</th>
<th>TextVQA val</th>
<th>DocVQA test</th>
<th>HallusionBench</th>
<th>Object HalBench</th>
</tr>
</thead>
<tbody align="center">
<tr>
<td colspan="15" align="left"><strong>Proprietary</strong></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">GPT-4o</td>
<td>-</td>
<td>1088</td>
<td>69.9</td>
<td>2328.7</td>
<td>69.1</td>
<td>736</td>
<td>69.2</td>
<td>61.3</td>
<td>82.2</td>
<td>84.6</td>
<td>-</td>
<td>92.8</td>
<td>55.0</td>
<td>17.6</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Claude 3.5 Sonnet</td>
<td>-</td>
<td>750</td>
<td>67.9</td>
<td>1920.0</td>
<td>66.0</td>
<td>788</td>
<td>65.9</td>
<td>61.6</td>
<td>78.5</td>
<td>80.2</td>
<td>-</td>
<td>95.2</td>
<td>49.9</td>
<td>13.8</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Gemini 1.5 Pro</td>
<td>-</td>
<td>-</td>
<td>64.4</td>
<td>2110.6</td>
<td>64.0</td>
<td>754</td>
<td>60.6</td>
<td>57.7</td>
<td>73.9</td>
<td>79.1</td>
<td>73.5</td>
<td>86.5</td>
<td>45.6</td>
<td>-</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">GPT-4o mini</td>
<td>-</td>
<td>1088</td>
<td>64.1</td>
<td>2003.4</td>
<td>66.9</td>
<td>785</td>
<td>60.0</td>
<td>52.4</td>
<td>76.0</td>
<td>77.8</td>
<td>-</td>
<td>-</td>
<td>46.1</td>
<td>12.4</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">GPT-4V</td>
<td>-</td>
<td>1088</td>
<td>63.5</td>
<td>2070.2</td>
<td>67.5</td>
<td>656</td>
<td>61.7</td>
<td>54.7</td>
<td>79.8</td>
<td>78.6</td>
<td>78.0</td>
<td>87.2</td>
<td>43.9</td>
<td>14.2</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Step-1V</td>
<td>-</td>
<td>-</td>
<td>59.5</td>
<td>2206.4</td>
<td>63.3</td>
<td>625</td>
<td>49.9</td>
<td>44.8</td>
<td>78.0</td>
<td>79.2</td>
<td>71.6</td>
<td>-</td>
<td>48.4</td>
<td>-</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Qwen-VL-Max</td>
<td>-</td>
<td>784</td>
<td>58.3</td>
<td>2281.7</td>
<td>61.8</td>
<td>684</td>
<td>52.0</td>
<td>43.4</td>
<td>74.6</td>
<td>75.7</td>
<td>79.5</td>
<td>93.1</td>
<td>41.2</td>
<td>13.4</td>
</tr>
<tr>
<td colspan="15" align="left"><strong>Open-source</strong></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">LLaVA-NeXT-Yi-34B</td>
<td>34B</td>
<td>157</td>
<td>55.0</td>
<td>2006.5</td>
<td>50.7</td>
<td>574</td>
<td>48.8</td>
<td>40.4</td>
<td>77.8</td>
<td>78.9</td>
<td>69.3</td>
<td>-</td>
<td>34.8</td>
<td>12.6</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Mini-Gemini-HD-34B</td>
<td>34B</td>
<td>157</td>
<td>-</td>
<td>2141</td>
<td>59.3</td>
<td>518</td>
<td>48.0</td>
<td>43.3</td>
<td>-</td>
<td>80.5</td>
<td>74.1</td>
<td>78.9</td>
<td>-</td>
<td>-</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Cambrian-34B</td>
<td>34B</td>
<td>1820</td>
<td>58.3</td>
<td>2049.9</td>
<td>53.2</td>
<td>591</td>
<td>50.4</td>
<td>50.3</td>
<td>77.8</td>
<td>79.5</td>
<td>76.7</td>
<td>75.5</td>
<td>41.6</td>
<td>14.7</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">GLM-4V-9B</td>
<td>13B</td>
<td>784</td>
<td>59.1</td>
<td>2018.8</td>
<td>58.0</td>
<td>776</td>
<td>46.9</td>
<td>51.1</td>
<td>67.9</td>
<td>71.2</td>
<td>-</td>
<td>-</td>
<td>45.0</td>
<td>-</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">InternVL2-8B</td>
<td>8B</td>
<td>706</td>
<td>64.1</td>
<td>2215.1</td>
<td>54.3</td>
<td>794</td>
<td><strong>51.2</strong></td>
<td>58.3</td>
<td><strong>79.4</strong></td>
<td><strong>83.6</strong></td>
<td>77.4</td>
<td><strong>91.6</strong></td>
<td>45.0</td>
<td>21.3</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">MiniCPM-Llama-V 2.5</td>
<td>8B</td>
<td>1882</td>
<td>58.8</td>
<td>2024.6</td>
<td>52.8</td>
<td>725</td>
<td>45.8</td>
<td>54.3</td>
<td>72.0</td>
<td>78.4</td>
<td>76.6</td>
<td>84.8</td>
<td>42.4</td>
<td>10.3</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">MiniCPM-V 2.6</td>
<td>8B</td>
<td><strong>2822</strong></td>
<td><strong>65.2</strong></td>
<td><strong>2348.4</strong>*</td>
<td><strong>60.0</strong></td>
<td><strong>852</strong>*</td>
<td>49.8*</td>
<td><strong>60.6</strong></td>
<td>78.0</td>
<td>82.1</td>
<td><strong>80.1<strong></td>
<td>90.8</td>
<td><strong>48.1</strong>*</td>
<td><strong>8.2</strong></td>
</tr>
</tbody>
</table>
</div>
* 我们使用思维链提示词来评估这些基准。
<sup>+</sup> Token Density：每个视觉 token 在最大分辨率下编码的像素数，即最大分辨率下的像素数 / 视觉 token 数。
注意：闭源模型的 Token Density 由 API 收费方式估算得到。
</details>
<details>
<summary>点击查看 Mantis Eval, BLINK, Mathverse mv, Sciverse mv, MIRB 上的多图评测结果详情。</summary>
<div align="center">
<table style="margin: 0px auto;">
<thead>
<tr>
<th align="left">Model</th>
<th>Size</th>
<th>Mantis Eval</th>
<th>BLINK val</th>
<th>Mathverse mv</th>
<th>Sciverse mv</th>
<th>MIRB</th>
</tr>
</thead>
<tbody align="center">
<tr>
<td colspan="7" align="left"><strong>Proprietary</strong></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">GPT-4V</td>
<td>-</td>
<td>62.7</td>
<td>54.6</td>
<td>60.3</td>
<td>66.9</td>
<td>53.1</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">LLaVA-NeXT-Interleave-14B</td>
<td>14B</td>
<td>66.4</td>
<td>52.6</td>
<td>32.7</td>
<td>30.2</td>
<td>-</td>
</tr>
<tr>
<td colspan="7" align="left"><strong>Open-source</strong></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Emu2-Chat</td>
<td>37B</td>
<td>37.8</td>
<td>36.2</td>
<td>-</td>
<td>27.2</td>
<td>-</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">CogVLM</td>
<td>17B</td>
<td>45.2</td>
<td>41.1</td>
<td>-</td>
<td>-</td>
<td>-</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">VPG-C</td>
<td>7B</td>
<td>52.4</td>
<td>43.1</td>
<td>24.3</td>
<td>23.1</td>
<td>-</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">VILA 8B</td>
<td>8B</td>
<td>51.2</td>
<td>39.3</td>
<td>-</td>
<td>36.5</td>
<td>-</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">InternLM-XComposer-2.5</td>
<td>8B</td>
<td>53.1*</td>
<td>48.9</td>
<td>32.1*</td>
<td>-</td>
<td>42.5</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">InternVL2-8B</td>
<td>8B</td>
<td>59.0*</td>
<td>50.9</td>
<td>30.5*</td>
<td>34.4*</td>
<td><strong>56.9*</strong></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">MiniCPM-V 2.6</td>
<td>8B</td>
<td><strong>69.1</strong></td>
<td><strong>53.0</strong></td>
<td><strong>84.9</strong></td>
<td><strong>74.9</strong></td>
<td>53.8</td>
</tr>
</tbody>
</table>
</div>
* 正式开源模型权重的评测结果。
</details>
<details>
<summary>点击查看 Video-MME 和 Video-ChatGPT 上的视频评测结果详情。</summary>
<div align="center">
<table style="margin: 0px auto;">
<thead>
<tr>
<th align="left">Model</th>
<th>Size</th>
<th colspan="2">Video-MME</th>
<th colspan="5">Video-ChatGPT</th>
</tr>
<tr>
<th align="left"></th>
<th></th>
<th>w/o subs</th>
<th>w subs</th>
<th>Correctness</th>
<th>Detail</th>
<th>Context</th>
<th>Temporal</th>
<th>Consistency</th>
</tr>
</thead>
<tbody align="center">
<tr>
<td colspan="9" align="left"><strong>Proprietary</strong></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">Claude 3.5 Sonnet</td>
<td>-</td>
<td>60.0</td>
<td>62.9</td>
<td>-</td>
<td>-</td>
<td>-</td>
<td>-</td>
<td>-</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">GPT-4V</td>
<td>-</td>
<td>59.9</td>
<td>63.3</td>
<td>-</td>
<td>-</td>
<td>-</td>
<td>-</td>
<td>-</td>
</tr>
<tr>
<td colspan="9" align="left"><strong>Open-source</strong></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">LLaVA-NeXT-7B</td>
<td>7B</td>
<td>-</td>
<td>-</td>
<td>3.39</td>
<td>3.29</td>
<td>3.92</td>
<td>2.60</td>
<td>3.12</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">LLaVA-NeXT-34B</td>
<td>34B</td>
<td>-</td>
<td>-</td>
<td>3.29</td>
<td>3.23</td>
<td>3.83</td>
<td>2.51</td>
<td>3.47</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">CogVLM2-Video</td>
<td>12B</td>
<td>-</td>
<td>-</td>
<td>3.49</td>
<td><strong>3.46</strong></td>
<td>3.23</td>
<td><strong>2.98</strong></td>
<td><strong>3.64</strong></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">LongVA</td>
<td>7B</td>
<td>52.4</td>
<td>54.3</td>
<td>3.05</td>
<td>3.09</td>
<td>3.77</td>
<td>2.44</td>
<td><strong>3.64</strong></td>
</tr>
<tr>
<td nowrap="nowrap" align="left">InternVL2-8B</td>
<td>8B</td>
<td>54.0</td>
<td>56.9</td>
<td>-</td>
<td>-</td>
<td>-</td>
<td>-</td>
<td>-</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">InternLM-XComposer-2.5</td>
<td>8B</td>
<td>55.8</td>
<td>-</td>
<td>-</td>
<td>-</td>
<td>-</td>
<td>-</td>
<td>-</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">LLaVA-NeXT-Video</td>
<td>32B</td>
<td>60.2</td>
<td>63.0</td>
<td>3.48</td>
<td>3.37</td>
<td><strong>3.95</strong></td>
<td>2.64</td>
<td>3.28</td>
</tr>
<tr>
<td nowrap="nowrap" align="left">MiniCPM-V 2.6</td>
<td>8B</td>
<td><strong>60.9</strong></td>
<td><strong>63.6</strong></td>
<td><strong>3.59</strong></td>
<td>3.28</td>
<td>3.93</td>
<td>2.73</td>
<td>3.62</td>
</tr>
</tbody>
</table>
</div>
</details>
<details>
<summary>点击查看 TextVQA, VizWiz, VQAv2, OK-VQA上的少样本评测结果详情。</summary>
<div align="center">
<table style="margin: 0px auto;">
<thead>
<tr>
<th align="left">Model</th>
<th>Size</th>
<th>Shot</th>
<th>TextVQA val</th>
<th>VizWiz test-dev</th>
<th>VQAv2 test-dev</th>
<th>OK-VQA val</th>
</tr>
</thead>
<tbody align="center">
<tr>
<td align="left" nowrap="nowrap" rowspan="3">Flamingo</td>
<td rowspan="3">80B</td>
<td>0*</td>
<td>35.0</td>
<td>31.6</td>
<td>56.3</td>
<td>40.6</td>
</tr>
<tr>
<td>4</td>
<td>36.5</td>
<td>39.6</td>
<td>63.1</td>
<td><strong>57.4</strong></td>
</tr>
<tr>
<td>8</td>
<td>37.3</td>
<td>44.8</td>
<td>65.6</td>
<td>57.5</td>
</tr>
<tr>
<td align="left" nowrap="nowrap" rowspan="3">IDEFICS</td>
<td rowspan="3">80B</td>
<td>0*</td>
<td>30.9</td>
<td>36.0</td>
<td>60.0</td>
<td>45.2</td>
</tr>
<tr>
<td>4</td>
<td>34.3</td>
<td>40.4</td>
<td>63.6</td>
<td>52.4</td>
</tr>
<tr>
<td>8</td>
<td>35.7</td>
<td>46.1</td>
<td>64.8</td>
<td>55.1</td>
</tr>
<tr>
<td align="left" nowrap="nowrap" rowspan="3">OmniCorpus</td>
<td rowspan="3">7B</td>
<td>0*</td>
<td>43.0</td>
<td>49.8</td>
<td>63.2</td>
<td>45.5</td>
</tr>
<tr>
<td>4</td>
<td>45.4</td>
<td>51.3</td>
<td>64.5</td>
<td>46.5</td>
</tr>
<tr>
<td>8</td>
<td>45.6</td>
<td>52.2</td>
<td>64.7</td>
<td>46.6</td>
</tr>
<tr>
<td align="left" nowrap="nowrap" rowspan="3">Emu2</td>
<td rowspan="3">37B</td>
<td>0</td>
<td>26.4</td>
<td>40.4</td>
<td>33.5</td>
<td>26.7</td>
</tr>
<tr>
<td>4</td>
<td>48.2</td>
<td>54.6</td>
<td>67.0</td>
<td>53.2</td>
</tr>
<tr>
<td>8</td>
<td>49.3</td>
<td>54.7</td>
<td>67.8</td>
<td>54.1</td>
</tr>
<tr>
<td align="left" nowrap="nowrap" rowspan="2">MM1</td>
<td rowspan="2">30B</td>
<td>0</td>
<td>26.2</td>
<td>40.4</td>
<td>48.9</td>
<td>26.7</td>
</tr>
<tr>
<td>8</td>
<td>49.3</td>
<td>54.7</td>
<td><strong>70.9</strong></td>
<td>54.1</td>
</tr>
<tr>
<td align="left" nowrap="nowrap" rowspan="3">MiniCPM-V 2.6<sup>+</sup></td>
<td rowspan="3">8B</td>
<td>0</td>
<td>43.9</td>
<td>33.8</td>
<td>45.4</td>
<td>23.9</td>
</tr>
<tr>
<td>4</td>
<td>63.6</td>
<td>60.5</td>
<td>65.5</td>
<td>50.1</td>
</tr>
<tr>
<td>8</td>
<td><strong>64.6</strong></td>
<td><strong>63.4</strong></td>
<td>68.2</td>
<td>51.4</td>
</tr>
</tbody>
</table>
</div>
* 使用 Flamingo 方式 zero image shot 和 two additional text shots 评估零样本性能。
<sup>+</sup> 我们在没有进行监督微调 (SFT) 的情况下评估预训练的模型权重 (ckpt)。
</details>
### 典型示例 <!-- omit in toc -->
<div style="display: flex; flex-direction: column; align-items: center;">
<img src="../assets/minicpmv2_6/multi_img-bike.png" alt="Bike" style="margin-bottom: 5px;">
<img src="../assets/minicpmv2_6/multi_img-menu.png" alt="Menu" style="margin-bottom: 5px;">
<img src="../assets/minicpmv2_6/multi_img-code.png" alt="Code" style="margin-bottom: 5px;">
<img src="../assets/minicpmv2_6/ICL-Mem.png" alt="Mem" style="margin-bottom: 5px;">
<img src="../assets/minicpmv2_6/multiling-medal.png" alt="medal" style="margin-bottom: 10px;">
</div>
<details>
<summary>点击查看更多示例。</summary>
<div style="display: flex; flex-direction: column; align-items: center;">
<img src="../assets/minicpmv2_6/ICL-elec.png" alt="elec" style="margin-bottom: 5px;">
<img src="../assets/minicpmv2_6/multiling-olympic.png" alt="Menu" style="margin-bottom: 10px;">
</div>
</details>
我们将 MiniCPM-V 2.6 部署在iPad Pro上，并录制了以下演示视频。
<table align="center">
<p align="center">
<img src="../assets/gif_cases/ai.gif" width=32%/>
&nbsp;&nbsp;&nbsp;&nbsp;
<img src="../assets/gif_cases/beer.gif" width=32%/>
</p>
</table>
<table align="center">
<p align="center">
<video src="https://github.com/user-attachments/assets/21f4b818-ede1-4822-920e-91281725c830" width="360" /> </video>
<!-- <video src="https://github.com/user-attachments/assets/c835f757-206b-4d9c-8e36-70d67b453628" width="360" /> </video> -->
</p>
</table>
</details>
### 模型库
| 模型 | 设备 | 资源 | &emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp; 简介 | 下载链接 |
|:--------------|:-:|:----------:|:-------------------|:---------------:|
| MiniCPM-V 2.6| GPU | 17 GB | 提供出色的端侧单图、多图、视频理解能力。 | [🤗](https://huggingface.co/openbmb/MiniCPM-V-2_6) &nbsp;&nbsp; [<img src="./assets/modelscope_logo.png" width="20px"></img>](https://modelscope.cn/models/OpenBMB/MiniCPM-V-2_6) |
| MiniCPM-V 2.6 gguf | CPU | 6 GB | gguf 版本，更低的内存占用和更高的推理效率。 | [🤗](https://huggingface.co/openbmb/MiniCPM-V-2_6-gguf) &nbsp;&nbsp; [<img src="./assets/modelscope_logo.png" width="20px"></img>](https://modelscope.cn/models/OpenBMB/MiniCPM-V-2_6-gguf) |
| MiniCPM-V 2.6 int4 | GPU | 7 GB | int4量化版，更低显存占用。 | [🤗](https://huggingface.co/openbmb/MiniCPM-V-2_6-int4) &nbsp;&nbsp; [<img src="./assets/modelscope_logo.png" width="20px"></img>](https://modelscope.cn/models/OpenBMB/MiniCPM-V-2_6-int4) |
Reference in New Issue View Git Blame Copy Permalink