Update README.md
Browse files
README.md
CHANGED
|
@@ -60,6 +60,83 @@ MiniCPM-V 4.5 can be easily used in various ways: (1) [llama.cpp](https://github
|
|
| 60 |
<img src="https://raw.githubusercontent.com/openbmb/MiniCPM-o/main/assets/minicpmv_4_5_evaluation_result.png" , width=100%>
|
| 61 |
</div>
|
| 62 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 63 |
### Examples
|
| 64 |
|
| 65 |
<div align="center">
|
|
|
|
| 60 |
<img src="https://raw.githubusercontent.com/openbmb/MiniCPM-o/main/assets/minicpmv_4_5_evaluation_result.png" , width=100%>
|
| 61 |
</div>
|
| 62 |
|
| 63 |
+
### Inference Efficiency
|
| 64 |
+
|
| 65 |
+
**OpenCompass**
|
| 66 |
+
<div align="left">
|
| 67 |
+
<table style="margin: 0px auto;">
|
| 68 |
+
<thead>
|
| 69 |
+
<tr>
|
| 70 |
+
<th align="left">Model</th>
|
| 71 |
+
<th>Size</th>
|
| 72 |
+
<th>Avg Score ↑</th>
|
| 73 |
+
<th>Total Inference Time ↓</th>
|
| 74 |
+
</tr>
|
| 75 |
+
</thead>
|
| 76 |
+
<tbody align="center">
|
| 77 |
+
<tr>
|
| 78 |
+
<td nowrap="nowrap" align="left">GLM-4.1V-9B-Thinking</td>
|
| 79 |
+
<td>10.3B</td>
|
| 80 |
+
<td>76.6</td>
|
| 81 |
+
<td>17.5h</td>
|
| 82 |
+
</tr>
|
| 83 |
+
<tr>
|
| 84 |
+
<td nowrap="nowrap" align="left">MiMo-VL-7B-RL</td>
|
| 85 |
+
<td>8.3B</td>
|
| 86 |
+
<td>76.4</td>
|
| 87 |
+
<td>11h</td>
|
| 88 |
+
</tr>
|
| 89 |
+
<tr>
|
| 90 |
+
<td nowrap="nowrap" align="left">MiniCPM-V 4.5</td>
|
| 91 |
+
<td>8.7B</td>
|
| 92 |
+
<td>77.0</td>
|
| 93 |
+
<td>7.5h</td>
|
| 94 |
+
</tr>
|
| 95 |
+
</tbody>
|
| 96 |
+
</table>
|
| 97 |
+
</div>
|
| 98 |
+
|
| 99 |
+
**Video-MME**
|
| 100 |
+
|
| 101 |
+
<div align="left">
|
| 102 |
+
<table style="margin: 0px auto;">
|
| 103 |
+
<thead>
|
| 104 |
+
<tr>
|
| 105 |
+
<th align="left">Model</th>
|
| 106 |
+
<th>Size</th>
|
| 107 |
+
<th>Avg Score ↑</th>
|
| 108 |
+
<th>Total Inference Time ↓</th>
|
| 109 |
+
<th>GPU Mem ↓</th>
|
| 110 |
+
</tr>
|
| 111 |
+
</thead>
|
| 112 |
+
<tbody align="center">
|
| 113 |
+
<tr>
|
| 114 |
+
<td nowrap="nowrap" align="left">Qwen2.5-VL-7B-Instruct</td>
|
| 115 |
+
<td>8.3B</td>
|
| 116 |
+
<td>71.6</td>
|
| 117 |
+
<td>3h</td>
|
| 118 |
+
<td>60G</td>
|
| 119 |
+
</tr>
|
| 120 |
+
<tr>
|
| 121 |
+
<td nowrap="nowrap" align="left">GLM-4.1V-9B-Thinking</td>
|
| 122 |
+
<td>10.3B</td>
|
| 123 |
+
<td>73.6</td>
|
| 124 |
+
<td>2.63h</td>
|
| 125 |
+
<td>32G</td>
|
| 126 |
+
</tr>
|
| 127 |
+
<tr>
|
| 128 |
+
<td nowrap="nowrap" align="left">MiniCPM-V 4.5</td>
|
| 129 |
+
<td>8.7B</td>
|
| 130 |
+
<td>73.5</td>
|
| 131 |
+
<td>0.26h</td>
|
| 132 |
+
<td>28G</td>
|
| 133 |
+
</tr>
|
| 134 |
+
</tbody>
|
| 135 |
+
</table>
|
| 136 |
+
</div>
|
| 137 |
+
|
| 138 |
+
Both Video-MME and OpenCompass were evaluated using 8×A100 GPUs for inference. The reported inference time of Video-MME excludes the cost of video frame extraction.
|
| 139 |
+
|
| 140 |
### Examples
|
| 141 |
|
| 142 |
<div align="center">
|