diff --git a/README.md b/README.md index 2a66c0b..5d7d269 100644 --- a/README.md +++ b/README.md @@ -171,6 +171,85 @@ MiniCPM-V 4.5 can be easily used in various ways: (1) [llama.cpp](https://github +### Inference Efficiency + + +**OpenCompass** +
+ + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
ModelSizeAvg Score ↑Total Inference Time ↓
GLM-4.1V-9B-Thinking10.3B76.617.5h
MiMo-VL-7B-RL8.3B76.411h
MiniCPM-V-4_58.7B77.07.5h
+
+ +**Video-MME** + +
+ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
ModelSizeAvg Score ↑Total Inference Time ↓GPU Mem ↓
Qwen2.5-VL-7B-Instruct8.3B71.63h60G
GLM-4.1V-9B-Thinking10.3B73.62.63h32G
MiniCPM-V-4_58.7B73.50.26h28G
+
+ +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. + + ### Examples
diff --git a/README_zh.md b/README_zh.md index 68d846d..eff0b4f 100644 --- a/README_zh.md +++ b/README_zh.md @@ -163,6 +163,84 @@
+### 推理效率 + + +**OpenCompass** +
+ + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
ModelSizeAvg Score ↑Total Inference Time ↓
GLM-4.1V-9B-Thinking10.3B76.617.5h
MiMo-VL-7B-RL8.3B76.411h
MiniCPM-V-4_58.7B77.07.5h
+
+ +**Video-MME** + +
+ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
ModelSizeAvg Score ↑Total Inference Time ↓GPU Mem ↓
Qwen2.5-VL-7B-Instruct8.3B71.63h60G
GLM-4.1V-9B-Thinking10.3B73.62.63h32G
MiniCPM-V-4_58.7B73.50.26h28G
+ + +OpenCompass 和 Video-MME 均采用 A100*8卡 推理，其中 Video-MME 的推理时间未统计视频抽帧时间 + ### 典型示例