Skip to content

Commit 3f99522

Browse files
committed
update to flag name and updates from re-testing
1 parent fbf9cc3 commit 3f99522

1 file changed

Lines changed: 4 additions & 4 deletions

File tree

  • content/learning-paths/mobile-graphics-and-gaming/voice-assistant

content/learning-paths/mobile-graphics-and-gaming/voice-assistant/7-performance.md

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -109,7 +109,7 @@ Parameters:
109109
110110
```
111111

112-
To get benchmark numbers with use of SME kernels, you can rerun the full Benchmarking LLM on Android phone section without using `MNN_SME` flag as follows, which will leave SME instructions enabled by default:
112+
To get benchmark numbers with use of SME kernels, you can rerun the full "Benchmarking LLM on Android phone" section without setting `MNN_SME2` to `OFF`. Omitting the `MNN_SME2` flag enables SME instructions by default.:
113113

114114
```
115115
cmake --preset=x-android-aarch64 -B build/ -DBUILD_BENCHMARK=ON -DLLM_FRAMEWORK=mnn
@@ -123,11 +123,11 @@ The table table shows the measurements taken on a Vivo X300 Android phone:
123123

124124
| LLM Framework | Model | Threads | Without SME2 | With SME2 | Uplift |
125125
|-------------------|-----------------------|---------|----------------|-----------|----------|
126-
| mnn | qwen25vl-3b | 1 | 33 | 134 | 306.06 % |
127-
| | | 2 | 51 | 140 | 174.51 % |
126+
| mnn | qwen25vl-3b | 1 | 85 | 134 | 57.65 % |
127+
| | | 2 | 95 | 140 | 47.37 % |
128128
| | llama-3.2-1B | 1 | 196 | 339 | 72.96 % |
129129
| | | 2 | 275 | 396 | 44.00 % |
130-
| llama.cpp | qwen-2-VL | 1 | 113 | 146 | 29.20 % |
130+
| llama.cpp | qwen-2-VL | 1 | 113 | 146 | 29.20 % |
131131
| | | 2 | 92 | 139 | 51.09 % |
132132
| | llama-3.2-1B | 1 | 148 | 173 | 16.89 % |
133133
| | | 2 | 124 | 191 | 54.03 % |

0 commit comments

Comments
 (0)