YingkunZhou

question about Firrtl(IR) optimization

11

```scala // See LICENSE.txt for license details. package solutions import chisel3._ // Problem: // // 'out' should be the sum of 'in0' and 'in1' // Adder width should be parametrized...

I use `conda install -c conda-forge spdlog` to install spdlog dependency, but it seems like that ld cannot find the spdlog library path. The error log is as following: ```bash...

MobileViTv2 colab doesn't work

1

and I changed `var_1216` to `var_1268`, it still cannot work

quantized.out int8量化后的模型在CPU后端运行resnet50出错（已更新到最新commit版本）

10

https://github.com/alibaba/MNN/issues/2614 初步测试mobilenet看起来是正常了许多，感谢开发者的努力！ 但和上述issue同样的复现方法，我又用resnet50试了试，得到的结果如下： 模型下载地址：https://github.com/onnx/models/raw/main/Computer_Vision/resnet50_Opset16_timm/resnet50_Opset16.onnx ```bash ./pictureRecognition.out resnet50.mnn daisy.jpg Load Cache file error. The device support i8sdot:1, support fp16:1, support i8mm: 0 Session Info: memory use 29.272980 MB, flops is 4096.363281...

INT8 量化模型在 ARM Android上的性能严重缩水，无法复现在benchmark.md中的加速比

2

按照官方文档https://github.com/PaddlePaddle/Paddle-Lite/blob/develop/docs/performance/benchmark_tools.md ```bash ./lite/tools/build_android.sh --toolchain=clang --with_benchmark=ON full_publish ``` 编译出benchmark_bin，在Android的termux上运行 使用的模型也是官方提供 - Resnet50: https://paddle-inference-dist.bj.bcebos.com/AI-Rank/mobile/ResNet50.tar.gz - ResNet50_quant: https://paddle-inference-dist.bj.bcebos.com/AI-Rank/mobile/ResNet50_quant.tar.gz 使用的开发板是khadas edge2, 其中大核是cortex-a76， 和 https://github.com/PaddlePaddle/Paddle-Lite/blob/develop/docs/performance/benchmark.md 中提到的Xiaomi MI9 骁龙 855 相同，主频我锁定在了2.2GHz 下面是运行的log Resnet50 ```bash ./benchmark_bin --uncombined_model_dir=ResNet50...

```bash $ cmake -DTENGINE_BUILD_QUANT_TOOL=ON .. # on my arm machine CMake Error at tools/quantize/CMakeLists.txt:68 (MESSAGE): quantization tool can only be built on x86 ```

mlmodel initialized with pretrained checkpoint is much slower than without pretrained checkpoint, wondering why

1

 the left one is initialized with pretrained checkpoint and right one is without