Zhikai Li comments

Results 9 comments of


                                            Zhikai Li

How long will the data generation take?

Hi, The whole process takes less than 4 min on an RTX 3090 GPU and most time is spent in the image generation stage, since the parameter calibration without training...

The speed after quantization looks same as fp32

Hi, This project aims to verify the accuracy of the quantized model using _simulated quantization_. To achieve deployment and acceleration on low-precision computing units, you need to resort to an...

您好，我这里所有图片的输出为什么都是这样的

Hello，得到的输出是：Top-5的分类标签。

latency result slower than tensorrt fp16

Hi. Our I-ViT TVM implementation is designed for the Turing Tensor Core (RTX 2080Ti), so there could be potential issues in the Ampere Tensor Core (RTX 3090) environment which could...

AttributeError: 'Attention' object has no attribute 'reduction'

Hi, There is no problem with your shell. If your model is deit, you should access line112 ('norm1') or 114 ('norm2'). My suggestion is to check if the model is...

Hello! May I inquire if the LayerNorm module has not been quantized in this model?

Hi, In this work, the output (activation) of LayerNorm is quantized, while retaining the floating-point computation of LayerNorm itself.

Hello! May I inquire if the LayerNorm module has not been quantized in this model?

Our method follows the settings of the previous PTQ4ViT and APQ-ViT, so comparisons with the PTQ4ViT and APQ-ViT are exactly fair.

Pytorch inference and TVM inference gives different result!

Hi, I think it could be a problem with the version of TVM or Timm. For example, different timm versions may import modules with different names. Please try installing the...

Pytorch inference and TVM inference gives different result!

Hi, could you try aligning the timm version with the recommended one (0.4.12)?