PaddlePaddle study notes 6

PaddlePaddle study notes 6

Multiple choice

[1] What are the advantages of fixed-point quantification?

A. Memory bandwidth B. Low power consumption C. Low computing resources D. Low storage volume

[2] In the routine distillation task, which of the following statements is correct:

A. Only the parameters of the teacher model need to be updated

B. Only the parameters of the student model need to be updated

C. The parameters of teacher model and student model need to be updated

D. The parameters of teacher model and student model do not need to be updated

[3] Can ResNet50 be distilled with MobileNetv1?

A: Yes

B: No

[4] Which of the following methods can reduce model reasoning time?

A. Only the weight weight is quantified

B. Distill the ResNet50 model to improve accuracy

C. Cut the model to reduce the number of channels

D. Quantify the weight and activation, and use INT8 to calculate the prediction

[5] The three key elements of NAS are:

A. Search space

B. Search algorithm

C. Model optimization

D. Model evaluation

Multiple choice answer sheet

Please write the answer to each multiple-choice question here:

【1】A B C

【2】B

【3】A

【4】A C D

【5】A B D

Image classification model quantification tutorial

This tutorial uses the image classification model MobileNetV1 as an example to illustrate how to quickly use the quantitative training interface .
The example includes the following steps:

1. Import dependencies
2. Build model
3. Define input data
4. Training model
5. Quantitative model 这个步骤中需要添加代码
6. Train and test the quantized model