Skip to content

A simple demo for the illustration of the quantization of NNs.

Notifications You must be signed in to change notification settings

chengquan/Quantization

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

16 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Quantization

A simple demo for the illustration of the quantization of NNs.
Top file: train_mnist_nas.py

Training step

Step 1: Training

Run the “model_train()” function in debug mode and set a breakpoint at “hook = 0”.
After the finish of training, the code will be stuck at the breakpoint.
Then, please use the “dump_file()” function to dump the quantized INT8 model.

Step 2: Verification

If you need to test the quantized INT8 model, please set correct model file and run the “model_eval()” function.

Step 3: Extraction

If you want to extract the dumped file as .txt, please use the “parse_hand_dumped_file()” function to decode the dumped file.
Some relevant .txt files could be generated.

Publication

This quantization method has already been used in the following publications:

@ARTICLE{9793397, author={Huang, Mingqiang and Liu, Yucen and Man, Changhai and Li, Kai and Cheng, Quan and Mao, Wei and Yu, Hao}, journal={IEEE Transactions on Circuits and Systems I: Regular Papers}, title={A High Performance Multi-Bit-Width Booth Vector Systolic Accelerator for NAS Optimized Deep Learning Neural Networks}, year={2022}, volume={}, number={}, pages={1-13}, doi={10.1109/TCSI.2022.3178474}}

@ARTICLE{9997088, author={Cheng, Quan and Dai, Liuyao and Huang, Mingqiang and Shen, Ao and Mao, Wei and Hashimoto, Masanori and Yu, Hao}, journal={IEEE Transactions on Circuits and Systems II: Express Briefs}, title={A Low-Power Sparse Convolutional Neural Network Accelerator with Pre-Encoding Radix-4 Booth Multiplier}, year={2022}, volume={}, number={}, pages={1-1}, doi={10.1109/TCSII.2022.3231361}}

@ARTICLE{10220121, author={Cheng, Quan and Huang, Mingqiang and Man, Changhai and Shen, Ao and Dai, Liuyao and Yu, Hao and Hashimoto, Masanori}, journal={IEEE Transactions on Circuits and Systems I: Regular Papers}, title={Reliability Exploration of System-on-Chip With Multi-Bit-Width Accelerator for Multi-Precision Deep Neural Networks}, year={2023}, volume={70}, number={10}, pages={3978-3991}, doi={10.1109/TCSI.2023.3300899}}

About

A simple demo for the illustration of the quantization of NNs.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages