Skip to content

Latest commit

 

History

History
29 lines (24 loc) · 1.82 KB

README.md

File metadata and controls

29 lines (24 loc) · 1.82 KB

tfzip

An experimental compression pipeline for TensorFlow.

Goal

This project aims to implement a TensorFlow version of the Deep Compression pipeline outlined by Han et al.

Current Progress

This project is still in the experimental stage, it is not currently ready for production use

Pruning

  • Initial results:
    • Running compression_test.py can generate uncompressed_model and compressed_model protobuf files from a simple MNIST model for comparison.
    • Note that there is no difference in the protobuf file sizes until you apply gzip or some other compression tool.
    • The provided protobuf files were generated using a LeNet-300-100 model trained for MNIST digit classification.
Parameters Parameter Compression Protobuf Size Protobuf Size (gzipped) Protobuf Compression Accuracy
Uncompressed ~267k 3.2MB ~2.8MB 98.19%
Compressed ~27k ~10x 3.2MB ~490kB ~6x 97.31%

Trained Quantization

  • Not started.

Huffman Coding

  • Not started.