Research tools for communication-efficient deep learning
Project description
Research tools for communication-efficient deep learning, developed by Muhang Lan at USTC.
Latest version: 0.1.0
Installation
Install using pip
pip install deepcom
Check version number
import deepcom as dc
print(dc.__version__)
Function list
Basic tools
- Batch processing with enumerating given argument values:
python -m deepcom config.json
Deep learning perspective
-
Convert model parameters to a numpy array:
model2params()
-
Load a numpy array as model parameters:
params2model()
Communication perspective
- Calculating mutual information:
mutual_info()
Compression for training model
Compression for post-training model
- SuRP algorithm as a sparse compression for Laplacian sequence:
surp_algorithm()
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
deepcom-0.1.0.tar.gz
(5.9 kB
view hashes)