v3.1.0

What's Changed

• Clustering by device performance by @maduc238 in #5

New feature:

• Clustering by device performance
• Remove hard code

Full Changelog: v3.0.0...v3.1.0

v3.0.0

New feature:

• Cluster by devices
• Multi-partition points for each cluster
  Enhancement:
• Change queue message
• Remove cluster by data

What's Changed

• Cluster devices by @maduc238 in #1

New Contributors

• @maduc238 made their first contribution in #1

Full Changelog: v2.0.1...v3.0.0

v2.0.0

New feature:

• Consolidate weights into a single *.pth file.
• Add event_time to debug events by time.

Enhancement:

• Modify the data loading mechanism to ensure all clients start training at the same time.
• Separate server-side validation into a dedicated class.
• Remove client-side validation.

v1.8.0

Split Learning v1.8.0

• Fix load state_dict from cpu to cuda
• Detect NaN loss
• Add training result and test result, retrain when false
• Average parameter with data size

Full Changelog: v1.7.1...v1.8.0

v1.7.1

Split Learning v1.7.1

• Split data for each client
• Log training time
• Profiling and partition algorithm
• Remove --num_layers on client side

v1.6.1

Split Learning v1.6.1

• Allow partition configuration
• Server only configuration
• Select model with name-based
• Refactor and package the server structure

v1.3.1

The stable version of Split Learning

• Manual configuration on server and clients
• Actively declare cut layer points for each different Class