Communication-efficient parallel split learning with aligned autoencoders.
-
Updated
Oct 21, 2026 - Python
Communication-efficient parallel split learning with aligned autoencoders.
SaveScope: audit PyTorch saved-tensor compression against gradients and actual AdamW updates. Reproducible CPU experiments and an interactive report.
Research implementation for memory-constrained causal-LM training: exact, activation-checkpointed, and INT8 residual-corrected reversible backends with audited GPU evidence.
RetainCode: One-Bit Exact Gated-MLP Activation Compression from Retained Products DOI: https://doi.org/10.5281/zenodo.21698623
To associate your repository with the activation-compression topic, visit your repo's landing page and select "manage topics."