Bnb/multi gpu#317

Merged

bnb32 merged 34 commits into

mainfrom

May 15, 2026

Commits on May 3, 2026

feat: implement multi-GPU support and related optimizations in Sup3r models
bnb32
committed
refactor: streamline loss calculation and update loss details in Sup3rGan
bnb32
committed
Refactor Sup3rX class for improved caching and feature handling
bnb32
committed
test: add assertions for queue state in eager and lazy batchers
bnb32
committed
feat: implement exception handling in TrainingSession with dedicated thread class
bnb32
committed
refactor: remove unnecessary tf.function decorator from calc_loss method in Sup3rGan
bnb32
committed
feat: enhance loss metrics with TensorFlow rank assertions and optimize kernel calculations
bnb32
committed
fix: ensure proper data type casting for high-resolution features in AbstractSingleModel
bnb32
committed
fix: remove unnecessary type casting for exo in AbstractSingleModel
bnb32
committed
fix: update error handling in loss functions to raise ValueError instead of AssertionError
bnb32
committed

Commits on May 5, 2026

refactor: change logger info to debug level for training loss and batch step messages in Sup3rGan
bnb32
committed

Commits on May 6, 2026

fix: update TensorFlow function decorators to reduce retracing and improve performance; adjust TrainingConfig defaults for checkpoint interval and adaptive update bounds
bnb32
committed

Commits on May 9, 2026

Enhance logging across multiple modules
bnb32
committed

Commits on May 10, 2026

Commits on May 11, 2026

Commits on May 12, 2026

fix: simplify loss_details dictionary comprehension to improve readability
bnb32
committed

Commits on May 13, 2026