Skip to content
Snippets Groups Projects

Repository graph

You can move around the graph by using the arrow keys.
Select Git revision
  • feat/adaptive-threshold-mechanism
  • main default protected
  • refactoring
3 results
Created with Raphaël 2.2.026Sep191029Aug282722212024Jul231817161511108432129Jun2827211913631May2916228Apr2624191126Mar151326Feb23222013105231Jan302826252420191817121110986227Dec22211918131265424Nov1615141311109876330Oct262219121165229Sep2827262522201918151276430Aug28252423227131Jul28Improved legend position in plottingrefactoringrefactoringLinked Wiki to README.mdmainmainAdded run averaging for plotting accuracy over timeRefactored device module: split up into one module per class, added documentationMerge branch 'fix-autoencoder' into 'main'Merge branch 'simulate_parallelism' into 'main'Merge branch 'fix-lr-schedule' into 'main'Renamed adaptive_learning_threshold to adaptive_threshold_valueRemoved printing optimizer stateAdded adaptive threshold for SwarmSL. Fixed setting the optimizer state for SwarmSL and PSLUpdated result generation notebookAdapted smart controller FLOP estimation to account for a potential AEAdjusted FLOP estimation to account for different parts of the model used in forward and backward passExcluded zero-gradients of AE from averaging in PSLFixed autoencoder: skipping AE layers during backpropPostprocessing fixfeat: add config for simulated parallelismExperiment postprocessing and plotting for simulated parallelismAdjusted experiment setupSimulating parallel execution by running everything sequentially while measuring the time and computing the expected parallel timeExperiment setupFixed smart controller bugexperiment setupsAvoid common prefixes at model loading for testingAdded lr scheduler for PSL clientsAdjusted lr scheduler milestonesCall LR scheduler before training in case a device misses out a roundMerge branch 'fix-cuda-error-when-loading-weights' into 'main'Fixed cuda error by loading AE weights explicitly on CPUMerge branch 'fix_experiment_group_naming' into 'main'Removing points from experiment group names to fix model saving and loadingMerge branch 'psl-performance' into 'main'PSL performanceMerge branch 'wip' into 'main'Merge branch 'main' into 'wip'Merge branch 'fix/psl' into 'wip'Merge branch 'cifar10' into 'wip'fix (psl): increased robustness, added test casefix(psl): gradient sendingfix(psl): gradient accumulation and sending