DeepSpeed
Checks for None tensors and skip them when splitting the buckets in zero stage 2.
#728
Merged
Overview
Commits
48
Changes
View On GitHub
Commits
  • add flops count profiler
    cli99 committed 5 years ago
  • add run script for the small test model
    cli99 committed 5 years ago
  • fix flops sum and add batch counter
    cli99 committed 5 years ago
  • remove hook handles
    cli99 committed 5 years ago
  • add conv2d flops
    cli99 committed 5 years ago
  • add flops compute for major functionals and rnn modules
    cli99 committed 5 years ago
  • work
    cli99 committed 5 years ago
  • fix flops count cal in post hook
    cli99 committed 5 years ago
  • add flops of embedding and dropout as 0
    cli99 committed 5 years ago
  • fix
    cli99 committed 5 years ago
  • add duration and throughput
    cli99 committed 5 years ago
  • add time and throughput
    cli99 committed 5 years ago
  • add basic tracer for wall clock time breakdown
    cli99 committed 5 years ago
  • Added top module info summary
    cli99 committed 5 years ago
  • refactor
    cli99 committed 5 years ago
  • refactor and add readme
    cli99 committed 5 years ago
  • reorg folders
    cli99 committed 5 years ago
  • update readme
    cli99 committed 5 years ago
  • fix xsp init import
    cli99 committed 5 years ago
  • update readme and rename batch to step
    cli99 committed 5 years ago
  • fix multiple steps calc
    cli99 committed 5 years ago
  • update readme
    cli99 committed 5 years ago
  • rename pytorch-profiler to flops-profiler
    cli99 committed 5 years ago
  • rename pytorch-profiler to flops-profiler
    cli99 committed 5 years ago
  • fix steps calc and update readme
    cli99 committed 5 years ago
  • update readme
    cli99 committed 5 years ago
  • update ds
    cli99 committed 5 years ago
  • remove tracer code
    cli99 committed 5 years ago
  • fix incorrect merging
    cli99 committed 5 years ago
  • fix formatting
    cli99 committed 5 years ago
  • fix
    cli99 committed 5 years ago
  • fix sort value divided by steps
    cli99 committed 5 years ago
  • print info when steps is 0
    cli99 committed 5 years ago
  • Merge branch 'master' into cheng/flops_profiler
    jeffra committed 5 years ago
  • add flops profiler class and unit test
    cli99 committed 5 years ago
  • delete examples
    cli99 committed 5 years ago
  • add reload functionals in end profile
    cli99 committed 5 years ago
  • use config dict for flops profiler config
    cli99 committed 5 years ago
  • add doc strings and update readme
    cli99 committed 5 years ago
  • Merge branch 'master' into cheng/flops_profiler
    cli99 committed 5 years ago
  • Merge branch 'master' into cheng/flops_profiler
    cli99 committed 5 years ago
  • Merge branch 'master' into cheng/flops_profiler
    cli99 committed 5 years ago
  • Merge remote-tracking branch 'upstream/master' into master
    cli99 committed 5 years ago
  • Merge remote-tracking branch 'upstream/master'
    cli99 committed 5 years ago
  • Merge remote-tracking branch 'upstream/master' into master
    cli99 committed 5 years ago
  • check none tensors when splitting buckets
    cli99 committed 5 years ago
  • Merge branch 'master' into fix_zero2_none_tensor
    cli99 committed 5 years ago
  • Merge branch 'master' into fix_zero2_none_tensor
    cli99 committed 5 years ago
FAQ Terms Privacy Refunds Impressum
Loading