Megatron-DeepSpeed
BLOOM Inference via DeepSpeed-Inference, Accelerate and DeepSpeed-ZeRO
#308
Merged
Overview
Commits
58
Changes
View On GitHub
Commits
  • hardcode the dtype depending on the model
    stas00 committed 3 years ago
  • change the mp based on the world_size
    Reza Yazdani committed 3 years ago
  • remove hardcoded world_size
    stas00 committed 3 years ago
  • add bigscience/bigscience-small-testing
    stas00 committed 3 years ago
  • Merge branch 'bloom-inference' of https://github.com/bigscience-workshop/Megatron-DeepSpeed into bloom-inference
    Reza Yazdani committed 3 years ago
  • fixes
    stas00 committed 3 years ago
  • add zero-inference script
    stas00 committed 3 years ago
  • fixes
    stas00 committed 3 years ago
  • fix
    stas00 committed 3 years ago
  • working script
    stas00 committed 3 years ago
  • renames
    stas00 committed 3 years ago
  • fixes
    stas00 committed 3 years ago
  • fix for offline use
    stas00 committed 3 years ago
  • add benchmark
    stas00 committed 3 years ago
  • add benchmark
    stas00 committed 3 years ago
  • update
    stas00 committed 3 years ago
  • cleanup
    stas00 committed 3 years ago
  • update
    stas00 committed 3 years ago
  • msecs
    stas00 committed 3 years ago
  • cleanup
    stas00 committed 3 years ago
  • improve
    stas00 committed 3 years ago
  • fix benchmark, add warmup
    stas00 committed 3 years ago
  • update
    stas00 committed 3 years ago
  • fix; thanks Michael Wyatt
    stas00 committed 3 years ago
  • clarify
    stas00 committed 3 years ago
  • Merge branch 'bloom-inference' of https://github.com/bigscience-workshop/Megatron-DeepSpeed into bloom-inference
    Reza Yazdani committed 3 years ago
  • add bloom batch-inference script
    Reza Yazdani committed 3 years ago
  • removed the names :-)
    Reza Yazdani committed 3 years ago
  • fold the bs functionality from the other script
    stas00 committed 3 years ago
  • fix
    stas00 committed 3 years ago
  • restore do_sample
    stas00 committed 3 years ago
  • dump generate args
    stas00 committed 3 years ago
  • fix
    stas00 committed 3 years ago
  • fix
    stas00 committed 3 years ago
  • support any batchsize
    stas00 committed 3 years ago
  • div by bs
    stas00 committed 3 years ago
  • mul by bs
    stas00 committed 3 years ago
  • add cpu_offload; sync scripts
    stas00 committed 3 years ago
  • wip
    stas00 committed 3 years ago
  • improvements
    stas00 committed 3 years ago
  • fixes
    stas00 committed 3 years ago
  • fixes
    stas00 committed 3 years ago
  • add accelerate script
    stas00 committed 3 years ago
  • fix
    stas00 committed 3 years ago
  • wip
    stas00 committed 3 years ago
  • wip
    stas00 committed 3 years ago
  • stats
    stas00 committed 3 years ago
  • add OnDevice and remove zero-inference (#316)
    jeffra committed 3 years ago
  • wip
    stas00 committed 3 years ago
  • rework generate + benchmark
    stas00 committed 3 years ago
  • figure out the memory map dynamically
    stas00 committed 3 years ago
  • bug fix
    stas00 committed 3 years ago
  • fix ds-zero-inference wrt device
    stas00 committed 3 years ago
  • bug fix
    stas00 committed 3 years ago
  • update
    stas00 committed 3 years ago
  • update
    stas00 committed 3 years ago
  • fix
    stas00 committed 3 years ago
  • Merge remote-tracking branch 'origin/main' into bloom-inference
    stas00 committed 3 years ago
FAQ Terms Privacy Refunds Impressum
Loading