Code

support deepspeed#1101

Code

Merged

kohya-ss merged 12 commits intokohya-ss:deep-speedkohya-ss/sd-scripts:deep-speedfrom BootsofLagrangian:deepspeedBootsofLagrangian/sd-scripts:deepspeedCopy head branch name to clipboardFeb 27, 2024

+233-56

Commits

Commits on Feb 3, 2024

• support deepspeed

  BootsofLagrangian

  committed
  dfe08f3

  Copy full SHA for dfe08f3

Commits on Feb 5, 2024

• fix offload_optimizer_device typo

  BootsofLagrangian

  committed
  64873c1

  Copy full SHA for 64873c1

• fix vae type error during training sdxl

  BootsofLagrangian

  committed
  2824312

  Copy full SHA for 2824312

• fix all trainer about vae

  BootsofLagrangian

  committed
  4295f91

  Copy full SHA for 4295f91

• maybe fix branch to run offloading

  BootsofLagrangian

  committed
  3970bf4

  Copy full SHA for 3970bf4

• apply offloading method runable for all trainer

  BootsofLagrangian

  committed
  7d2a926

  Copy full SHA for 7d2a926

Commits on Feb 7, 2024

• fix full_fp16 compatible and train_step

  BootsofLagrangian

  committed
  6255661

  Copy full SHA for 6255661

• remove test requirements

  BootsofLagrangian

  committed
  2445a5b

  Copy full SHA for 2445a5b

• forgot setting mixed_precision for deepspeed. sorry

  BootsofLagrangian

  committed
  a98feca

  Copy full SHA for a98feca

Commits on Feb 9, 2024

• the reason not working grad accum steps found. it was becasue of my accelerate settings

  BootsofLagrangian

  committed
  03f0816

  Copy full SHA for 03f0816

Commits on Feb 22, 2024

• refactored codes, some function moved into train_utils.py

  BootsofLagrangian

  committed
  4d5186d

  Copy full SHA for 4d5186d

Commits on Feb 27, 2024

• Merge branch 'deep-speed' into deepspeed

  kohya-ss

  authored
  eefb3cc

  Copy full SHA for eefb3cc