v0.9.3.9

版本发布时间: 2024-08-27 21:50:20

bghira/SimpleTuner最新发布版本:v1.1.1(2024-10-05 08:37:33)

What's Changed

lycoris model card updates by @bghira in https://github.com/bghira/SimpleTuner/pull/820
lycoris: resume should use less vram now by @bghira in https://github.com/bghira/SimpleTuner/pull/835
update lycoris to dev branch, 3.0.1dev3 by @bghira in https://github.com/bghira/SimpleTuner/pull/828
(#644) temporarily block training on multi-gpu setup with quanto + PEFT, inform user to go with lycoris instead by @bghira in https://github.com/bghira/SimpleTuner/pull/837

Generate and store attention masks for T5 for flux by @AmericanPresidentJimmyCarter in https://github.com/bghira/SimpleTuner/pull/821
backwards-compatible flux embedding cache masks by @bghira in https://github.com/bghira/SimpleTuner/pull/823
remove clip warning by @bghira in https://github.com/bghira/SimpleTuner/pull/827
fix quanto resuming by @bghira in https://github.com/bghira/SimpleTuner/pull/834
(#864) add flux final export for full tune by @bghira in https://github.com/bghira/SimpleTuner/pull/871
flux: mobius-style training via augmented guidance scale by @bghira in https://github.com/bghira/SimpleTuner/pull/880
track flux cfg via wandb by @bghira in https://github.com/bghira/SimpleTuner/pull/881

add schedulefree optim w/ kahan summation by @bghira in https://github.com/bghira/SimpleTuner/pull/857
schedulefree: return to previous stable settings and add a new preset for aggressive training by @bghira in https://github.com/bghira/SimpleTuner/pull/862
(#519) add side by side comparison with base model by @bghira in https://github.com/bghira/SimpleTuner/pull/865
wandb gallery mode by @bghira in https://github.com/bghira/SimpleTuner/pull/872
combine the vae cache clear logic with bucket rebuild logic by @bghira in https://github.com/bghira/SimpleTuner/pull/879

parquet add width and height columns by @frankchieng in https://github.com/bghira/SimpleTuner/pull/825
quanto: remove warnings about int8/fp8 confusion as it happened so long ago now; add warning about int4 by @bghira in https://github.com/bghira/SimpleTuner/pull/826
Fix caption_with_blip3.py on CUDA by @anhi in https://github.com/bghira/SimpleTuner/pull/833
quanto + deepspeed minor fixes for multigpu training by @bghira in https://github.com/bghira/SimpleTuner/pull/839
deepspeed sharding fixes by @bghira in https://github.com/bghira/SimpleTuner/pull/840
fix: only run save full model on main process by @ErwannMillon in https://github.com/bghira/SimpleTuner/pull/838
(#851) remove shard merge code on load hook by @bghira in https://github.com/bghira/SimpleTuner/pull/853
csv backend updates by @williamzhuk in https://github.com/bghira/SimpleTuner/pull/645
csv fixes by @bghira in https://github.com/bghira/SimpleTuner/pull/856
fix validation image filename only using resolution from first img, and, unreadable/untypeable parenthesis by @bghira in https://github.com/bghira/SimpleTuner/pull/863
sdxl: dtype inference followup fix by @bghira in https://github.com/bghira/SimpleTuner/pull/873

Full Changelog: https://github.com/bghira/SimpleTuner/compare/v0.9.8.3.2...v0.9.3.9