0.57721566490

What's Changed

• [FIX] render wildcards for each generation in batch + add metadata to png by @s1dlx in #83
• [feat] Latin Hypercube Sampling by @s1dlx in #84
• no classifiers + weighted prompt score + scorer_device by @s1dlx in #85
• Move merge functions to meh package by @s1dlx in #79
• feat(requirements.txt): add openai-clip package to requirements by @axsddlr in #88
• fix: merge on cfg.device by @ljleb in #90
• feat: Merge models in-memory by @ljleb in #91
• feat: rebasin by @s1dlx in #93

Full Changelog: 0.577215664...0.57721566490

0.577215664

What's Changed

• refactor(bounds.py): simplify get_value method and add None check for… by @axsddlr in #66
• Implement TensorSum by @mariaWitch in #54
• User score in image name + catch missing guides by @s1dlx in #74
• Initial Implementation of Adaptive TPE by @mariaWitch in #73

New Contributors

• @axsddlr made their first contribution in #66

Full Changelog: 0.57721566...0.577215664

0.57721566

What's Changed

• Group blocks by @s1dlx in #53
• Manual scorer method by @hoblin in #59
• Add weighted_subtraction method by @bmaltais in #64
• Fix groups by @s1dlx in #65

New Contributors

• @hoblin made their first contribution in #59
• @bmaltais made their first contribution in #64

Full Changelog: 0.5772156...0.57721566

0.5772156

What's Changed

• Remove unused alias PathT from scorer.py by @wbclark in #38
• Fix a payload parameter by @jnyfil in #48
• Fix merge by @s1dlx in #50
• Guided optimization by @s1dlx in #51

New Contributors

• @wbclark made their first contribution in #38
• @jnyfil made their first contribution in #48

Full Changelog: 0.577215...0.5772156

0.577215

What's Changed

• More merging methods by @s1dlx in #27

closing #14 by adding support for

• add_difference
• triple_sum: related to #21
• sum_twice

Also:

• fixing and closing #33
• add bounds_transformer
• update logging and plotting with beta params

Full Changelog: 0.57721...0.577215

0.57721

What's Changed

• Generate multiple images for each api call by @s1dlx in #26

In particular:

• switch to hydra for config management
• remove click cli
  these two let us define easily all the parameters in the payload: closing #24

also

• move docs to wiki
• keep best merge and log weights while running: closing #31

Full Changelog: 0.5772...0.57721

0.5772

What's Changed

• switch to hydra a single .ini config (hydra is great but overkill for this project) source
• (optional) save generated images
• print best base_alpha and weights in a copypastable format at the end of the run
• random bugfixes

closes #20

Full Changelog: 0.577...0.5772

0.577

What's Changed

• More scorers by @s1dlx in #12

as discussed in #6

• AestheticScorer abstract class
• all chad models:
  • sac+logos+ava1-l14-linearMSE.pth
  • ava+logos-l14-linearMSE.pth
  • ava+logos-l14-reluMSE.pth
• one laion model:
  • laion-sac-logos-ava-v2.safetensors
• one aesthetic model:
  • aes-B32-v0.safetensor
• cafe_aesthetic models:
  • cafe_aesthetic
  • cafe_style
  • cafe_waifu

as suggested in #10

• add install.py

Full Changelog: 0.57...0.577

0.57

What's Changed

Bugfix

• rename sampler -> sampler_name by @ljleb in #15

New functionalities

• Save best weights at the end of the run by @s1dlx in #8
• TPE optimiser by @JonnoFTW in #19

New Contributors

• @ljleb made their first contribution in #15
• @JonnoFTW made their first contribution in #7

Full Changelog: 0.5...0.57

0.5

The basic algorithm is implemented

• pick two models
• set some number of iterations
• run the optimiser
• get results