Version 0.12.0

What's Changed

• fix: restore tags-only caption handling in FineTuningDataset with enable_wildcard by @kohya-ss in #2389
• fix: respect caption_separator when rejoining shuffled caption tokens by @kohya-ss in #2390
• Fix caption token rejoining to respect custom caption_separator by @kohya-ss in #2391
• [Fix] lora_flux: ValueError unpacking get_qkv_mlp_split_dims()[:3] on qkv-only path (#2220) by @yushuosun in #2396
• fix: enhance weight renaming logic for anima model loading by @kohya-ss in #2404
• Update IPEX libs by @WhitePr in #2406
• Fix typos and exclude vendored ipex directory from typos check by @kohya-ss in #2408
• Update IPEX libraries and fix typos in code by @kohya-ss in #2409
• feat: add per-subset timestep_bias for semantic-aware timestep sampling by @okdsf in #2401
• Add --show_timesteps_offset and document offset behavior under shift/flux_shift by @kohya-ss in #2410
• Add per-subset timestep sampling offset and update documentation by @kohya-ss in #2411
• OFTv2 / BOFT のネットワークモジュールを追加 by @umisetokikaze in #2357
• OFTv2/BOFT follow-up: direct model_io import, docs, README by @kohya-ss in #2427
• Add OFTv2 and BOFT network modules by @kohya-ss in #2428
• Make opencv-python an optional dependency (Windows on ARM64 / RTX Spark support) by @kohya-ss in #2430
• Bump transformers/schedulefree/safetensors for Windows on ARM64 wheels, make dadaptation optional in tests by @kohya-ss in #2431
• Sync requirements-no-opencv.txt with the dependency bumps from #2431 by @kohya-ss in #2432
• Merge requirements-no-opencv.txt into requirements.txt with environment markers (Windows on ARM64 / RTX Spark support) by @kohya-ss in #2433
• Fix tokenizer sources that are no longer accessible on the Hugging Face Hub (SD2, Lumina) by @kohya-ss in #2435
• Add local-only regression harness for transformers / diffusers upgrades (tests/local) by @kohya-ss in #2434
• Update transformers to 5.5.4, diffusers to 0.40.0, accelerate to 1.15.0, huggingface-hub to 1.32.0 by @kohya-ss in #2436
• Support transformers 5.6+ (CLIPTextModel layout wrapper, eager T5 attention) by @kohya-ss in #2437
• Fix the dpmsolver / dpmsingle samplers and give a clear error for lms without scipy by @kohya-ss in #2438
• Remove gen_img_diffusers.py, the old SD1.x / SD2.x image generation script by @kohya-ss in #2439
• docs: Add version 0.12.0 to the change history by @kohya-ss in #2440
• Version 0.12.0 by @kohya-ss in #2441

New Contributors

• @yushuosun made their first contribution in #2396
• @okdsf made their first contribution in #2401

Full Changelog: v0.11.1...v0.12.0