* docs: add XL (4B DiT) model documentation (#993)
Update English documentation to cover the new XL (4B) DiT models
(acestep-v15-xl-base, acestep-v15-xl-sft, acestep-v15-xl-turbo):
- README.md: Add XL models to Model Zoo table, update GPU recommendation
with DiT column and XL VRAM guidance
- INSTALL.md: Add XL download commands, available models table entries,
and update "Which Model Should I Choose?" with XL recommendations
- GPU_COMPATIBILITY.md: Add "XL (4B) DiT" column to tier table showing
support level per tier (❌/⚠️/✅)
- Tutorial.md: Add XL model section and expand DiT selection summary
with XL variants
Key facts documented:
- XL weights ~9GB (bf16) vs ~4.7GB for 2B
- Min VRAM: 16GB with offload, 20GB+ recommended
- All LM models (0.6B/1.7B/4B) fully compatible with XL
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: address Codex review suggestions for XL documentation
- Update min VRAM from ≥16GB to ≥12GB (offload makes 12GB viable)
- Fix 16-20GB tier note: "XL requires CPU offload below 20GB"
- GPU_COMPATIBILITY: clarify Tier 5 marginal note for 12-16GB range
- Tutorial: use full model names (acestep-v15-xl-*) instead of short
- INSTALL: fix comment dash style in download section
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* docs: add XL (4B DiT) model documentation in zh/ja/ko translations
Update translated documentation with XL model support:
- zh: INSTALL.md (model recommendation table), GPU_COMPATIBILITY.md
(tier table with XL column), Tutorial.md (XL section + summary)
- ja: INSTALL.md, GPU_COMPATIBILITY.md, Tutorial.md (same updates)
- ko: GPU_COMPATIBILITY.md, Tutorial.md (XL section + summary table added)
All translations consistent with English docs: ≥12GB VRAM with offload,
20GB+ recommended, all LM models compatible with XL.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: address remaining Codex review suggestions for XL docs
- ≥24GB tier: note xl-base for extract/lego/complete (en/zh/ja)
- Fix pre-existing Korean GPU tier batch size discrepancies:
Tier 3: 1/2 → 2/2, Tier 5: 2/4 → 4/4, Tier 6b: 4/8 → 8/8
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
