Default GPU flavor -> rtx-pro-6000 (Blackwell)

Upgrade to PR#237 trainer (torch 2.9.1/cu128) + enable RTX PRO 6000 + 6h timeout (#1)

UI: flatten dataset groups (uniform shade) + file drop zone single input-shade + readable placeholders

UI: file-upload drop zone matches other components (readable text, no distinct gray)

Update README.md

UI tweaks: caption tips above the boxes; lighter info/placeholder text; Training-tab auto-picked note

Remove broken welcome splash (blank-page bug); drop incompatible rtx-pro-6000 flavor (Blackwell vs cu126 torch)

UI redesign: dark LTX-cinematic theme + HF citrus accent, glass cards, pill tabs, welcome splash

#2 Validate dataset button; GPU flavor auto-adapts to profile (+hint); collapse caption tips into info disclosure

#1 GPU flavor guidance: single-GPU fits only (drop multi-GPU) + flavor guide table

Rebrand UI to LTX-2.3 (it trains LTX-2.3 models)

Point SRC_BUCKET default at ltx-community after org transfer

#6 concise recommended-value hints (rank/alpha, batch, grad-accum, flavor, timeout)

#5 validation defaults: inference_steps 16->30; 'validate every N steps' field (default 250, clamped to <= steps)

#4 captioning guidance + optional trigger word (prepended to upload captions; Hub captions untouched)

#3 performance profile (Quality/Low-VRAM/Custom; sets rank+memory recipe; flavor suggests; preserve manual edits)

#2 resolution presets (editable dropdown, configs-grounded)

Mode-aware smart defaults (LR/Steps per mode; preserve manual edits via dirty flags)

Use huggingface_hub Python API for buckets+jobs (fixes Space hf-CLI import crash on bucket sync)

Tabs + Citrus theme; dataset source (upload or Hub dataset); per-clip .txt captions

OAuth login (no pasted tokens) + ai-toolkit-style UI + generic IC-LoRA (paired references, drop Canny)

Fix: let HF Spaces set host/port (remove hardcoded server_port=7861 that caused RUNTIME_ERROR)

Fix job script: apt libgl1 (opencv) + keep_last_n=-1 (avoid checkpoint double-save delete)

LTX-2 LoRA trainer Space (HF Jobs + buckets)

initial commit