Luis Rojas 080b06e183 Training Step 3 revisions (ClickUp 868k1qape) ...

Labels/copy:
- Drop "[v1.0]" from Anima label; remove redundant "Preview —" prefix from
  Anima/HiDream O1 descriptions (the experimental-build Note already says it).
- Remove duplicate ecosystem "New" badge; capitalize "This" in the Note.

Step 3 sample prompts:
- Completion progress badge: "X/Y ready", and "Required · X/Y ready" when incomplete.
- Auto-sizing prompt textareas.
- Move sample CFG / LoRA strength into a nested "Advanced sample settings" collapsible.
- Repopulate prompts on a direct Step-3 landing: prefer persisted prompts, else
  fetch captions on demand from the training-data zip (shared helper).

Training params:
- Click-triggered info bubbles (was hover); convert per-row tooltips to bubbles.
- Always-visible Steps + Checkpoints in a separated-accordion card matching
  "Training Parameters"; Steps info icon on the label line, "seen ~X times" subtext below.
- "Checkpoints" (count, default 10) replaces "Save every"; steps default = 10× dataset
  size floored at the ecosystem minimum.
- Gate advanced params behind a refund-forfeit acknowledgment checkbox.

Continuing-from-epoch:
- Hide Model Details, Base Model selection, Engine selection, and Multi-training;
  keep Steps + Checkpoints editable.

Other:
- Base-model change pop-confirm when training params are dirty.
- Fix cropped Create button icon on the epoch results page (GenerateButton py).
- Add orchestrator-data-fetch design doc + Step 3 design mockup.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

2026-06-17 16:49:02 -04:00

..

orchestrator-data-fetch-design.md

Training Step 3 revisions (ClickUp 868k1qape)

2026-06-17 16:49:02 -04:00