ceos/unsloth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Align LR-schedule copy with the diffusion two-card chart layout

Browse source 
DiffusionCharts deliberately renders only Training Loss + Gradient Norm (the LR
curve is the deterministic schedule the user picked), but the settings copy and two
comments still promised a live LR chart. Reword them so the UI no longer references a
chart that was intentionally dropped.
This commit is contained in:
Daniel Han 2026-07-06 11:21:06 +00:00
commit a1114bfdc3
1 changed files with 3 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

6
studio/frontend/src/features/images/train/diffusion-train-panel.tsx
View file

@ -153,7 +153,7 @@ function mergeFamilies(reported?: DiffusionTrainableFamily[]): FamilyPreset[] {
}
// A full-page training workspace: left = configure (family, dataset, labeling, settings),
// right = live run (progress, loss/LR charts, completion + deploy). Kept mounted with the
// right = live run (progress, loss/grad-norm charts, completion + deploy). Kept mounted with the
// page so a long run survives Create/Train tab switches; polling is gated on `active`.
export function DiffusionTrainPanel({
active,
@ -235,7 +235,7 @@ export function DiffusionTrainPanel({
const [batchSize, setBatchSize] = useState(1);
const [gradAccum, setGradAccum] = useState(1);
const [seed, setSeed] = useState(42);
// LR schedule (PR E wired get_scheduler into the loop, so the LR chart reflects this).
// LR schedule (PR E wired get_scheduler into the loop, so the LR follows the chosen curve).
// Warmup only applies to the non-constant schedules; plain "constant" ignores it.
const [lrScheduler, setLrScheduler] = useState<
"constant" | "constant_with_warmup" | "cosine" | "linear"
@ -842,7 +842,7 @@ export function DiffusionTrainPanel({
<option value="linear">Linear decay</option>
</select>
<p className="text-[11px] leading-snug text-muted-foreground">
How the learning rate evolves over the run (shown live in the LR chart).
How the learning rate evolves over the run.
</p>
</div>
{lrScheduler !== "constant" &&