flux-brandbloom step-budget A/B

Local evals with the official evaluator image (combined L2 = 0.25text_guided + 0.75no_text, lower is better). Tournament winner reference on this task: 0.01403.

Before

model combined L2 notes
flux-brandbloom-v2 0.01515 baseline (LR 8e-5, dropout 0.1, 140 steps)
flux-brandbloom-v3 0.01519 dropout 0.15 A/B (no gain)
flux-brandbloom-v4 0.01493 boss LR pair 5e-5/5e-6, dropout 0.1, 140 steps

After

model combined L2 notes
flux-brandbloom-v5 0.01405 deeper step budget (240 planned, deadline-capped), boss LR pair, dropout 0.1

v5 breakdown: text_guided=0.01253, no_text=0.01455. Step-band git change: kept (v5 better).

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support