A three-term loss (initial instantaneous, initial average, and terminal velocity matching) distills flow matching models into one-evaluation two-timed flow models, outperforming LFMD, EFMD, and PID baselines on most tested datasets.
We tried a student architecture with such layers, but we found that samples quality degraded significantly
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Distilling Two-Timed Flow Models by Separately Matching Initial and Terminal Velocities
A three-term loss (initial instantaneous, initial average, and terminal velocity matching) distills flow matching models into one-evaluation two-timed flow models, outperforming LFMD, EFMD, and PID baselines on most tested datasets.