DMAD: Distribution Matching as Adversarial Distillation for Fast Visual Generation
↑ +66 today
Discover the best Distillation open source repositories on GitHub. Explore trending tools, live star counts, source code, and developer projects updated daily.
DMAD: Distribution Matching as Adversarial Distillation for Fast Visual Generation
An 8M-parameter English text-to-speech model distilled from Kokoro-82M. 9 MB, runs 18x real time on one CPU thread.
A unified inference and post-training framework for accelerated video generation.
Harness-Zero: Harness Distillation via Agent-as-Harness