MuSID-KD paper improves semantic-ID generative recommendation via distillation
The MuSID-KD paper proposes distillation objectives that transfer item-level scoring knowledge from autoregressive recommenders to parallel semantic-ID generative models, cutting inference cost while improving accuracy.
2026-10-26 ~ 2026-10-28 · 2 related posts
- Knowledge Distillation across Item and Code Spaces for Parallel Generative Recommendation — _reachsumit · 2026-10-26
- MuSID-KD paper distills semantic-ID generative recommenders by ranking impact — _reachsumit · 2026-10-28