VaLiDRec uses variable-length LLM-aligned IDs and runs 87.49× faster than LC-Rec

_reachsumit · x · 2026-07-29

VaLiDRec introduces variable-length semantic IDs built from native LLM vocabulary tokens, then predicts them with graph-aware soft prompts instead of autoregressive SID decoding.

Original post →

More from Infra

Infra channel →