All news
3 articles- blogMeta Engineering·19 Oct 20264.4Meta builds a stable token vocabulary for ads so its generative recommenders can reason over ad content without ID churn.
- paperarXiv (systems, data, ML)·23 Oct 20264.1Quantizes n-gram embedding tables in open LLMs to low bits.
- paperarXiv (systems, data, ML)·22 Oct 20264.1Compresses per-layer embeddings with cross-layer latent vector quantization.