# Extreme Compression of Per-Layer Embeddings via Cross-Layer Latent Vector Quantization

> Source: <https://aiflash.com/news/140227/>
> Published: 2026-10-27 03:01:09+00:00

Scaling embedding capacity alongside Transformer computation is an emerging trend in language models. This trend makes high compression rates increasingly useful. Hence, we study extreme compression of Per-Layer Embeddings (PLE), which follow this trend and account for 35.2% of the parameters of Gem