{"slug": "extreme-compression-of-per-layer-embeddings-via-cross-layer-latent-vector", "title": "Extreme Compression of Per-Layer Embeddings via Cross-Layer Latent Vector Quantization", "summary": "A research paper proposes cross-layer latent vector quantization to achieve extreme compression of Per-Layer Embeddings (PLE), which account for 35.2% of the parameters of Gemma, according to the paper's authors. The work targets the growing parameter cost of scaling embedding capacity alongside Transformer computation, a trend the authors say makes high compression rates increasingly useful. The paper's headline result is the 35.2% parameter share held by PLE in Gemma.", "body_md": "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", "url": "https://wpnews.pro/news/extreme-compression-of-per-layer-embeddings-via-cross-layer-latent-vector", "canonical_source": "https://aiflash.com/news/140227/", "published_at": "2026-10-27 03:01:09+00:00", "updated_at": "2026-10-27 03:13:10.163324+00:00", "lang": "en", "topics": ["machine-learning", "large-language-models", "ai-research", "natural-language-processing"], "entities": ["Gemma", "Per-Layer Embeddings"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/extreme-compression-of-per-layer-embeddings-via-cross-layer-latent-vector", "markdown": "https://wpnews.pro/news/extreme-compression-of-per-layer-embeddings-via-cross-layer-latent-vector.md", "text": "https://wpnews.pro/news/extreme-compression-of-per-layer-embeddings-via-cross-layer-latent-vector.txt", "jsonld": "https://wpnews.pro/news/extreme-compression-of-per-layer-embeddings-via-cross-layer-latent-vector.jsonld"}}