Gemma 3n
An on-device-oriented derivative model of Gemma released by Google, which introduces per-layer embeddings (PLE) and an audio modality for the first time.
Timeline (last 90 days)
The audio modality first appeared in Gemma 3n and in Gemma 4 is extended to the E2B, E4B and 12B models, supporting more than 35 languages out of the box.
Per-Layer Embedding (PLE) is a parameter-efficient architecture first introduced by Google in Gemma 3n; it reads an additional shared embedding table at each layer, decoupling nominal parameter count from active parameter count.
All Facts (2)
The audio modality first appeared in Gemma 3n and in Gemma 4 is extended to the E2B, E4B and 12B models, supporting more than 35 languages out of the box.
50%UnverifiedPer-Layer Embedding (PLE) is a parameter-efficient architecture first introduced by Google in Gemma 3n; it reads an additional shared embedding table at each layer, decoupling nominal parameter count from active parameter count.
50%