Gemma 4 E2B Architecture
Back to Gemma 4 E2B in the gallery
Smallest Gemma 4 edge model keeps the family's hybrid attention stack and adds native audio on a phone-scale multimodal footprint. Uses per-layer embeddings, which add small layer-specific token vectors without scaling the full compute path, so its compute footprint is closer to 2.3B than a full 5.1B dense model.
Architecture and specifications
- Scale
- 5.1B parameters (2.3B effective)
- Context (tokens)
- 131,072
- Vocabulary
- 262,144 (~262k)
- License
- Apache License 2.0
- Date
- 2026-04-02
- Decoder type
- Dense
- Attention
- Multi-query attention with QK-Norm, cross-layer KV sharing, p-RoPE on global layers, and 4:1 sliding-window/global attention
- Layer mix
- 28 sliding-window + 7 global
- KV cache / token (bf16) info
- 18 KiB · Very low
- Key detail
- The first 15 layers have a 6,144-wide GELU MLP. The final 20 reuse earlier KV tensors and double the MLP width to 12,288.
Training and benchmarks
- Pretraining data
- Not disclosed No per-model pretraining-token count is published. Gemma 4 model card
- Post-training methods
- Instruction tuning IT variant; the report refers to Gemma 3-like post-training without itemizing alignment stages. Gemma 4 report