Architecture
Retrieved 2026-09-01 at pinned commit 62a522b2dcb8.
- Architecture
- gemma3_text
- Layers
- 62
- Hidden size
- 5,376
- Attention heads
- 32
- KV heads
- 16
- Head dimension
- 128
- Feed-forward width
- 21,504
- Vocabulary
- 262,208
- Context ceiling
- 131,072
- Sliding window
- 1,024
- RoPE theta
- 1,000,000
Where the memory goes
Grouped-query attention shares each key/value head across 2 query heads, so the KV cache is 50% of what multi-head attention would need at the same context.
This is a multimodal checkpoint (vision); the diagram and memory sizing above cover the text decoder. Encoder/audio/image activations are not included in the KV or weight figures.