Gemma 4 E2B
vision
5,1B
Parameters
3,2 Go
Q4_K_M file
≥ 4,7 Go
Recommended VRAM
131 072
Max context
Speed by graphics card
Simulate in the “Configurator” →Standard llama-bench run (prompt 512 / generation 128 tokens), model fully in VRAM. Click a card for details. Electricity cost at the default price (0,25 €/kWh).
| Card · Q4_K_M | VRAM | Generation | Prompt | VRAM used | Power | € elec. / M tok | Price |
|---|---|---|---|---|---|---|---|
| RTX 4070 Ti | 12 Go | 198 tok/s | 12 570 | 2,3 Go | 100 W | 0,04 €/M tok | — |
Not measured on these cards yet — click for an estimate:
Technical sheet
- Published
- 2026-03-02
- Architecture
- 35 layers · 1 KV heads · dimension 256
- Attention
- Sliding window: 7 of 35 layers keep the whole context, the others only the last 512 tokens — a much lighter cache.
- Context cache (f16)
- ≈ 7 MB per 1,000 tokens · 0,2 Go for 32,768 tokens
- GGUF file
- bartowski/google_gemma-4-E2B-it-GGUF · google_gemma-4-E2B-it-Q4_K_M.gguf Download (3,2 Go)
- Official repo
- google/gemma-4-E2B-it
Public Hugging Face data (official config.json, GGUF repo). The context cache only counts full-attention layers.