开源权重本地模型
Google Gemma 3 4B IT
Compact multimodal option. Text estimates exclude image encoder and vision activation peaks.
- 参数量
- 4.3B
- 推荐量化
- Q4_K_M
- 权重估算
- 3.3GB
- 最大上下文
- 128K
默认上下文下的量化显存
| 量化 | 权重 | KV cache | 总显存 |
|---|---|---|---|
| FP16 | 8.80GB | 1.14GB | 10.82GB |
| Q8_0 | 5.10GB | 1.14GB | 7.04GB |
| Q6_K | 3.67GB | 1.14GB | 5.61GB |
| Q5_K_M | 3.10GB | 1.14GB | 5.04GB |
| Q4_K_M | 3.30GB | 1.14GB | 5.24GB |
| Q3_K_M | 2.80GB | 1.14GB | 4.74GB |