开源权重本地模型
Google Gemma 3 12B IT
Balanced multimodal model. Text estimates exclude image encoder and vision activation peaks.
- 参数量
- 12B
- 推荐量化
- Q4_K_M
- 权重估算
- 8.2GB
- 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 | 24.50GB | 1.61GB | 28.56GB |
| Q8_0 | 13.50GB | 1.61GB | 16.46GB |
| Q6_K | 10.24GB | 1.61GB | 12.87GB |
| Q5_K_M | 8.66GB | 1.61GB | 11.14GB |
| Q4_K_M | 8.20GB | 1.61GB | 10.63GB |
| Q3_K_M | 6.80GB | 1.61GB | 9.21GB |