开源权重本地模型
Qwen 3 14B
General reasoning model comfortable on 16GB GPUs at 4-bit with moderate context.
- 参数量
- 14.768B
- 推荐量化
- Q4_K_M
- 权重估算
- 9.3GB
- 最大上下文
- 40K
默认上下文下的量化显存
| 量化 | 权重 | KV cache | 总显存 |
|---|---|---|---|
| FP16 | 29.80GB | 1.34GB | 34.12GB |
| Q8_0 | 16.40GB | 1.34GB | 19.38GB |
| Q6_K | 12.60GB | 1.34GB | 15.20GB |
| Q5_K_M | 10.66GB | 1.34GB | 13.07GB |
| Q4_K_M | 9.30GB | 1.34GB | 11.57GB |
| Q3_K_M | 7.70GB | 1.34GB | 9.84GB |
| Q2_K | 4.85GB | 1.34GB | 6.99GB |