开源权重本地模型
DeepSeek R1 Distill Qwen 14B
Reasoning-focused dense model. Long reasoning output can make generation feel slower than chat models.
- 参数量
- 14.77B
- 推荐量化
- Q4_K_M
- 权重估算
- 9.3GB
- 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 | 29.80GB | 1.61GB | 34.39GB |
| Q8_0 | 16.40GB | 1.61GB | 19.65GB |
| Q6_K | 12.60GB | 1.61GB | 15.47GB |
| Q5_K_M | 10.66GB | 1.61GB | 13.34GB |
| Q4_K_M | 9.30GB | 1.61GB | 11.84GB |
| Q3_K_M | 7.70GB | 1.61GB | 10.11GB |
| Q2_K | 4.85GB | 1.61GB | 7.26GB |