请教Qwen3.8模式部署的问题

手上有一台服务器，配置如下

显卡是4090*2

通过llama.cpp 部署的 Qwen3.8-27B-Uncensored-Q5_K_M.gguf

启动参数为：

ExecStart=/usr/local/bin/llama-server \

    -m /home/xxxx/ollama/qwen3.8/Qwen3.8-27B-Uncensored-Q5_K_M.gguf \

    --mmproj /home/xxxx/ollama/qwen3.8/mmproj-Qwen3.8-27B-Uncensored-f16.gguf \

    -a Qwen3.8-27B \

    --host 0.0.0.0 \

    --port 8888 \

    -ngl all \

    -np 1 \

    -c 262144 \

    -kvu \

    --temp 1.00 \

    --top-k 20 \

    --top-p 0.95 \

    --min-p 0.00 \

    --repeat-penalty 1.00 \

    --presence-penalty 0.00 \

    --frequency-penalty 0.00 \

    --spec-type draft-mtp \

    --spec-draft-n-max 1 \

    --load-mode mlock \

    --reasoning-preserve \

    --reasoning-effort medium

GPU内存占用：

+-----------------------------------------------------------------------------------------+

| NVIDIA-SMI 595.84                 Driver Version: 595.84         CUDA Version: 13.2     |

+-----------------------------------------+------------------------+----------------------+

| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |

| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |

|                                         |                        |               MIG M. |

|=========================================+========================+======================|

|   0  NVIDIA GeForce RTX 4090        Off |   00000000:02:00.0 Off |                  Off |

| 30%   48C    P0             82W /  450W |   20761MiB /  24564MiB |      7%      Default |

|                                         |                        |                  N/A |

+-----------------------------------------+------------------------+----------------------+

|   1  NVIDIA GeForce RTX 4090        Off |   00000000:03:00.0 Off |                  Off |

| 30%   35C    P8             21W /  450W |   21566MiB /  24564MiB |      0%      Default |

|                                         |                        |                  N/A |

+-----------------------------------------+------------------------+----------------------+



+-----------------------------------------------------------------------------------------+

| Processes:                                                                              |

|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |

|        ID   ID                                                               Usage      |

|=========================================================================================|

|    0   N/A  N/A            2374      G   /usr/lib/xorg/Xorg                        4MiB |

|    0   N/A  N/A          621097      G   /usr/bin/gnome-shell                    406MiB |

|    0   N/A  N/A          621387    C+G   ...c/gnome-remote-desktop-daemon        477MiB |

|    0   N/A  N/A          621481      G   /usr/bin/Xwayland                        11MiB |

|    0   N/A  N/A          779483      C   /usr/local/bin/llama-server           19732MiB |

|    0   N/A  N/A          793036      G   ...tigravity-ide/antigravity-ide         30MiB |

|    0   N/A  N/A          892515      G   ...y/Antigravity-x64/antigravity         11MiB |

|    1   N/A  N/A            2374      G   /usr/lib/xorg/Xorg                        4MiB |

|    1   N/A  N/A          779483      C   /usr/local/bin/llama-server           21536MiB |

+-----------------------------------------------------------------------------------------+

部署完的模型接入到sub2api，在opencode中使用，

opencode中的这个模型的配置如下：

 "Qwen3.8": {

    "name": "Qwen3.8",

    "thinking": true,

    "attachment": true,

    "limit": {

            "context": 262144,

            "output": 65535

    }

}

目前发现一个问题，每次会话即将要压缩，或者压缩的时候， opencode中就会出现超时, 服务器的风扇狂转， 请问一下各位佬是什么原因？哪里可以优化一下 ^-^

sub2api前面的nginx的错误如下：

upstream timed out (110: Connection timed out) while reading response header from upstream, client: 192.168.110.234, server: xx.xxxxxx.xx, request: "POST /sub2api/v1/chat/completions HTTP/1.1", upstream: "http://192.168.110.80:8080/v1/chat/completions", host: "xx.xxxxxx.xx:11443"