Ollama安装#

先装个Ollama

1
curl -fsSL https://ollama.com/install.sh | sh

还是先下载下来吧

1
cd /opt
2
proxychains curl -fsSL https://ollama.com/install.sh -o ollama_install.sh
3
#这里一定要加sudo，不然useradd会未找到
4
#这里需要代理，因为下载Linux bundle太慢了
5
sudo proxychains bash ollama_install.sh

坐等
提示出

1
>>> The Ollama API is now available at 127.0.0.1:11434.
2
>>> Install complete. Run "ollama" from the command line.

就ok了
注意，下面几个连续操作都是改变ollama.service，你可以一次配完

开放一下api#

1
nano /etc/systemd/system/ollama.service
2
#在[Service]里面加
3
Environment="OLLAMA_HOST=0.0.0.0"
4
Environment="OLLAMA_ORIGINS=*"
5
#重启一下服务
6
systemctl daemon-reload
7
systemctl restart ollama
8
systemctl status ollama

修改模型路径#

1
chown -R ollama:ollama /mnt/model
2
chmod -R 775 /mnt/model
3

4
#在[Service]里面加
5
Environment="OLLAMA_MODELS=/mnt/model"

添加一下代理#

1
#在[Service]里面加
2
Environment="https_proxy=192.168.40.1:7890"

保持加载#

1
#在[Service]里面加
2
Environment="OLLAMA_KEEP_ALIVE=-1"

上面的方法保持效果不好，用下面的

1
nano /etc/systemd/system/ollama-run.service
2
[Unit]
3
Description=Ollama Run Service
4
After=network.target
5

6
[Service]
7
ExecStart=/usr/local/bin/ollama run qwq
8
Restart=always
9
User=root
10
WorkingDirectory=/root
11
Environment="PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin"
12

13
[Install]
14
WantedBy=multi-user.target
15

16
systemctl daemon-reload
17
systemctl enable --now ollama-run

安装模型#

1
ollama pull qwq:32b

1
curl http://localhost:11434/api/generate -d '{"model": "qwq", "keep_alive": -1}'

ok，pull下来的话就可以使用了

调整layer#

1
#在[Service]里面加
2
Environment="OLLAMA_GPU_LAYER=cuda"
3
Environment="CUDA_VISIBLE_DEVICES=GPU-6b53d422-019c-11f0-b3ec-82c4d041ae01"