Create a new fine-tuning instance, how to use local GPUs resources

[tianj0522]

[tianj0522]

version： v1.0.0

[Rader]

suppose you have managed GPUs with k8s, please copy the kube config file to .kube folder of service csghub_server_runner.

.kube folder mounted in docker compose yaml config:

  csghub_server_runner:
   
...
    volumes:
      - ./.kube:/root/.kube:r

[Rader]

resource list are read from table space_resources, change the config according to your real GPU instances.

[tianj0522]

资源列表从表中读取space_resources，根据您的实际 GPU 实例更改配置。

Are there any standards for modification? For example, the GPU instance used in my k8s cluster is NVIDIA-GeForce-RTX-4070 4 cards, which are managed uniformly using gpu-operator. After adding relevant information, it still cannot be selected. The space_resources table configuration is shown in the figure below