From a564c984186b9bba8128490f3a12ce94e9696b5b Mon Sep 17 00:00:00 2001 From: Dr Amy Krause Date: Mon, 21 Oct 2024 13:37:06 +0000 Subject: [PATCH] Update README.md --- README.md | 35 ++++++++++++++++++++++++++++++++++- 1 file changed, 34 insertions(+), 1 deletion(-) diff --git a/README.md b/README.md index fafc70e..5c6ad65 100644 --- a/README.md +++ b/README.md @@ -24,12 +24,45 @@ The kubeconfig for the remote Kubernetes cluster must be available in the user s Launch a new pod: After launch a new Jupyter kernel is created in the container and utilities required for remote execution are transferred to the pod. The image for the pod must run Jupyter at launch. This could be a Jupyter Docker stack image (see https://jupyter-docker-stacks.readthedocs.io/en/latest/using/selecting.html) or an extension of those. -Example with Pytorch CUDA 12 in namespace eidfxxxns on the GPU Service and pod name prefix `amy-`. +Example with Pytorch CUDA 12 in namespace eidfxxxns on the GPU Service and pod name prefix `amy-`. By default, it requests one GPU of type `NVIDIA-A100-SXM4-40GB-MIG-1g.5gb`. ``` %kube_exec launch -k /path/to/kube/config -ns eidfxxxns -pp amy- -i quay.io/jupyter/pytorch-notebook:cuda12-latest ``` +## Commands and options + +### `launch` + +Launch a new pod in the Kubernetes cluster. + +* `-k/--kubeconfig`: Path to the kubeconfig (default is `.kube/config`) +* `-ns/--namespace`: Kubernetes namespace +* `-i/--image`: Container image +* `-pp/--pod-prefix`: Pod name prefix (a UUID is appended, optional) +* `-g/--gpus`: Number of GPUs +* `-gp/--gpu-product`: GPU product, currently available: + * `NVIDIA-A100-SXM4-40GB-MIG-1g.5gb` (default) + * `NVIDIA-A100-SXM4-40GB-MIG-3g.20gb` + * `NVIDIA-A100-SXM4-40GB` + * `NVIDIA-A100-SXM4-80GB` + * `NVIDIA-H100-80GB-HBM3` + +### `connect` + +Connect to an existing pod. + +* `-p/--pod-name`: Name of a running pod to connect to +* `-ns/--namespace`: Kubernetes namespace +* `-k/--kubeconfig`: Path to the kubeconfig + +### `list` + +List pods in the namespace. + +* `-ns/--namespace`: Kubernetes namespace +* `-k/--kubeconfig`: Path to the kubeconfig + ## Re-use container Connect to an existing container. A Jupyter server must be running on this container. -- GitLab