lambda-labs-gpu-cloud

Provision on-demand GPU cloud instances for ML training by using the Lambda Cloud API to launch, monitor, and terminate instances with pre-installed ML stacks. You can configure SSH keys and attach persistent storage for data and checkpoints.

Yes, you can launch multi-node GPU clusters for distributed ML workloads using the 1-Click Clusters feature. This allows scalable training across Lambda Labs regions with pre-configured SSH access and persistent filesystems.

Yes, Lambda Labs GPU cloud supports persistent storage by allowing you to attach filesystems. This ensures your data, model checkpoints, and outputs persist across sessions even after terminating instances.

Connect to a Lambda Labs GPU instance via SSH by configuring SSH keys through the Lambda Cloud API. This provides secure, direct command-line administration for your training and inference workflows.

Available GPUs for on-demand ML inference and training include B200, H100, GH200, A100, A10, A6000, and V100. These instances come with a pre-installed ML stack featuring PyTorch, CUDA, cuDNN, and NCCL.

Yes, you need the lambda-cloud-client dependency to interact with the Lambda Cloud API. It enables you to programmatically launch, monitor, and terminate GPU instances, attach filesystems, and configure SSH access.