What problem does it solve? Running large-scale ML training and inference requires dedicated GPU infrastructure, but provisioning, configuring, and managing cloud GPU instances involves complex API calls, SSH setup, storage management, and distributed training configuration that slows down ML workflows. ## Core Features & Use Cases - GPU Instance Management: Launch, monitor, and terminate on-demand GPU instances (B200, H100, A100, A10, V100) via the Lambda Cloud Python API or curl-based CLI. - Persistent Storage & Filesystems: Attach NFS filesystems at launch to persist datasets, checkpoints, and models across instance restarts. - Distributed Training & 1-Click Clusters: Configure multi-GPU DDP/FSDP training and multi-node Slurm clusters with InfiniBand for 16-512 GPU workloads. - Use Case: Fine-tune a 70B LLM by launching an 8x H100 instance with an attached filesystem, running accelerate-based training with checkpoints saved to persistent storage, then terminating the instance when done. ## Quick Start Ask the agent to launch a Lambda Labs GPU instance with your SSH key and attached filesystem, then connect via SSH to start training.
