What problem does it solve?
Lambda Labs users need a reliable way to provision dedicated GPU instances (and persistent storage) for ML training and inference without spending time on low-level infrastructure setup.
Core Features & Use Cases
- Dedicated GPU instances with SSH: Provision single-GPU or multi-GPU machines for interactive development, long training jobs, and inference runs.
- Lambda Stack ML environment: Use pre-installed CUDA/NCCL and common ML frameworks (PyTorch, TensorFlow, JAX, JupyterLab) to reduce setup time.
- Persistent filesystems: Keep datasets, checkpoints, and outputs across instance restarts and reuse them for iterative workflows.
- 1-Click Slurm clusters: Run high-performance multi-node/distributed training with Slurm (16–512 GPUs) for large-scale experiments.
Use Case Example: Fine-tune an LLM on an 8x H100 setup, save checkpoints to a mounted filesystem for safe resumption, then run batch inference from the same persisted model artifacts.
Quick Start
Launch an H100 instance in the Lambda Labs console, attach (or create) a filesystem for persistent checkpoints, wait for it to become active, then connect via SSH using the instance IP shown in the dashboard.
