[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M74YWANDYWH8RG72VM7R55H2":3,"skill-same:01M74YWANDYWH8RG72VM7R55H2":68},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":30,"commitSha":31,"skillDirectory":32,"forksCount":33,"repositoryHtmlUrl":34,"repositoryHomepageUrl":35,"ownerHtmlUrl":36,"ownerBlogUrl":37,"ownerDescription":38,"updatedAt":39,"install":40,"resources":53},"01M74YWANDYWH8RG72VM7R55H2","lambda-labs-gpu-cloud",null,"Reserved and on-demand GPU cloud instances for ML training and inference. Use when you need dedicated GPU instances with simple SSH access, persistent filesystems, or high-performance multi-node clusters for large-scale training.","KalarisLabs\u002Fresearch-agent-skills","KalarisLabs","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F300878568?v=4","01M74Y7MT23GXXC56VM6TVSXZY","Kalarislabs","01M74Y7MTQM9REXZX4SNXWX95C",132,"MIT","https:\u002F\u002Fgithub.com\u002FKalarisLabs\u002Fresearch-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Flambda-labs-gpu-cloud\u002FSKILL.md",[],[],"2026-11-12 13:55:43",false,"---\nname: lambda-labs-gpu-cloud\ndescription: Reserved and on-demand GPU cloud instances for ML training and inference. Use when you need dedicated GPU instances with simple SSH access, persistent filesystems, or high-performance multi-node clusters for large-scale training.\nlicense: MIT\nmetadata:\n  version: 1.0.0\n  category: ml-inference-and-ops\n  maintainer: Kalaris Labs\n  tags: Infrastructure, GPU Cloud, Training, Inference, Lambda Labs\n  dependencies: lambda-cloud-client>=1.0.0\n---\n\n# Lambda Labs GPU Cloud\n\nComprehensive guide to running ML workloads on Lambda Labs GPU cloud with on-demand instances and 1-Click Clusters.\n\n## When to use Lambda Labs\n\n**Use Lambda Labs when:**\n- Need dedicated GPU instances with full SSH access\n- Running long training jobs (hours to days)\n- Want simple pricing with no egress fees\n- Need persistent storage across sessions\n- Require high-performance multi-node clusters (16-512 GPUs)\n- Want pre-installed ML stack (Lambda Stack with PyTorch, CUDA, NCCL)\n\n**Key features:**\n- **GPU variety**: B200, H100, GH200, A100, A10, A6000, V100\n- **Lambda Stack**: Pre-installed PyTorch, TensorFlow, CUDA, cuDNN, NCCL\n- **Persistent filesystems**: Keep data across instance restarts\n- **1-Click Clusters**: 16-512 GPU Slurm clusters with InfiniBand\n- **Simple pricing**: Pay-per-minute, no egress fees\n- **Global regions**: 12+ regions worldwide\n\n**Use alternatives instead:**\n- **Modal**: For serverless, auto-scaling workloads\n- **SkyPilot**: For multi-cloud orchestration and cost optimization\n- **RunPod**: For cheaper spot instances and serverless endpoints\n- **Vast.ai**: For GPU marketplace with lowest prices\n\n## Quick start\n\n### Account setup\n\n1. Create account at https:\u002F\u002Flambda.ai\n2. Add payment method\n3. Generate API key from dashboard\n4. Add SSH key (required before launching instances)\n\n### Launch via console\n\n1. Go to https:\u002F\u002Fcloud.lambda.ai\u002Finstances\n2. Click \"Launch instance\"\n3. Select GPU type and region\n4. Choose SSH key\n5. Optionally attach filesystem\n6. Launch and wait 3-15 minutes\n\n### Connect via SSH\n\n```bash\n# Get instance IP from console\nssh ubuntu@\u003CINSTANCE-IP>\n\n# Or with specific key\nssh -i ~\u002F.ssh\u002Flambda_key ubuntu@\u003CINSTANCE-IP>\n```\n\n## GPU instances\n\n### Available GPUs\n\n| GPU | VRAM | Price\u002FGPU\u002Fhr | Best For |\n|-----|------|--------------|----------|\n| B200 SXM6 | 180 GB | $4.99 | Largest models, fastest training |\n| H100 SXM | 80 GB | $2.99-3.29 | Large model training |\n| H100 PCIe | 80 GB | $2.49 | Cost-effective H100 |\n| GH200 | 96 GB | $1.49 | Single-GPU large models |\n| A100 80GB | 80 GB | $1.79 | Production training |\n| A100 40GB | 40 GB | $1.29 | Standard training |\n| A10 | 24 GB | $0.75 | Inference, fine-tuning |\n| A6000 | 48 GB | $0.80 | Good VRAM\u002Fprice ratio |\n| V100 | 16 GB | $0.55 | Budget training |\n\n### Instance configurations\n\n```\n8x GPU: Best for distributed training (DDP, FSDP)\n4x GPU: Large models, multi-GPU training\n2x GPU: Medium workloads\n1x GPU: Fine-tuning, inference, development\n```\n\n### Launch times\n\n- Single-GPU: 3-5 minutes\n- Multi-GPU: 10-15 minutes\n\n## Lambda Stack\n\nAll instances come with Lambda Stack pre-installed:\n\n```bash\n# Included software\n- Ubuntu 22.04 LTS\n- NVIDIA drivers (latest)\n- CUDA 12.x\n- cuDNN 8.x\n- NCCL (for multi-GPU)\n- PyTorch (latest)\n- TensorFlow (latest)\n- JAX\n- JupyterLab\n```\n\n### Verify installation\n\n```bash\n# Check GPU\nnvidia-smi\n\n# Check PyTorch\npython -c \"import torch; print(torch.cuda.is_available())\"\n\n# Check CUDA version\nnvcc --version\n```\n\n## Python API\n\nDetails, code examples and parameter tables: [references\u002Fpython-api.md](references\u002Fpython-api.md). Read it when this step applies.\n\n## CLI with curl\n\n### List instance types\n\n```bash\ncurl -u $LAMBDA_API_KEY: \\\n  https:\u002F\u002Fcloud.lambdalabs.com\u002Fapi\u002Fv1\u002Finstance-types | jq\n```\n\n### Launch instance\n\n```bash\ncurl -u $LAMBDA_API_KEY: \\\n  -X POST https:\u002F\u002Fcloud.lambdalabs.com\u002Fapi\u002Fv1\u002Finstance-operations\u002Flaunch \\\n  -H \"Content-Type: application\u002Fjson\" \\\n  -d '{\n    \"region_name\": \"us-west-1\",\n    \"instance_type_name\": \"gpu_1x_h100_sxm5\",\n    \"ssh_key_names\": [\"my-key\"]\n  }' | jq\n```\n\n### Terminate instance\n\n```bash\ncurl -u $LAMBDA_API_KEY: \\\n  -X POST https:\u002F\u002Fcloud.lambdalabs.com\u002Fapi\u002Fv1\u002Finstance-operations\u002Fterminate \\\n  -H \"Content-Type: application\u002Fjson\" \\\n  -d '{\"instance_ids\": [\"\u003CINSTANCE-ID>\"]}' | jq\n```\n\n## Persistent storage\n\n### Filesystems\n\nFilesystems persist data across instance restarts:\n\n```bash\n# Mount location\n\u002Flambda\u002Fnfs\u002F\u003CFILESYSTEM_NAME>\n\n# Example: save checkpoints\npython train.py --checkpoint-dir \u002Flambda\u002Fnfs\u002Fmy-storage\u002Fcheckpoints\n```\n\n### Create filesystem\n\n1. Go to Storage in Lambda console\n2. Click \"Create filesystem\"\n3. Select region (must match instance region)\n4. Name and create\n\n### Attach to instance\n\nFilesystems must be attached at instance launch time:\n- Via console: Select filesystem when launching\n- Via API: Include `file_system_names` in launch request\n\n### Best practices\n\n```bash\n# Store on filesystem (persists)\n\u002Flambda\u002Fnfs\u002Fstorage\u002F\n  ├── datasets\u002F\n  ├── checkpoints\u002F\n  ├── models\u002F\n  └── outputs\u002F\n\n# Local SSD (faster, ephemeral)\n\u002Fhome\u002Fubuntu\u002F\n  └── working\u002F  # Temporary files\n```\n\n## SSH configuration\n\n### Add SSH key\n\n```bash\n# Generate key locally\nssh-keygen -t ed25519 -f ~\u002F.ssh\u002Flambda_key\n\n# Add public key to Lambda console\n# Or via API\n```\n\n### Multiple keys\n\n```bash\n# On instance, add more keys\necho 'ssh-rsa AAAA...' >> ~\u002F.ssh\u002Fauthorized_keys\n```\n\n### Import from GitHub\n\n```bash\n# On instance\nssh-import-id gh:username\n```\n\n### SSH tunneling\n\n```bash\n# Forward Jupyter\nssh -L 8888:localhost:8888 ubuntu@\u003CIP>\n\n# Forward TensorBoard\nssh -L 6006:localhost:6006 ubuntu@\u003CIP>\n\n# Multiple ports\nssh -L 8888:localhost:8888 -L 6006:localhost:6006 ubuntu@\u003CIP>\n```\n\n## JupyterLab\n\n### Launch from console\n\n1. Go to Instances page\n2. Click \"Launch\" in Cloud IDE column\n3. JupyterLab opens in browser\n\n### Manual access\n\n```bash\n# On instance\njupyter lab --ip=0.0.0.0 --port=8888\n\n# From local machine with tunnel\nssh -L 8888:localhost:8888 ubuntu@\u003CIP>\n# Open http:\u002F\u002Flocalhost:8888\n```\n\n## Training workflows\n\n### Single-GPU training\n\n```bash\n# SSH to instance\nssh ubuntu@\u003CIP>\n\n# Clone repo\ngit clone https:\u002F\u002Fgithub.com\u002Fuser\u002Fproject\ncd project\n\n# Install dependencies\npip install -r requirements.txt\n\n# Train\npython train.py --epochs 100 --checkpoint-dir \u002Flambda\u002Fnfs\u002Fstorage\u002Fcheckpoints\n```\n\n### Multi-GPU training (single node)\n\n```python\n# train_ddp.py\nimport torch\nimport torch.distributed as dist\nfrom torch.nn.parallel import DistributedDataParallel as DDP\n\ndef main():\n    dist.init_process_group(\"nccl\")\n    rank = dist.get_rank()\n    device = rank % torch.cuda.device_count()\n\n    model = MyModel().to(device)\n    model = DDP(model, device_ids=[device])\n\n    # Training loop...\n\nif __name__ == \"__main__\":\n    main()\n```\n\n```bash\n# Launch with torchrun (8 GPUs)\ntorchrun --nproc_per_node=8 train_ddp.py\n```\n\n### Checkpoint to filesystem\n\n```python\nimport os\n\ncheckpoint_dir = \"\u002Flambda\u002Fnfs\u002Fmy-storage\u002Fcheckpoints\"\nos.makedirs(checkpoint_dir, exist_ok=True)\n\n# Save checkpoint\ntorch.save({\n    'epoch': epoch,\n    'model_state_dict': model.state_dict(),\n    'optimizer_state_dict': optimizer.state_dict(),\n    'loss': loss,\n}, f\"{checkpoint_dir}\u002Fcheckpoint_{epoch}.pt\")\n```\n\n## 1-Click Clusters\n\n### Overview\n\nHigh-performance Slurm clusters with:\n- 16-512 NVIDIA H100 or B200 GPUs\n- NVIDIA Quantum-2 400 Gb\u002Fs InfiniBand\n- GPUDirect RDMA at 3200 Gb\u002Fs\n- Pre-installed distributed ML stack\n\n### Included software\n\n- Ubuntu 22.04 LTS + Lambda Stack\n- NCCL, Open MPI\n- PyTorch with DDP and FSDP\n- TensorFlow\n- OFED drivers\n\n### Storage\n\n- 24 TB NVMe per compute node (ephemeral)\n- Lambda filesystems for persistent data\n\n### Multi-node training\n\n```bash\n# On Slurm cluster\nsrun --nodes=4 --ntasks-per-node=8 --gpus-per-node=8 \\\n  torchrun --nnodes=4 --nproc_per_node=8 \\\n  --rdzv_backend=c10d --rdzv_endpoint=$MASTER_ADDR:29500 \\\n  train.py\n```\n\n## Networking\n\n### Bandwidth\n\n- Inter-instance (same region): up to 200 Gbps\n- Internet outbound: 20 Gbps max\n\n### Firewall\n\n- Default: Only port 22 (SSH) open\n- Configure additional ports in Lambda console\n- ICMP traffic allowed by default\n\n### Private IPs\n\n```bash\n# Find private IP\nip addr show | grep 'inet '\n```\n\n## Common workflows\n\n### Workflow 1: Fine-tuning LLM\n\n```bash\n# 1. Launch 8x H100 instance with filesystem\n\n# 2. SSH and setup\nssh ubuntu@\u003CIP>\npip install transformers accelerate peft\n\n# 3. Download model to filesystem\npython -c \"\nfrom transformers import AutoModelForCausalLM\nmodel = AutoModelForCausalLM.from_pretrained('meta-llama\u002FLlama-2-7b-hf')\nmodel.save_pretrained('\u002Flambda\u002Fnfs\u002Fstorage\u002Fmodels\u002Fllama-2-7b')\n\"\n\n# 4. Fine-tune with checkpoints on filesystem\naccelerate launch --num_processes 8 train.py \\\n  --model_path \u002Flambda\u002Fnfs\u002Fstorage\u002Fmodels\u002Fllama-2-7b \\\n  --output_dir \u002Flambda\u002Fnfs\u002Fstorage\u002Foutputs \\\n  --checkpoint_dir \u002Flambda\u002Fnfs\u002Fstorage\u002Fcheckpoints\n```\n\n### Workflow 2: Batch inference\n\n```bash\n# 1. Launch A10 instance (cost-effective for inference)\n\n# 2. Run inference\npython inference.py \\\n  --model \u002Flambda\u002Fnfs\u002Fstorage\u002Fmodels\u002Ffine-tuned \\\n  --input \u002Flambda\u002Fnfs\u002Fstorage\u002Fdata\u002Finputs.jsonl \\\n  --output \u002Flambda\u002Fnfs\u002Fstorage\u002Fdata\u002Foutputs.jsonl\n```\n\n## Cost optimization\n\n### Choose right GPU\n\n| Task | Recommended GPU |\n|------|-----------------|\n| LLM fine-tuning (7B) | A100 40GB |\n| LLM fine-tuning (70B) | 8x H100 |\n| Inference | A10, A6000 |\n| Development | V100, A10 |\n| Maximum performance | B200 |\n\n### Reduce costs\n\n1. **Use filesystems**: Avoid re-downloading data\n2. **Checkpoint frequently**: Resume interrupted training\n3. **Right-size**: Don't over-provision GPUs\n4. **Terminate idle**: No auto-stop, manually terminate\n\n### Monitor usage\n\n- Dashboard shows real-time GPU utilization\n- API for programmatic monitoring\n\n## Common issues\n\n| Issue | Solution |\n|-------|----------|\n| Instance won't launch | Check region availability, try different GPU |\n| SSH connection refused | Wait for instance to initialize (3-15 min) |\n| Data lost after terminate | Use persistent filesystems |\n| Slow data transfer | Use filesystem in same region |\n| GPU not detected | Reboot instance, check drivers |\n\n## References\n\n- **[Advanced Usage](references\u002Fadvanced-usage.md)** - Multi-node training, API automation\n- **[Troubleshooting](references\u002Ftroubleshooting.md)** - Common issues and solutions\n\n## Resources\n\n- **Documentation**: https:\u002F\u002Fdocs.lambda.ai\n- **Console**: https:\u002F\u002Fcloud.lambda.ai\n- **Pricing**: https:\u002F\u002Flambda.ai\u002Finstances\n- **Support**: https:\u002F\u002Fsupport.lambdalabs.com\n- **Blog**: https:\u002F\u002Flambda.ai\u002Fblog\n\n## Related skills\n\n- `modal-serverless-gpu`: Serverless GPU cloud platform for running ML workloads.\n- `skypilot-multi-cloud-orchestration`: Multi-cloud orchestration for ML workloads with automatic cost optimization.\n- `modal`: Modal is a serverless cloud platform for running Python on demand, including on-demand GPUs.\n","# Lambda Labs GPU Cloud\n\nComprehensive guide to running ML workloads on Lambda Labs GPU cloud with on-demand instances and 1-Click Clusters.\n\n## When to use Lambda Labs\n\n**Use Lambda Labs when:**\n- Need dedicated GPU instances with full SSH access\n- Running long training jobs (hours to days)\n- Want simple pricing with no egress fees\n- Need persistent storage across sessions\n- Require high-performance multi-node clusters (16-512 GPUs)\n- Want pre-installed ML stack (Lambda Stack with PyTorch, CUDA, NCCL)\n\n**Key features:**\n- **GPU variety**: B200, H100, GH200, A100, A10, A6000, V100\n- **Lambda Stack**: Pre-installed PyTorch, TensorFlow, CUDA, cuDNN, NCCL\n- **Persistent filesystems**: Keep data across instance restarts\n- **1-Click Clusters**: 16-512 GPU Slurm clusters with InfiniBand\n- **Simple pricing**: Pay-per-minute, no egress fees\n- **Global regions**: 12+ regions worldwide\n\n**Use alternatives instead:**\n- **Modal**: For serverless, auto-scaling workloads\n- **SkyPilot**: For multi-cloud orchestration and cost optimization\n- **RunPod**: For cheaper spot instances and serverless endpoints\n- **Vast.ai**: For GPU marketplace with lowest prices\n\n## Quick start\n\n### Account setup\n\n1. Create account at https:\u002F\u002Flambda.ai\n2. Add payment method\n3. Generate API key from dashboard\n4. Add SSH key (required before launching instances)\n\n### Launch via console\n\n1. Go to https:\u002F\u002Fcloud.lambda.ai\u002Finstances\n2. Click \"Launch instance\"\n3. Select GPU type and region\n4. Choose SSH key\n5. Optionally attach filesystem\n6. Launch and wait 3-15 minutes\n\n### Connect via SSH\n\n```bash\n# Get instance IP from console\nssh ubuntu@\u003CINSTANCE-IP>\n\n# Or with specific key\nssh -i ~\u002F.ssh\u002Flambda_key ubuntu@\u003CINSTANCE-IP>\n```\n\n## GPU instances\n\n### Available GPUs\n\n| GPU | VRAM | Price\u002FGPU\u002Fhr | Best For |\n|-----|------|--------------|----------|\n| B200 SXM6 | 180 GB | $4.99 | Largest models, fastest training |\n| H100 SXM | 80 GB | $2.99-3.29 | Large model training |\n| H100 PCIe | 80 GB | $2.49 | Cost-effective H100 |\n| GH200 | 96 GB | $1.49 | Single-GPU large models |\n| A100 80GB | 80 GB | $1.79 | Production training |\n| A100 40GB | 40 GB | $1.29 | Standard training |\n| A10 | 24 GB | $0.75 | Inference, fine-tuning |\n| A6000 | 48 GB | $0.80 | Good VRAM\u002Fprice ratio |\n| V100 | 16 GB | $0.55 | Budget training |\n\n### Instance configurations\n\n```\n8x GPU: Best for distributed training (DDP, FSDP)\n4x GPU: Large models, multi-GPU training\n2x GPU: Medium workloads\n1x GPU: Fine-tuning, inference, development\n```\n\n### Launch times\n\n- Single-GPU: 3-5 minutes\n- Multi-GPU: 10-15 minutes\n\n## Lambda Stack\n\nAll instances come with Lambda Stack pre-installed:\n\n```bash\n# Included software\n- Ubuntu 22.04 LTS\n- NVIDIA drivers (latest)\n- CUDA 12.x\n- cuDNN 8.x\n- NCCL (for multi-GPU)\n- PyTorch (latest)\n- TensorFlow (latest)\n- JAX\n- JupyterLab\n```\n\n### Verify installation\n\n```bash\n# Check GPU\nnvidia-smi\n\n# Check PyTorch\npython -c \"import torch; print(torch.cuda.is_available())\"\n\n# Check CUDA version\nnvcc --version\n```\n\n## Python API\n\nDetails, code examples and parameter tables: [references\u002Fpython-api.md](references\u002Fpython-api.md). Read it when this step applies.\n\n## CLI with curl\n\n### List instance types\n\n```bash\ncurl -u $LAMBDA_API_KEY: \\\n  https:\u002F\u002Fcloud.lambdalabs.com\u002Fapi\u002Fv1\u002Finstance-types | jq\n```\n\n### Launch instance\n\n```bash\ncurl -u $LAMBDA_API_KEY: \\\n  -X POST https:\u002F\u002Fcloud.lambdalabs.com\u002Fapi\u002Fv1\u002Finstance-operations\u002Flaunch \\\n  -H \"Content-Type: application\u002Fjson\" \\\n  -d '{\n    \"region_name\": \"us-west-1\",\n    \"instance_type_name\": \"gpu_1x_h100_sxm5\",\n    \"ssh_key_names\": [\"my-key\"]\n  }' | jq\n```\n\n### Terminate instance\n\n```bash\ncurl -u $LAMBDA_API_KEY: \\\n  -X POST https:\u002F\u002Fcloud.lambdalabs.com\u002Fapi\u002Fv1\u002Finstance-operations\u002Fterminate \\\n  -H \"Content-Type: application\u002Fjson\" \\\n  -d '{\"instance_ids\": [\"\u003CINSTANCE-ID>\"]}' | jq\n```\n\n## Persistent storage\n\n### Filesystems\n\nFilesystems persist data across instance restarts:\n\n```bash\n# Mount location\n\u002Flambda\u002Fnfs\u002F\u003CFILESYSTEM_NAME>\n\n# Example: save checkpoints\npython train.py --checkpoint-dir \u002Flambda\u002Fnfs\u002Fmy-storage\u002Fcheckpoints\n```\n\n### Create filesystem\n\n1. Go to Storage in Lambda console\n2. Click \"Create filesystem\"\n3. Select region (must match instance region)\n4. Name and create\n\n### Attach to instance\n\nFilesystems must be attached at instance launch time:\n- Via console: Select filesystem when launching\n- Via API: Include `file_system_names` in launch request\n\n### Best practices\n\n```bash\n# Store on filesystem (persists)\n\u002Flambda\u002Fnfs\u002Fstorage\u002F\n  ├── datasets\u002F\n  ├── checkpoints\u002F\n  ├── models\u002F\n  └── outputs\u002F\n\n# Local SSD (faster, ephemeral)\n\u002Fhome\u002Fubuntu\u002F\n  └── working\u002F  # Temporary files\n```\n\n## SSH configuration\n\n### Add SSH key\n\n```bash\n# Generate key locally\nssh-keygen -t ed25519 -f ~\u002F.ssh\u002Flambda_key\n\n# Add public key to Lambda console\n# Or via API\n```\n\n### Multiple keys\n\n```bash\n# On instance, add more keys\necho 'ssh-rsa AAAA...' >> ~\u002F.ssh\u002Fauthorized_keys\n```\n\n### Import from GitHub\n\n```bash\n# On instance\nssh-import-id gh:username\n```\n\n### SSH tunneling\n\n```bash\n# Forward Jupyter\nssh -L 8888:localhost:8888 ubuntu@\u003CIP>\n\n# Forward TensorBoard\nssh -L 6006:localhost:6006 ubuntu@\u003CIP>\n\n# Multiple ports\nssh -L 8888:localhost:8888 -L 6006:localhost:6006 ubuntu@\u003CIP>\n```\n\n## JupyterLab\n\n### Launch from console\n\n1. Go to Instances page\n2. Click \"Launch\" in Cloud IDE column\n3. JupyterLab opens in browser\n\n### Manual access\n\n```bash\n# On instance\njupyter lab --ip=0.0.0.0 --port=8888\n\n# From local machine with tunnel\nssh -L 8888:localhost:8888 ubuntu@\u003CIP>\n# Open http:\u002F\u002Flocalhost:8888\n```\n\n## Training workflows\n\n### Single-GPU training\n\n```bash\n# SSH to instance\nssh ubuntu@\u003CIP>\n\n# Clone repo\ngit clone https:\u002F\u002Fgithub.com\u002Fuser\u002Fproject\ncd project\n\n# Install dependencies\npip install -r requirements.txt\n\n# Train\npython train.py --epochs 100 --checkpoint-dir \u002Flambda\u002Fnfs\u002Fstorage\u002Fcheckpoints\n```\n\n### Multi-GPU training (single node)\n\n```python\n# train_ddp.py\nimport torch\nimport torch.distributed as dist\nfrom torch.nn.parallel import DistributedDataParallel as DDP\n\ndef main():\n    dist.init_process_group(\"nccl\")\n    rank = dist.get_rank()\n    device = rank % torch.cuda.device_count()\n\n    model = MyModel().to(device)\n    model = DDP(model, device_ids=[device])\n\n    # Training loop...\n\nif __name__ == \"__main__\":\n    main()\n```\n\n```bash\n# Launch with torchrun (8 GPUs)\ntorchrun --nproc_per_node=8 train_ddp.py\n```\n\n### Checkpoint to filesystem\n\n```python\nimport os\n\ncheckpoint_dir = \"\u002Flambda\u002Fnfs\u002Fmy-storage\u002Fcheckpoints\"\nos.makedirs(checkpoint_dir, exist_ok=True)\n\n# Save checkpoint\ntorch.save({\n    'epoch': epoch,\n    'model_state_dict': model.state_dict(),\n    'optimizer_state_dict': optimizer.state_dict(),\n    'loss': loss,\n}, f\"{checkpoint_dir}\u002Fcheckpoint_{epoch}.pt\")\n```\n\n## 1-Click Clusters\n\n### Overview\n\nHigh-performance Slurm clusters with:\n- 16-512 NVIDIA H100 or B200 GPUs\n- NVIDIA Quantum-2 400 Gb\u002Fs InfiniBand\n- GPUDirect RDMA at 3200 Gb\u002Fs\n- Pre-installed distributed ML stack\n\n### Included software\n\n- Ubuntu 22.04 LTS + Lambda Stack\n- NCCL, Open MPI\n- PyTorch with DDP and FSDP\n- TensorFlow\n- OFED drivers\n\n### Storage\n\n- 24 TB NVMe per compute node (ephemeral)\n- Lambda filesystems for persistent data\n\n### Multi-node training\n\n```bash\n# On Slurm cluster\nsrun --nodes=4 --ntasks-per-node=8 --gpus-per-node=8 \\\n  torchrun --nnodes=4 --nproc_per_node=8 \\\n  --rdzv_backend=c10d --rdzv_endpoint=$MASTER_ADDR:29500 \\\n  train.py\n```\n\n## Networking\n\n### Bandwidth\n\n- Inter-instance (same region): up to 200 Gbps\n- Internet outbound: 20 Gbps max\n\n### Firewall\n\n- Default: Only port 22 (SSH) open\n- Configure additional ports in Lambda console\n- ICMP traffic allowed by default\n\n### Private IPs\n\n```bash\n# Find private IP\nip addr show | grep 'inet '\n```\n\n## Common workflows\n\n### Workflow 1: Fine-tuning LLM\n\n```bash\n# 1. Launch 8x H100 instance with filesystem\n\n# 2. SSH and setup\nssh ubuntu@\u003CIP>\npip install transformers accelerate peft\n\n# 3. Download model to filesystem\npython -c \"\nfrom transformers import AutoModelForCausalLM\nmodel = AutoModelForCausalLM.from_pretrained('meta-llama\u002FLlama-2-7b-hf')\nmodel.save_pretrained('\u002Flambda\u002Fnfs\u002Fstorage\u002Fmodels\u002Fllama-2-7b')\n\"\n\n# 4. Fine-tune with checkpoints on filesystem\naccelerate launch --num_processes 8 train.py \\\n  --model_path \u002Flambda\u002Fnfs\u002Fstorage\u002Fmodels\u002Fllama-2-7b \\\n  --output_dir \u002Flambda\u002Fnfs\u002Fstorage\u002Foutputs \\\n  --checkpoint_dir \u002Flambda\u002Fnfs\u002Fstorage\u002Fcheckpoints\n```\n\n### Workflow 2: Batch inference\n\n```bash\n# 1. Launch A10 instance (cost-effective for inference)\n\n# 2. Run inference\npython inference.py \\\n  --model \u002Flambda\u002Fnfs\u002Fstorage\u002Fmodels\u002Ffine-tuned \\\n  --input \u002Flambda\u002Fnfs\u002Fstorage\u002Fdata\u002Finputs.jsonl \\\n  --output \u002Flambda\u002Fnfs\u002Fstorage\u002Fdata\u002Foutputs.jsonl\n```\n\n## Cost optimization\n\n### Choose right GPU\n\n| Task | Recommended GPU |\n|------|-----------------|\n| LLM fine-tuning (7B) | A100 40GB |\n| LLM fine-tuning (70B) | 8x H100 |\n| Inference | A10, A6000 |\n| Development | V100, A10 |\n| Maximum performance | B200 |\n\n### Reduce costs\n\n1. **Use filesystems**: Avoid re-downloading data\n2. **Checkpoint frequently**: Resume interrupted training\n3. **Right-size**: Don't over-provision GPUs\n4. **Terminate idle**: No auto-stop, manually terminate\n\n### Monitor usage\n\n- Dashboard shows real-time GPU utilization\n- API for programmatic monitoring\n\n## Common issues\n\n| Issue | Solution |\n|-------|----------|\n| Instance won't launch | Check region availability, try different GPU |\n| SSH connection refused | Wait for instance to initialize (3-15 min) |\n| Data lost after terminate | Use persistent filesystems |\n| Slow data transfer | Use filesystem in same region |\n| GPU not detected | Reboot instance, check drivers |\n\n## References\n\n- **[Advanced Usage](references\u002Fadvanced-usage.md)** - Multi-node training, API automation\n- **[Troubleshooting](references\u002Ftroubleshooting.md)** - Common issues and solutions\n\n## Resources\n\n- **Documentation**: https:\u002F\u002Fdocs.lambda.ai\n- **Console**: https:\u002F\u002Fcloud.lambda.ai\n- **Pricing**: https:\u002F\u002Flambda.ai\u002Finstances\n- **Support**: https:\u002F\u002Fsupport.lambdalabs.com\n- **Blog**: https:\u002F\u002Flambda.ai\u002Fblog\n\n## Related skills\n\n- `modal-serverless-gpu`: Serverless GPU cloud platform for running ML workloads.\n- `skypilot-multi-cloud-orchestration`: Multi-cloud orchestration for ML workloads with automatic cost optimization.\n- `modal`: Modal is a serverless cloud platform for running Python on demand, including on-demand GPUs.\n",{"name":5,"license":15,"metadata":24,"description":7},{"tags":25,"version":26,"category":27,"maintainer":28,"dependencies":29},"Infrastructure, GPU Cloud, Training, Inference, Lambda Labs","1.0.0","ml-inference-and-ops","Kalaris Labs","lambda-cloud-client>=1.0.0","main","29fe36d26127205057d5e0ed8425f6b0e9a8c13a","skills\u002Flambda-labs-gpu-cloud",16,"https:\u002F\u002Fgithub.com\u002FKalarisLabs\u002Fresearch-agent-skills","https:\u002F\u002Fdocs.kalarislabs.com\u002Fresearch-agent-skills","https:\u002F\u002Fgithub.com\u002FKalarisLabs","kalarislabs.com","Applied AI lab building the infrastructure for the agentic era","2026-11-14 17:49:08",{"source":8,"skillName":5,"commands":41,"prompt":51,"zipAvailable":52},[42,45,48],{"tool":43,"command":44},"npx","npx skills add KalarisLabs\u002Fresearch-agent-skills --skill lambda-labs-gpu-cloud",{"tool":46,"command":47},"bunx","bunx skills add KalarisLabs\u002Fresearch-agent-skills --skill lambda-labs-gpu-cloud",{"tool":49,"command":50},"pnpm","pnpm dlx skills add KalarisLabs\u002Fresearch-agent-skills --skill lambda-labs-gpu-cloud","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FKalarisLabs\u002Fresearch-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Flambda-labs-gpu-cloud\u002FSKILL.md",true,[54,60,64],{"path":55,"resourceType":56,"resourceTypeDesc":57,"sizeBytes":58,"sourceHtmlUrl":59},"references\u002Fadvanced-usage.md","REFERENCE","参考文档",15100,"https:\u002F\u002Fgithub.com\u002FKalarisLabs\u002Fresearch-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Flambda-labs-gpu-cloud\u002Freferences\u002Fadvanced-usage.md",{"path":61,"resourceType":56,"resourceTypeDesc":57,"sizeBytes":62,"sourceHtmlUrl":63},"references\u002Fpython-api.md",1765,"https:\u002F\u002Fgithub.com\u002FKalarisLabs\u002Fresearch-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Flambda-labs-gpu-cloud\u002Freferences\u002Fpython-api.md",{"path":65,"resourceType":56,"resourceTypeDesc":57,"sizeBytes":66,"sourceHtmlUrl":67},"references\u002Ftroubleshooting.md",11683,"https:\u002F\u002Fgithub.com\u002FKalarisLabs\u002Fresearch-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Flambda-labs-gpu-cloud\u002Freferences\u002Ftroubleshooting.md",[69,76,83,90,97,104],{"skillNo":70,"name":71,"nameZh":6,"description":72,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":73,"categories":74,"keywords":75,"publishedAt":19},"01M74Z8MGT305R11SNY5G28QH5","plos","Prepare manuscripts for PLOS journals (PLOS ONE, PLOS Biology, PLOS Computational Biology, PLOS Genetics, PLOS Medicine, PLOS Pathogens, PLOS Neglected Tropical Diseases, PLOS Global Public Health and others), covering structure, abstract and author summary, Vancouver-style references, the mandatory data availability policy, ethics and financial disclosure statements, figure requirements (PACE), the PLOS LaTeX template, reporting guidelines and PLOS ONE's publication criteria. Use when targeting any PLOS journal or checking compliance with PLOS policies.","https:\u002F\u002Fgithub.com\u002FKalarisLabs\u002Fresearch-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fplos\u002FSKILL.md",[],[],{"skillNo":77,"name":78,"nameZh":6,"description":79,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":80,"categories":81,"keywords":82,"publishedAt":19},"01M74Z8CCN9APXYHNQ95V1EBGT","pkpd-modeling","Pharmacokinetic and pharmacodynamic modelling and simulation - non-compartmental analysis, compartmental and population PK, PK\u002FPD and exposure-response, TMDD, PBPK orientation, bioequivalence, allometric scaling and first-in-human dose, drug interaction prediction, and Bayesian therapeutic drug monitoring. Use when analysing concentration-time data, deriving exposure metrics, fitting PK or PD models, or evaluating dosing regimens. Triggers include \"pharmacokinetics\", \"pharmacodynamics\", \"PK\u002FPD\", \"NCA\", \"non-compartmental\", \"AUC\", \"Cmax\", \"lambda z\", \"half-life\", \"clearance\", \"volume of distribution\", \"compartmental model\", \"population PK\", \"popPK\", \"NONMEM\", \"nlmixr2\", \"Pharmpy\", \"Monolix\", \"exposure-response\", \"Emax\", \"EC50\", \"indirect response\", \"effect compartment\", \"TMDD\", \"PBPK\", \"bioequivalence\", \"RSABE\", \"ABEL\", \"allometric scaling\", \"first-in-human\", \"MABEL\", \"drug-drug interaction\", \"DDI\", \"ICH M12\", \"concentration-QTc\", \"therapeutic drug monitoring\", \"MIPD\", and \"dosing regimen\".","https:\u002F\u002Fgithub.com\u002FKalarisLabs\u002Fresearch-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fpkpd-modeling\u002FSKILL.md",[],[],{"skillNo":84,"name":85,"nameZh":6,"description":86,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":87,"categories":88,"keywords":89,"publishedAt":19},"01M74Z86YHJJ1FM3X87BHN76KZ","pinecone","Guides use of Pinecone, a managed serverless vector database, through its Python client and the LangChain and LlamaIndex integrations. Covers creating indexes, upserting and querying vectors, metadata filtering, namespaces, hybrid dense and sparse search, index management, and deleting vectors. Use when building a production RAG system on a hosted vector store, adding semantic search or recommendations without running infrastructure, isolating per-user or per-tenant data with namespaces, combining dense and sparse vectors in one query, or filtering results by metadata. Do not use for self-hosted or local stores (Chroma, Weaviate) or offline similarity search (FAISS).","https:\u002F\u002Fgithub.com\u002FKalarisLabs\u002Fresearch-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fpinecone\u002FSKILL.md",[],[],{"skillNo":91,"name":92,"nameZh":6,"description":93,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":94,"categories":95,"keywords":96,"publishedAt":19},"01M74Z80T6E89SAJ8MYDD0767Q","pi-agent","Build with and use Pi, the minimal terminal coding harness. Use for installing Pi, configuring providers\u002Fmodels\u002Fsettings\u002Fenvironment variables, creating Pi skills\u002Fextensions\u002Fpackages\u002Fthemes\u002Fprompt templates, embedding Pi through the SDK, integrating over RPC or JSON event streams, parsing sessions, running local models through the llama.cpp router, developing custom Pi providers and TUI components, or using ecosystem packages such as pi-subagents (delegation\u002Forchestration), pi-mcp-adapter (MCP servers), pi-interview (interactive forms), and pi-web-access (web search, fetching, video understanding).","https:\u002F\u002Fgithub.com\u002FKalarisLabs\u002Fresearch-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fpi-agent\u002FSKILL.md",[],[],{"skillNo":98,"name":99,"nameZh":6,"description":100,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":101,"categories":102,"keywords":103,"publishedAt":19},"01M74Z7WF1Y8Q3VFAQPPPRAFMV","phoenix-observability","Open-source AI observability platform for LLM tracing, evaluation, and monitoring. Use when debugging LLM applications with detailed traces, running evaluations on datasets, or monitoring production AI systems with real-time insights.","https:\u002F\u002Fgithub.com\u002FKalarisLabs\u002Fresearch-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fphoenix-observability\u002FSKILL.md",[],[],{"skillNo":105,"name":106,"nameZh":6,"description":107,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":108,"categories":109,"keywords":110,"publishedAt":19},"01M74Z7HNEH1A2AVYR3G1X2QV6","peft-fine-tuning","Fine-tunes LLMs with Hugging Face PEFT, using LoRA, QLoRA, IA3, AdaLoRA, prefix tuning, and prompt tuning so that under 1% of parameters are trained. Covers rank, alpha, and target module selection, loading and merging adapters, multi-adapter serving, and integration with TRL SFTTrainer, Axolotl, and vLLM. Use when fine-tuning 7B-70B models on limited GPU memory, when running QLoRA on a single 24GB GPU, when choosing LoRA rank and alpha, when merging or swapping adapters on one base model, or when debugging CUDA OOM or an adapter that does not apply. Not for full fine-tuning of small models under 1B parameters or cases needing all weights updated.","https:\u002F\u002Fgithub.com\u002FKalarisLabs\u002Fresearch-agent-skills\u002Fblob\u002Fmain\u002Fskills\u002Fpeft-fine-tuning\u002FSKILL.md",[],[]]