[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7FCH1VAF9VVSTB2Z8QH4BBB":3,"skill-same:01M7FCH1VAF9VVSTB2Z8QH4BBB":67},{"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":35,"commitSha":36,"skillDirectory":37,"forksCount":38,"repositoryHtmlUrl":39,"repositoryHomepageUrl":6,"ownerHtmlUrl":40,"ownerBlogUrl":6,"ownerDescription":41,"updatedAt":42,"install":43,"resources":56},"01M7FCH1VAF9VVSTB2Z8QH4BBB","skypilot-multi-cloud-orchestration",null,"Multi-cloud orchestration for ML workloads with automatic cost optimization. Use when you need to run training or batch jobs across multiple clouds, leverage spot instances with auto-recovery, or optimize GPU costs across providers.","CZX-LBJ\u002Fdeepseek-science","CZX-LBJ","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F204594689?v=4","01M7FC59WQ9PFXT04MKGJDCFFT","czx","01M7FC59XB2GW0BKWN9JTDBQS7",1,"MIT","https:\u002F\u002Fgithub.com\u002FCZX-LBJ\u002Fdeepseek-science\u002Fblob\u002Fmain\u002Fpackages\u002Fskill\u002Fskill-science\u002Fassets\u002Fsources\u002Fai-research-skills\u002F09-infrastructure\u002Fskypilot\u002FSKILL.md",[],[],"2026-11-16 10:05:34",false,"---\nname: skypilot-multi-cloud-orchestration\ndescription: Multi-cloud orchestration for ML workloads with automatic cost optimization. Use when you need to run training or batch jobs across multiple clouds, leverage spot instances with auto-recovery, or optimize GPU costs across providers.\nversion: 1.0.0\nauthor: Orchestra Research\nlicense: MIT\ntags: [Infrastructure, Multi-Cloud, Orchestration, GPU, Cost Optimization, SkyPilot]\ndependencies: [skypilot>=0.7.0]\n---\n\n# SkyPilot Multi-Cloud Orchestration\n\nComprehensive guide to running ML workloads across clouds with automatic cost optimization using SkyPilot.\n\n## When to use SkyPilot\n\n**Use SkyPilot when:**\n- Running ML workloads across multiple clouds (AWS, GCP, Azure, etc.)\n- Need cost optimization with automatic cloud\u002Fregion selection\n- Running long jobs on spot instances with auto-recovery\n- Managing distributed multi-node training\n- Want unified interface for 20+ cloud providers\n- Need to avoid vendor lock-in\n\n**Key features:**\n- **Multi-cloud**: AWS, GCP, Azure, Kubernetes, Lambda, RunPod, 20+ providers\n- **Cost optimization**: Automatic cheapest cloud\u002Fregion selection\n- **Spot instances**: 3-6x cost savings with automatic recovery\n- **Distributed training**: Multi-node jobs with gang scheduling\n- **Managed jobs**: Auto-recovery, checkpointing, fault tolerance\n- **Sky Serve**: Model serving with autoscaling\n\n**Use alternatives instead:**\n- **Modal**: For simpler serverless GPU with Python-native API\n- **RunPod**: For single-cloud persistent pods\n- **Kubernetes**: For existing K8s infrastructure\n- **Ray**: For pure Ray-based orchestration\n\n## Quick start\n\n### Installation\n\n```bash\npip install \"skypilot[aws,gcp,azure,kubernetes]\"\n\n# Verify cloud credentials\nsky check\n```\n\n### Hello World\n\nCreate `hello.yaml`:\n```yaml\nresources:\n  accelerators: T4:1\n\nrun: |\n  nvidia-smi\n  echo \"Hello from SkyPilot!\"\n```\n\nLaunch:\n```bash\nsky launch -c hello hello.yaml\n\n# SSH to cluster\nssh hello\n\n# Terminate\nsky down hello\n```\n\n## Core concepts\n\n### Task YAML structure\n\n```yaml\n# Task name (optional)\nname: my-task\n\n# Resource requirements\nresources:\n  cloud: aws              # Optional: auto-select if omitted\n  region: us-west-2       # Optional: auto-select if omitted\n  accelerators: A100:4    # GPU type and count\n  cpus: 8+                # Minimum CPUs\n  memory: 32+             # Minimum memory (GB)\n  use_spot: true          # Use spot instances\n  disk_size: 256          # Disk size (GB)\n\n# Number of nodes for distributed training\nnum_nodes: 2\n\n# Working directory (synced to ~\u002Fsky_workdir)\nworkdir: .\n\n# Setup commands (run once)\nsetup: |\n  pip install -r requirements.txt\n\n# Run commands\nrun: |\n  python train.py\n```\n\n### Key commands\n\n| Command | Purpose |\n|---------|---------|\n| `sky launch` | Launch cluster and run task |\n| `sky exec` | Run task on existing cluster |\n| `sky status` | Show cluster status |\n| `sky stop` | Stop cluster (preserve state) |\n| `sky down` | Terminate cluster |\n| `sky logs` | View task logs |\n| `sky queue` | Show job queue |\n| `sky jobs launch` | Launch managed job |\n| `sky serve up` | Deploy serving endpoint |\n\n## GPU configuration\n\n### Available accelerators\n\n```yaml\n# NVIDIA GPUs\naccelerators: T4:1\naccelerators: L4:1\naccelerators: A10G:1\naccelerators: L40S:1\naccelerators: A100:4\naccelerators: A100-80GB:8\naccelerators: H100:8\n\n# Cloud-specific\naccelerators: V100:4         # AWS\u002FGCP\naccelerators: TPU-v4-8       # GCP TPUs\n```\n\n### GPU fallbacks\n\n```yaml\nresources:\n  accelerators:\n    H100: 8\n    A100-80GB: 8\n    A100: 8\n  any_of:\n    - cloud: gcp\n    - cloud: aws\n    - cloud: azure\n```\n\n### Spot instances\n\n```yaml\nresources:\n  accelerators: A100:8\n  use_spot: true\n  spot_recovery: FAILOVER  # Auto-recover on preemption\n```\n\n## Cluster management\n\n### Launch and execute\n\n```bash\n# Launch new cluster\nsky launch -c mycluster task.yaml\n\n# Run on existing cluster (skip setup)\nsky exec mycluster another_task.yaml\n\n# Interactive SSH\nssh mycluster\n\n# Stream logs\nsky logs mycluster\n```\n\n### Autostop\n\n```yaml\nresources:\n  accelerators: A100:4\n  autostop:\n    idle_minutes: 30\n    down: true  # Terminate instead of stop\n```\n\n```bash\n# Set autostop via CLI\nsky autostop mycluster -i 30 --down\n```\n\n### Cluster status\n\n```bash\n# All clusters\nsky status\n\n# Detailed view\nsky status -a\n```\n\n## Distributed training\n\n### Multi-node setup\n\n```yaml\nresources:\n  accelerators: A100:8\n\nnum_nodes: 4  # 4 nodes × 8 GPUs = 32 GPUs total\n\nsetup: |\n  pip install torch torchvision\n\nrun: |\n  torchrun \\\n    --nnodes=$SKYPILOT_NUM_NODES \\\n    --nproc_per_node=$SKYPILOT_NUM_GPUS_PER_NODE \\\n    --node_rank=$SKYPILOT_NODE_RANK \\\n    --master_addr=$(echo \"$SKYPILOT_NODE_IPS\" | head -n1) \\\n    --master_port=12355 \\\n    train.py\n```\n\n### Environment variables\n\n| Variable | Description |\n|----------|-------------|\n| `SKYPILOT_NODE_RANK` | Node index (0 to num_nodes-1) |\n| `SKYPILOT_NODE_IPS` | Newline-separated IP addresses |\n| `SKYPILOT_NUM_NODES` | Total number of nodes |\n| `SKYPILOT_NUM_GPUS_PER_NODE` | GPUs per node |\n\n### Head-node-only execution\n\n```bash\nrun: |\n  if [ \"${SKYPILOT_NODE_RANK}\" == \"0\" ]; then\n    python orchestrate.py\n  fi\n```\n\n## Managed jobs\n\n### Spot recovery\n\n```bash\n# Launch managed job with spot recovery\nsky jobs launch -n my-job train.yaml\n```\n\n### Checkpointing\n\n```yaml\nname: training-job\n\nfile_mounts:\n  \u002Fcheckpoints:\n    name: my-checkpoints\n    store: s3\n    mode: MOUNT\n\nresources:\n  accelerators: A100:8\n  use_spot: true\n\nrun: |\n  python train.py \\\n    --checkpoint-dir \u002Fcheckpoints \\\n    --resume-from-latest\n```\n\n### Job management\n\n```bash\n# List jobs\nsky jobs queue\n\n# View logs\nsky jobs logs my-job\n\n# Cancel job\nsky jobs cancel my-job\n```\n\n## File mounts and storage\n\n### Local file sync\n\n```yaml\nworkdir: .\u002Fmy-project  # Synced to ~\u002Fsky_workdir\n\nfile_mounts:\n  \u002Fdata\u002Fconfig.yaml: .\u002Fconfig.yaml\n  ~\u002F.vimrc: ~\u002F.vimrc\n```\n\n### Cloud storage\n\n```yaml\nfile_mounts:\n  # Mount S3 bucket\n  \u002Fdatasets:\n    source: s3:\u002F\u002Fmy-bucket\u002Fdatasets\n    mode: MOUNT  # Stream from S3\n\n  # Copy GCS bucket\n  \u002Fmodels:\n    source: gs:\u002F\u002Fmy-bucket\u002Fmodels\n    mode: COPY  # Pre-fetch to disk\n\n  # Cached mount (fast writes)\n  \u002Foutputs:\n    name: my-outputs\n    store: s3\n    mode: MOUNT_CACHED\n```\n\n### Storage modes\n\n| Mode | Description | Best For |\n|------|-------------|----------|\n| `MOUNT` | Stream from cloud | Large datasets, read-heavy |\n| `COPY` | Pre-fetch to disk | Small files, random access |\n| `MOUNT_CACHED` | Cache with async upload | Checkpoints, outputs |\n\n## Sky Serve (Model Serving)\n\n### Basic service\n\n```yaml\n# service.yaml\nservice:\n  readiness_probe: \u002Fhealth\n  replica_policy:\n    min_replicas: 1\n    max_replicas: 10\n    target_qps_per_replica: 2.0\n\nresources:\n  accelerators: A100:1\n\nrun: |\n  python -m vllm.entrypoints.openai.api_server \\\n    --model meta-llama\u002FLlama-2-7b-chat-hf \\\n    --port 8000\n```\n\n```bash\n# Deploy\nsky serve up -n my-service service.yaml\n\n# Check status\nsky serve status\n\n# Get endpoint\nsky serve status my-service\n```\n\n### Autoscaling policies\n\n```yaml\nservice:\n  replica_policy:\n    min_replicas: 1\n    max_replicas: 10\n    target_qps_per_replica: 2.0\n    upscale_delay_seconds: 60\n    downscale_delay_seconds: 300\n  load_balancing_policy: round_robin\n```\n\n## Cost optimization\n\n### Automatic cloud selection\n\n```yaml\n# SkyPilot finds cheapest option\nresources:\n  accelerators: A100:8\n  # No cloud specified - auto-select cheapest\n```\n\n```bash\n# Show optimizer decision\nsky launch task.yaml --dryrun\n```\n\n### Cloud preferences\n\n```yaml\nresources:\n  accelerators: A100:8\n  any_of:\n    - cloud: gcp\n      region: us-central1\n    - cloud: aws\n      region: us-east-1\n    - cloud: azure\n```\n\n### Environment variables\n\n```yaml\nenvs:\n  HF_TOKEN: $HF_TOKEN  # Inherited from local env\n  WANDB_API_KEY: $WANDB_API_KEY\n\n# Or use secrets\nsecrets:\n  - HF_TOKEN\n  - WANDB_API_KEY\n```\n\n## Common workflows\n\n### Workflow 1: Fine-tuning with checkpoints\n\n```yaml\nname: llm-finetune\n\nfile_mounts:\n  \u002Fcheckpoints:\n    name: finetune-checkpoints\n    store: s3\n    mode: MOUNT_CACHED\n\nresources:\n  accelerators: A100:8\n  use_spot: true\n\nsetup: |\n  pip install transformers accelerate\n\nrun: |\n  python train.py \\\n    --checkpoint-dir \u002Fcheckpoints \\\n    --resume\n```\n\n### Workflow 2: Hyperparameter sweep\n\n```yaml\nname: hp-sweep-${RUN_ID}\n\nenvs:\n  RUN_ID: 0\n  LEARNING_RATE: 1e-4\n  BATCH_SIZE: 32\n\nresources:\n  accelerators: A100:1\n  use_spot: true\n\nrun: |\n  python train.py \\\n    --lr $LEARNING_RATE \\\n    --batch-size $BATCH_SIZE \\\n    --run-id $RUN_ID\n```\n\n```bash\n# Launch multiple jobs\nfor i in {1..10}; do\n  sky jobs launch sweep.yaml \\\n    --env RUN_ID=$i \\\n    --env LEARNING_RATE=$(python -c \"import random; print(10**random.uniform(-5,-3))\")\ndone\n```\n\n## Debugging\n\n```bash\n# SSH to cluster\nssh mycluster\n\n# View logs\nsky logs mycluster\n\n# Check job queue\nsky queue mycluster\n\n# View managed job logs\nsky jobs logs my-job\n```\n\n## Common issues\n\n| Issue | Solution |\n|-------|----------|\n| Quota exceeded | Request quota increase, try different region |\n| Spot preemption | Use `sky jobs launch` for auto-recovery |\n| Slow file sync | Use `MOUNT_CACHED` mode for outputs |\n| GPU not available | Use `any_of` for fallback clouds |\n\n## References\n\n- **[Advanced Usage](references\u002Fadvanced-usage.md)** - Multi-cloud, optimization, production patterns\n- **[Troubleshooting](references\u002Ftroubleshooting.md)** - Common issues and solutions\n\n## Resources\n\n- **Documentation**: https:\u002F\u002Fdocs.skypilot.co\n- **GitHub**: https:\u002F\u002Fgithub.com\u002Fskypilot-org\u002Fskypilot\n- **Slack**: https:\u002F\u002Fslack.skypilot.co\n- **Examples**: https:\u002F\u002Fgithub.com\u002Fskypilot-org\u002Fskypilot\u002Ftree\u002Fmaster\u002Fexamples\n","# SkyPilot Multi-Cloud Orchestration\n\nComprehensive guide to running ML workloads across clouds with automatic cost optimization using SkyPilot.\n\n## When to use SkyPilot\n\n**Use SkyPilot when:**\n- Running ML workloads across multiple clouds (AWS, GCP, Azure, etc.)\n- Need cost optimization with automatic cloud\u002Fregion selection\n- Running long jobs on spot instances with auto-recovery\n- Managing distributed multi-node training\n- Want unified interface for 20+ cloud providers\n- Need to avoid vendor lock-in\n\n**Key features:**\n- **Multi-cloud**: AWS, GCP, Azure, Kubernetes, Lambda, RunPod, 20+ providers\n- **Cost optimization**: Automatic cheapest cloud\u002Fregion selection\n- **Spot instances**: 3-6x cost savings with automatic recovery\n- **Distributed training**: Multi-node jobs with gang scheduling\n- **Managed jobs**: Auto-recovery, checkpointing, fault tolerance\n- **Sky Serve**: Model serving with autoscaling\n\n**Use alternatives instead:**\n- **Modal**: For simpler serverless GPU with Python-native API\n- **RunPod**: For single-cloud persistent pods\n- **Kubernetes**: For existing K8s infrastructure\n- **Ray**: For pure Ray-based orchestration\n\n## Quick start\n\n### Installation\n\n```bash\npip install \"skypilot[aws,gcp,azure,kubernetes]\"\n\n# Verify cloud credentials\nsky check\n```\n\n### Hello World\n\nCreate `hello.yaml`:\n```yaml\nresources:\n  accelerators: T4:1\n\nrun: |\n  nvidia-smi\n  echo \"Hello from SkyPilot!\"\n```\n\nLaunch:\n```bash\nsky launch -c hello hello.yaml\n\n# SSH to cluster\nssh hello\n\n# Terminate\nsky down hello\n```\n\n## Core concepts\n\n### Task YAML structure\n\n```yaml\n# Task name (optional)\nname: my-task\n\n# Resource requirements\nresources:\n  cloud: aws              # Optional: auto-select if omitted\n  region: us-west-2       # Optional: auto-select if omitted\n  accelerators: A100:4    # GPU type and count\n  cpus: 8+                # Minimum CPUs\n  memory: 32+             # Minimum memory (GB)\n  use_spot: true          # Use spot instances\n  disk_size: 256          # Disk size (GB)\n\n# Number of nodes for distributed training\nnum_nodes: 2\n\n# Working directory (synced to ~\u002Fsky_workdir)\nworkdir: .\n\n# Setup commands (run once)\nsetup: |\n  pip install -r requirements.txt\n\n# Run commands\nrun: |\n  python train.py\n```\n\n### Key commands\n\n| Command | Purpose |\n|---------|---------|\n| `sky launch` | Launch cluster and run task |\n| `sky exec` | Run task on existing cluster |\n| `sky status` | Show cluster status |\n| `sky stop` | Stop cluster (preserve state) |\n| `sky down` | Terminate cluster |\n| `sky logs` | View task logs |\n| `sky queue` | Show job queue |\n| `sky jobs launch` | Launch managed job |\n| `sky serve up` | Deploy serving endpoint |\n\n## GPU configuration\n\n### Available accelerators\n\n```yaml\n# NVIDIA GPUs\naccelerators: T4:1\naccelerators: L4:1\naccelerators: A10G:1\naccelerators: L40S:1\naccelerators: A100:4\naccelerators: A100-80GB:8\naccelerators: H100:8\n\n# Cloud-specific\naccelerators: V100:4         # AWS\u002FGCP\naccelerators: TPU-v4-8       # GCP TPUs\n```\n\n### GPU fallbacks\n\n```yaml\nresources:\n  accelerators:\n    H100: 8\n    A100-80GB: 8\n    A100: 8\n  any_of:\n    - cloud: gcp\n    - cloud: aws\n    - cloud: azure\n```\n\n### Spot instances\n\n```yaml\nresources:\n  accelerators: A100:8\n  use_spot: true\n  spot_recovery: FAILOVER  # Auto-recover on preemption\n```\n\n## Cluster management\n\n### Launch and execute\n\n```bash\n# Launch new cluster\nsky launch -c mycluster task.yaml\n\n# Run on existing cluster (skip setup)\nsky exec mycluster another_task.yaml\n\n# Interactive SSH\nssh mycluster\n\n# Stream logs\nsky logs mycluster\n```\n\n### Autostop\n\n```yaml\nresources:\n  accelerators: A100:4\n  autostop:\n    idle_minutes: 30\n    down: true  # Terminate instead of stop\n```\n\n```bash\n# Set autostop via CLI\nsky autostop mycluster -i 30 --down\n```\n\n### Cluster status\n\n```bash\n# All clusters\nsky status\n\n# Detailed view\nsky status -a\n```\n\n## Distributed training\n\n### Multi-node setup\n\n```yaml\nresources:\n  accelerators: A100:8\n\nnum_nodes: 4  # 4 nodes × 8 GPUs = 32 GPUs total\n\nsetup: |\n  pip install torch torchvision\n\nrun: |\n  torchrun \\\n    --nnodes=$SKYPILOT_NUM_NODES \\\n    --nproc_per_node=$SKYPILOT_NUM_GPUS_PER_NODE \\\n    --node_rank=$SKYPILOT_NODE_RANK \\\n    --master_addr=$(echo \"$SKYPILOT_NODE_IPS\" | head -n1) \\\n    --master_port=12355 \\\n    train.py\n```\n\n### Environment variables\n\n| Variable | Description |\n|----------|-------------|\n| `SKYPILOT_NODE_RANK` | Node index (0 to num_nodes-1) |\n| `SKYPILOT_NODE_IPS` | Newline-separated IP addresses |\n| `SKYPILOT_NUM_NODES` | Total number of nodes |\n| `SKYPILOT_NUM_GPUS_PER_NODE` | GPUs per node |\n\n### Head-node-only execution\n\n```bash\nrun: |\n  if [ \"${SKYPILOT_NODE_RANK}\" == \"0\" ]; then\n    python orchestrate.py\n  fi\n```\n\n## Managed jobs\n\n### Spot recovery\n\n```bash\n# Launch managed job with spot recovery\nsky jobs launch -n my-job train.yaml\n```\n\n### Checkpointing\n\n```yaml\nname: training-job\n\nfile_mounts:\n  \u002Fcheckpoints:\n    name: my-checkpoints\n    store: s3\n    mode: MOUNT\n\nresources:\n  accelerators: A100:8\n  use_spot: true\n\nrun: |\n  python train.py \\\n    --checkpoint-dir \u002Fcheckpoints \\\n    --resume-from-latest\n```\n\n### Job management\n\n```bash\n# List jobs\nsky jobs queue\n\n# View logs\nsky jobs logs my-job\n\n# Cancel job\nsky jobs cancel my-job\n```\n\n## File mounts and storage\n\n### Local file sync\n\n```yaml\nworkdir: .\u002Fmy-project  # Synced to ~\u002Fsky_workdir\n\nfile_mounts:\n  \u002Fdata\u002Fconfig.yaml: .\u002Fconfig.yaml\n  ~\u002F.vimrc: ~\u002F.vimrc\n```\n\n### Cloud storage\n\n```yaml\nfile_mounts:\n  # Mount S3 bucket\n  \u002Fdatasets:\n    source: s3:\u002F\u002Fmy-bucket\u002Fdatasets\n    mode: MOUNT  # Stream from S3\n\n  # Copy GCS bucket\n  \u002Fmodels:\n    source: gs:\u002F\u002Fmy-bucket\u002Fmodels\n    mode: COPY  # Pre-fetch to disk\n\n  # Cached mount (fast writes)\n  \u002Foutputs:\n    name: my-outputs\n    store: s3\n    mode: MOUNT_CACHED\n```\n\n### Storage modes\n\n| Mode | Description | Best For |\n|------|-------------|----------|\n| `MOUNT` | Stream from cloud | Large datasets, read-heavy |\n| `COPY` | Pre-fetch to disk | Small files, random access |\n| `MOUNT_CACHED` | Cache with async upload | Checkpoints, outputs |\n\n## Sky Serve (Model Serving)\n\n### Basic service\n\n```yaml\n# service.yaml\nservice:\n  readiness_probe: \u002Fhealth\n  replica_policy:\n    min_replicas: 1\n    max_replicas: 10\n    target_qps_per_replica: 2.0\n\nresources:\n  accelerators: A100:1\n\nrun: |\n  python -m vllm.entrypoints.openai.api_server \\\n    --model meta-llama\u002FLlama-2-7b-chat-hf \\\n    --port 8000\n```\n\n```bash\n# Deploy\nsky serve up -n my-service service.yaml\n\n# Check status\nsky serve status\n\n# Get endpoint\nsky serve status my-service\n```\n\n### Autoscaling policies\n\n```yaml\nservice:\n  replica_policy:\n    min_replicas: 1\n    max_replicas: 10\n    target_qps_per_replica: 2.0\n    upscale_delay_seconds: 60\n    downscale_delay_seconds: 300\n  load_balancing_policy: round_robin\n```\n\n## Cost optimization\n\n### Automatic cloud selection\n\n```yaml\n# SkyPilot finds cheapest option\nresources:\n  accelerators: A100:8\n  # No cloud specified - auto-select cheapest\n```\n\n```bash\n# Show optimizer decision\nsky launch task.yaml --dryrun\n```\n\n### Cloud preferences\n\n```yaml\nresources:\n  accelerators: A100:8\n  any_of:\n    - cloud: gcp\n      region: us-central1\n    - cloud: aws\n      region: us-east-1\n    - cloud: azure\n```\n\n### Environment variables\n\n```yaml\nenvs:\n  HF_TOKEN: $HF_TOKEN  # Inherited from local env\n  WANDB_API_KEY: $WANDB_API_KEY\n\n# Or use secrets\nsecrets:\n  - HF_TOKEN\n  - WANDB_API_KEY\n```\n\n## Common workflows\n\n### Workflow 1: Fine-tuning with checkpoints\n\n```yaml\nname: llm-finetune\n\nfile_mounts:\n  \u002Fcheckpoints:\n    name: finetune-checkpoints\n    store: s3\n    mode: MOUNT_CACHED\n\nresources:\n  accelerators: A100:8\n  use_spot: true\n\nsetup: |\n  pip install transformers accelerate\n\nrun: |\n  python train.py \\\n    --checkpoint-dir \u002Fcheckpoints \\\n    --resume\n```\n\n### Workflow 2: Hyperparameter sweep\n\n```yaml\nname: hp-sweep-${RUN_ID}\n\nenvs:\n  RUN_ID: 0\n  LEARNING_RATE: 1e-4\n  BATCH_SIZE: 32\n\nresources:\n  accelerators: A100:1\n  use_spot: true\n\nrun: |\n  python train.py \\\n    --lr $LEARNING_RATE \\\n    --batch-size $BATCH_SIZE \\\n    --run-id $RUN_ID\n```\n\n```bash\n# Launch multiple jobs\nfor i in {1..10}; do\n  sky jobs launch sweep.yaml \\\n    --env RUN_ID=$i \\\n    --env LEARNING_RATE=$(python -c \"import random; print(10**random.uniform(-5,-3))\")\ndone\n```\n\n## Debugging\n\n```bash\n# SSH to cluster\nssh mycluster\n\n# View logs\nsky logs mycluster\n\n# Check job queue\nsky queue mycluster\n\n# View managed job logs\nsky jobs logs my-job\n```\n\n## Common issues\n\n| Issue | Solution |\n|-------|----------|\n| Quota exceeded | Request quota increase, try different region |\n| Spot preemption | Use `sky jobs launch` for auto-recovery |\n| Slow file sync | Use `MOUNT_CACHED` mode for outputs |\n| GPU not available | Use `any_of` for fallback clouds |\n\n## References\n\n- **[Advanced Usage](references\u002Fadvanced-usage.md)** - Multi-cloud, optimization, production patterns\n- **[Troubleshooting](references\u002Ftroubleshooting.md)** - Common issues and solutions\n\n## Resources\n\n- **Documentation**: https:\u002F\u002Fdocs.skypilot.co\n- **GitHub**: https:\u002F\u002Fgithub.com\u002Fskypilot-org\u002Fskypilot\n- **Slack**: https:\u002F\u002Fslack.skypilot.co\n- **Examples**: https:\u002F\u002Fgithub.com\u002Fskypilot-org\u002Fskypilot\u002Ftree\u002Fmaster\u002Fexamples\n",{"name":5,"tags":24,"author":31,"license":15,"version":32,"description":7,"dependencies":33},[25,26,27,28,29,30],"Infrastructure","Multi-Cloud","Orchestration","GPU","Cost Optimization","SkyPilot","Orchestra Research","1.0.0",[34],"skypilot>=0.7.0","main","18193d71ea3441d7c033dbf195b8a7c75ac96f75","packages\u002Fskill\u002Fskill-science\u002Fassets\u002Fsources\u002Fai-research-skills\u002F09-infrastructure\u002Fskypilot",0,"https:\u002F\u002Fgithub.com\u002FCZX-LBJ\u002Fdeepseek-science","https:\u002F\u002Fgithub.com\u002FCZX-LBJ","一个没事喜欢随便写点代码的小菜鸡\r\n","2026-10-22 09:48:04",{"source":8,"skillName":5,"commands":44,"prompt":54,"zipAvailable":55},[45,48,51],{"tool":46,"command":47},"npx","npx skills add CZX-LBJ\u002Fdeepseek-science --skill skypilot-multi-cloud-orchestration",{"tool":49,"command":50},"bunx","bunx skills add CZX-LBJ\u002Fdeepseek-science --skill skypilot-multi-cloud-orchestration",{"tool":52,"command":53},"pnpm","pnpm dlx skills add CZX-LBJ\u002Fdeepseek-science --skill skypilot-multi-cloud-orchestration","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FCZX-LBJ\u002Fdeepseek-science\u002Fblob\u002Fmain\u002Fpackages\u002Fskill\u002Fskill-science\u002Fassets\u002Fsources\u002Fai-research-skills\u002F09-infrastructure\u002Fskypilot\u002FSKILL.md",true,[57,63],{"path":58,"resourceType":59,"resourceTypeDesc":60,"sizeBytes":61,"sourceHtmlUrl":62},"references\u002Fadvanced-usage.md","REFERENCE","参考文档",7469,"https:\u002F\u002Fgithub.com\u002FCZX-LBJ\u002Fdeepseek-science\u002Fblob\u002Fmain\u002Fpackages\u002Fskill\u002Fskill-science\u002Fassets\u002Fsources\u002Fai-research-skills\u002F09-infrastructure\u002Fskypilot\u002Freferences\u002Fadvanced-usage.md",{"path":64,"resourceType":59,"resourceTypeDesc":60,"sizeBytes":65,"sourceHtmlUrl":66},"references\u002Ftroubleshooting.md",10493,"https:\u002F\u002Fgithub.com\u002FCZX-LBJ\u002Fdeepseek-science\u002Fblob\u002Fmain\u002Fpackages\u002Fskill\u002Fskill-science\u002Fassets\u002Fsources\u002Fai-research-skills\u002F09-infrastructure\u002Fskypilot\u002Freferences\u002Ftroubleshooting.md",[68,75,82,89,96,103],{"skillNo":69,"name":70,"nameZh":6,"description":71,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":72,"categories":73,"keywords":74,"publishedAt":19},"01M7FEEKSNMRMKW6JEK93PFDFY","bio-free-energy-calculations","Performs alchemical free-energy calculations including relative binding free energy (RBFE \u002F FEP+) and absolute binding free energy (ABFE) via OpenFE, FEP+, GROMACS, AMBER pmemd, and OpenMM with explicit lambda scheduling, soft-core potentials, MBAR\u002FBAR analysis, cycle-closure validation, and protocol-appropriate enhanced sampling. Compares ML alternatives (Boltz-2 affinity, DeepDock). Use when ranking analogs by binding affinity beyond docking accuracy, performing prospective lead optimization, or validating SAR predictions.","https:\u002F\u002Fgithub.com\u002FCZX-LBJ\u002Fdeepseek-science\u002Fblob\u002Fmain\u002Fpackages\u002Fskill\u002Fskill-science\u002Fassets\u002Fsources\u002Fbioskills\u002Fchemoinformatics\u002Ffree-energy-calculations\u002FSKILL.md",[],[],{"skillNo":76,"name":77,"nameZh":6,"description":78,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":79,"categories":80,"keywords":81,"publishedAt":19},"01M7FEE5CTE5QK55CEMXNNVWY9","bio-conformer-generation","Generates 3D conformer ensembles using RDKit ETKDGv3 with knowledge-enhanced distance geometry, MMFF94\u002FUFF force-field optimization, CREST + GFN2-xTB semi-empirical refinement, and macrocycle-aware torsion preferences. Provides explicit decision rules for single vs ensemble conformer use, RMSD pruning, energy windows, conformer count, and force-field choice. Use when preparing 3D ligands for docking, generating descriptor input for 3D QSAR, or sampling macrocycle\u002Fpeptide conformational ensembles.","https:\u002F\u002Fgithub.com\u002FCZX-LBJ\u002Fdeepseek-science\u002Fblob\u002Fmain\u002Fpackages\u002Fskill\u002Fskill-science\u002Fassets\u002Fsources\u002Fbioskills\u002Fchemoinformatics\u002Fconformer-generation\u002FSKILL.md",[],[],{"skillNo":83,"name":84,"nameZh":6,"description":85,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":86,"categories":87,"keywords":88,"publishedAt":19},"01M7FEDNGFEMFWD1MSEDR5H7CY","bio-admet-prediction","Predicts ADMET properties using ADMETlab 3.0 (119 platform features, including 77 prediction models with modeled-endpoint uncertainty), ADMET-AI, DeepChem MolNet, and chemprop D-MPNN with explicit handling of OECD QSAR principles, applicability domain assessment, calibration, hERG\u002FCYP\u002FAMES endpoints, and PAINS \u002F Lipinski \u002F Ro5 \u002F Veber \u002F BBB druglikeness filters. Use when filtering compounds for drug-likeness, prioritizing leads by predicted safety, or building an in-house ADMET QSAR model.","https:\u002F\u002Fgithub.com\u002FCZX-LBJ\u002Fdeepseek-science\u002Fblob\u002Fmain\u002Fpackages\u002Fskill\u002Fskill-science\u002Fassets\u002Fsources\u002Fbioskills\u002Fchemoinformatics\u002Fadmet-prediction\u002FSKILL.md",[],[],{"skillNo":90,"name":91,"nameZh":6,"description":92,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":93,"categories":94,"keywords":95,"publishedAt":19},"01M7FED69GFB54FRF7AV5066DN","writing-systems-papers","Paragraph-level structural blueprint for 10-12 page systems papers targeting OSDI, SOSP, ASPLOS, NSDI, and EuroSys. Provides page allocation, paragraph templates, and writing patterns. Use when user says \"写系统论文\", \"systems paper structure\", \"OSDI paper\", \"SOSP paper\", or wants fine-grained structural guidance for a systems conference submission.","https:\u002F\u002Fgithub.com\u002FCZX-LBJ\u002Fdeepseek-science\u002Fblob\u002Fmain\u002Fpackages\u002Fskill\u002Fskill-science\u002Fassets\u002Fsources\u002Faris\u002Fskills\u002Fwriting-systems-papers\u002FSKILL.md",[],[],{"skillNo":97,"name":98,"nameZh":6,"description":99,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":100,"categories":101,"keywords":102,"publishedAt":19},"01M7FECVVDWZ0X0VEDTCE4B4DZ","wiki-enrich","Fill in the per-paper TODO sections of research-wiki\u002Fpapers\u002F\u003Cslug>.md pages that literature-ingest skills leave as bare scaffolds. Use when user says 'enrich wiki', 'fill paper TODOs', 'wiki body 補完', '把 paper 摘要寫進 wiki', 'research-wiki 自動填', or after a batch ingest that left papers\u002F as TODO scaffolds.","https:\u002F\u002Fgithub.com\u002FCZX-LBJ\u002Fdeepseek-science\u002Fblob\u002Fmain\u002Fpackages\u002Fskill\u002Fskill-science\u002Fassets\u002Fsources\u002Faris\u002Fskills\u002Fwiki-enrich\u002FSKILL.md",[],[],{"skillNo":104,"name":105,"nameZh":6,"description":106,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":107,"categories":108,"keywords":109,"publishedAt":19},"01M7FECD4BCJ2MTSBDDSE3ZAJ5","web-debug-search","Search GitHub, Stack Exchange, Chinese technical communities, official documentation, and general developer web sources for software errors, compatibility problems, API usage questions, and real-world workarounds. Use for debugging and discovery only; results are not paper-citation evidence.","https:\u002F\u002Fgithub.com\u002FCZX-LBJ\u002Fdeepseek-science\u002Fblob\u002Fmain\u002Fpackages\u002Fskill\u002Fskill-science\u002Fassets\u002Fsources\u002Faris\u002Fskills\u002Fweb-debug-search\u002FSKILL.md",[],[]]