vastai

vastai

在 Vast.ai 上管理 GPU 实例、模板、卷、无服务器端点、SSH 密钥和账单。

命令为 vastai（小写）。机器可读的 JSON 输出始终使用 --raw。

安装

# Linux/macOS (recommended, no Python required)
curl -fsSL https://vast.ai/install.sh | bash

# or via PyPI (Windows, or if you'd rather use pip)
pip install vastai

快速开始

vastai set api-key <YOUR_API_KEY>                   # Authenticate (one-time); Create API Key in account at https://console.vast.ai/manage-keys/
vastai show user                                    # Verify auth + check balance
vastai create ssh-key ~/.ssh/id_ed25519.pub         # Register SSH key (do BEFORE create)
vastai search offers 'gpu_name=RTX_4090 num_gpus=1 verified=true direct_port_count>=1 rentable=true' -o 'dlperf_usd-'
# Note the offer ID from the output
vastai create instance <OFFER_ID> --image pytorch/pytorch:@vastai-automatic-tag --disk 20 --ssh --direct
# Automatically grab appropriate image tag; Response:  {"success": true, "new_contract": <INSTANCE_ID>}
vastai show instance <INSTANCE_ID>                  # Poll until actual_status == "running" (see Instance status values below)
vastai ssh-url <INSTANCE_ID>                        # Get SSH connection string
vastai copy local:./data/ <INSTANCE_ID>:/workspace/ # Upload files
vastai destroy instance <INSTANCE_ID> -y             # Clean up (stops all billing; -y skips confirmation)

API 密钥：https://console.vast.ai/manage-keys/

全局标志

所有命令均可使用：

--api-key KEY    Override stored API key
--raw            Output machine-readable JSON (agents should always use this)
--full           Print full results (don't page with less)
--explain        Show underlying API calls (useful for debugging)
--curl           Show equivalent curl command
--no-color       Disable colored output
--url URL        Override server REST API URL
--retry RETRY    Set retry limit for API calls
--version        Show CLI version

查询语法

搜索命令接受筛选表达式。运算符：=、!=、>、>=、<、<=、in、notin。

# Examples
'gpu_name=RTX_4090 num_gpus=1'           # Exact match + numeric
'gpu_ram>=48 reliability>0.95'           # Greater-than filters
'geolocation=EU dph_total<=2.0'          # Region + price cap

常用筛选字段：num_gpus、gpu_name、gpu_ram、cpu_ram、disk_space、reliability、compute_cap、inet_up、inet_down、dph_total、geolocation、direct_port_count、verified、rentable

常用排序字段：score（默认值 — 综合性价比）、dlperf_usd（每美元的 DL 性能）、dph_total（价格）、num_gpus、reliability

命令

实例

vastai show instances                                    # List all your instances (fetches every page, no prompts)
vastai show instances --status running loading           # Filter by status
vastai show instances --gpu-name 'RTX 4090'              # Filter by GPU
vastai show instances --label training                   # Filter by label
vastai show instances --order-by start_date desc         # Sort by column
vastai show instances --cols id,status,gpu,dph           # Custom columns
vastai show instances --raw                              # Flat JSON list of all instances (scripting)
vastai show instance <id>                                # Poll single instance (use for status checks)
vastai create instance <offer-id> --image pytorch/pytorch:2.4.0-cuda12.4-cudnn9-runtime --disk 20 --ssh --direct
# Response includes "new_contract": <id> — that is your instance ID
vastai launch instance --gpu-name RTX_4090 --num-gpus 1 --image pytorch/pytorch
vastai start instance <id>                               # Start stopped instance
vastai stop instance <id>                                # Stop (preserves disk, no GPU charges)
vastai reboot instance <id>                              # Stop + start
vastai destroy instance <id> -y                          # Delete permanently (irreversible; -y required for non-interactive use)
vastai destroy instances <id1> <id2> -y                  # Batch delete (-y skips confirmation prompt)
vastai label instance <id> --label "training-run-1"      # Tag instance
vastai update instance <id>                              # Recreate from updated template
vastai prepay instance <id>                              # Deposit credits into reserved instance
vastai recycle instance <id>                             # Destroy + recreate

推荐的 Vast.ai 镜像（使用 @vastai-automatic-tag 自动为机器选择合适的标签；在 https://vast.ai/model-library 浏览预配置模型）：

vastai/base-image:@vastai-automatic-tag          # Minimal Ubuntu base
vastai/pytorch:@vastai-automatic-tag             # PyTorch + CUDA
vastai/linux-desktop:@vastai-automatic-tag       # Linux desktop (VNC/RDP)

# vLLM — set model via env vars with huggingface model example
vastai create instance <id> --image vastai/vllm:@vastai-automatic-tag --disk 40 --ssh --direct \
  --env '-e MODEL_NAME=Qwen/Qwen2.5-3B-Instruct -e HF_TOKEN=hf_xxx'

# ComfyUI — set model checkpoint via env vars with huggingface model example
vastai create instance <id> --image vastai/comfy:@vastai-automatic-tag --disk 40 --ssh --direct \
  --env '-e CHECKPOINT_MODEL=black-forest-labs/FLUX.1-schnell -e HF_TOKEN=hf_xxx'

create instance 标志：

• --image IMAGE — Docker 镜像
• --disk DISK — 本地磁盘，单位为 GB
• --ssh / --jupyter — 连接类型
• --direct — 更快的直连
• --label LABEL — 实例标签
• --env ENV — 环境变量和端口映射，例如 '-e TZ=UTC -p 8080:8080'
• --onstart FILE — 启动脚本文件的路径
• --onstart-cmd CMD — 内联字符串形式的启动脚本（较长的脚本请使用 --onstart，或通过 gzip+base64 编码）
• --bid_price PRICE — 可中断（竞价）价格，单位为美元/小时
• --template_hash HASH — 根据模板创建
• --create-volume ID — 挂载新卷
• --link-volume ID — 挂载现有卷
• --cancel-unavail — 无可用机器时失败（而非创建已停止的实例）

实例状态值：

actual_status	含义
null	正在配置
created	实例已创建，尚未配置
loading	正在下载镜像 / 启动容器
running	活跃 — 收取 GPU 费用
stopped	已停止 — 仅收取磁盘费用
frozen	已暂停并保留内存 — 收取 GPU 费用
exited	容器进程意外退出
rebooting	正在重启（临时状态）
unknown	主机没有最近的心跳信号
offline	主机已断开与 Vast 服务器的连接

**轮询循环警告：**如果 actual_status 变为 exited、unknown 或 offline，它将永远无法达到 running。务必添加超时和错误分支，否则脚本会无限循环，同时持续产生磁盘费用。销毁该实例并使用其他报价重试。

**费用：**存储费用从创建时开始计收。GPU 费用从状态达到 running 时开始计收。

可中断（竞价）租用

可中断（竞价）实例的价格低于按需实例，但随时可能被出价更低的其他用户中断。注意：search offers --type bid 会显示 min_bid，但 create instance 默认使用 dph_total 按需创建，除非传入 --bid_price <floor>。在使用 --type bid 搜索后，务必传入 --bid_price，否则租用的将是按需实例 / 按需价格，而不是可中断实例。

被超过出价后，实例会变为 stopped（而非销毁），存储费用仍会继续计收。可以通过 update instance --bid_price 提高出价来恢复实例。

搜索

vastai search offers                                     # Default: verified, on-demand, sorted by score
vastai search offers 'gpu_name=RTX_4090 num_gpus=1 verified=true direct_port_count>=1' -o 'dlperf_usd-'
vastai search offers 'num_gpus>=4 reliability>0.99' -o 'num_gpus-'
vastai search offers --type bid                          # Interruptible (spot) pricing
vastai search offers --type reserved                     # Reserved pricing
vastai search offers -n 'gpu_name=H100_SXM'             # No default filters
vastai search volumes                                    # Search volume offers
vastai search templates "pytorch"                        # Search templates
vastai search benchmarks                                 # Search benchmarks
vastai search invoices                                   # Search invoice history

search offers 标志：--type on-demand|reserved|bid、--order/-o FIELD[-]、--limit、--storage GB、--no-default/-n

SSH 和文件传输

vastai ssh-url <id>                                      # Get ssh:// connection URL
vastai scp-url <id>                                      # Get scp:// URL
vastai attach ssh <id> "ssh-ed25519 AAAA..."             # Attach SSH key to instance
vastai detach ssh <id> <ssh_key_id>                      # Remove SSH key (ssh_key_id from show ssh-keys)
vastai show ssh-keys                                     # List account SSH keys
vastai create ssh-key ~/.ssh/id_ed25519.pub              # Add SSH key from file (do BEFORE create instance)
vastai create ssh-key                                    # Generate new key if you don't have one
vastai create ssh-key "ssh-ed25519 AAAA..."              # Add SSH key inline
vastai delete ssh-key <id>                               # Remove SSH key from account
vastai update ssh-key <id> "ssh-ed25519 AAAA..."         # Update SSH key value

注意：ssh-url 会返回连接字符串，不会打开交互式会话。使用 $(vastai ssh-url <id>) 提取 URL，或解析 --raw JSON 输出。

文件复制

vastai copy <src> <dst>                                  # Copy between instance and local
vastai copy local:./data/ <id>:/workspace/data/          # Local → instance (preferred format)
vastai copy <id>:/workspace/results/ local:./results/    # Instance → local
vastai copy <id-a>:/workspace/ <id-b>:/workspace/        # Instance → instance
vastai copy s3.101:/data/ C.<id>:/workspace/             # Cloud service → instance (s3.<connection-id>)
vastai copy V.1234:/file C.<id>:/workspace/              # Volume → instance
vastai copy V.1234:/file s3.101:/workspace/              # Volume → cloud service
# Legacy format also works: vastai copy 12345:./data ./local-data
vastai cloud copy --src ./data --dst s3://bucket/path \
  --instance 12345 --connection <conn-id> \
  --transfer "Instance To Cloud"                         # To cloud storage (add connection in UI on settings page)
vastai cancel copy <dst-id>                              # Cancel in-progress copy

云端复制标志： --src、--dst、--instance、--connection、--transfer

位置格式： [instance_id:]path（旧版）、C.instance_id:path、V.volume_id:path、cloud_service[.id]:path、local:path。卷复制目前仅支持复制到其他卷、实例或云服务，不支持复制到本地。切勿复制到 /root 或 /——这会破坏实例上的 ssh 权限，并导致后续复制失败。

日志与执行

vastai logs <id>                                         # Container logs (last 1000 lines)
vastai logs <id> --tail 100                              # Last 100 lines
vastai logs <id> --filter "error"                        # Grep filter
vastai execute <id> "nvidia-smi"                         # Run command on instance
vastai execute <id> "ls /workspace" --schedule DAILY     # Scheduled execution

卷

vastai search volumes                                    # Search available volume offers
vastai show volumes                                      # List your volumes
vastai create volume <offer_id> [-s SIZE] [-n NAME]      # Create volume (offer_id from search volumes)
vastai clone volume <source_id> <dest_id> [-s SIZE]      # Clone volume (dest_id from search volumes)
vastai delete volume <id>                                # Delete volume
vastai create network-volume ...                         # Create network volume
vastai list network-volume                               # List network volumes
vastai take snapshot <instance_id> --repo REPO --docker_login_user USER --docker_login_pass PASS  # Snapshot instance to volume

无服务器与部署

vastai show endpoints                                    # List endpoint groups
vastai create endpoint --name "my-ep" ...                # Create endpoint
vastai update endpoint <id> ...                          # Update endpoint
vastai delete endpoint <id>                              # Delete endpoint
vastai get endpt-logs <id>                               # Endpoint logs

vastai show workergroups                                 # List worker groups
vastai create workergroup --name "wg" ...                # Create worker group
vastai update workergroup <id> ...                       # Update worker group
vastai update workers <id>                               # Rolling update of workers
vastai delete workergroup <id>                           # Delete worker group
vastai get wrkgrp-logs <id>                              # Worker group logs

vastai show deployments                                  # List deployments
vastai show deployment <id>                              # Deployment details
vastai show deployment-versions <id>                     # Version history
vastai delete deployment <id>                            # Delete deployment

vastai show scheduled-jobs                               # List scheduled jobs
vastai delete scheduled-job <id>                         # Delete scheduled job

模板

vastai search templates "pytorch"                        # Search templates
vastai create template --name "x" --image "img"          # Create template
vastai update template <hash_id> ...                     # Update template (edits create a new version)
vastai delete template <id>                              # Delete template

账户与 API 密钥

vastai set api-key <key>                                 # Save API key locally
vastai show api-key <id>                                 # Show a specific key
vastai show api-keys                                     # List all your API keys
vastai create api-key --name "ci" --permissions '{...}'  # Create restricted key
vastai delete api-key <id>                               # Delete key
vastai reset api-key                                     # Reset main key (get new from console)
vastai show user                                         # Account info, credit balance
vastai show audit-logs                                   # Account action history
vastai show connections                                  # Cloud storage connections
vastai show ipaddrs                                      # IP address history

账单

vastai show invoices-v1                                  # Charges/invoices (paginated)
vastai show invoices-v1 --charges                        # Charges only
vastai show invoices-v1 --invoices                       # Invoices only
vastai show invoices-v1 --start-date 2026-01-01 --end-date 2026-02-01
vastai show invoices-v1 --limit 50 --latest-first
vastai show deposit <id>                                 # Reserved instance deposit info

团队

vastai create team --team-name "myteam"                  # Create team; your key stays on your personal account
vastai show members                                      # List team members
vastai invite member --email [email protected]            # Invite member
vastai remove member <id>                                # Remove member
vastai create team-role --name "viewer" ...              # Create role
vastai show team-role <id>                               # Role details
vastai show team-roles                                   # List roles
vastai update team-role <id> ...                         # Update role
vastai remove team-role <id>                             # Remove role
vastai destroy team                                      # Delete team

环境变量

vastai show env-vars                                     # List user env vars
vastai create env-var KEY val                             # Create env var
vastai update env-var KEY newval                          # Update env var
vastai delete env-var KEY                                 # Delete env var

机器管理（主机）

vastai show machines                                     # List your machines
vastai list machine <id>                                 # List/unlist machine on marketplace
vastai show machine <id>                                 # Machine details
vastai cleanup machine <id>                              # Clean up machine state
vastai schedule maint <id> ...                           # Schedule maintenance window
vastai cancel maint <id>                                 # Cancel scheduled maintenance
vastai show maints                                       # List maintenance windows
vastai unlist machine <id>                               # Remove from marketplace

常见错误

错误	原因	修复方法
401 Unauthorized	API 密钥无效或已过期	vastai set api-key <new-key>
Insufficient credits	账户余额过低	在 https://cloud.vast.ai/billing/ 添加额度
No offers found	筛选条件过于严格	放宽筛选条件，尝试 --no-default/-n
Permission denied	未附加 SSH 密钥	在 create instance 前运行 vastai create ssh-key
Connection refused	实例尚未运行	轮询 show instance <id>，直到 actual_status == "running"
destroy instance 时卡住	确认提示正在等待输入	添加 -y 标志：vastai destroy instance <id> -y

URL

控制台

https://console.vast.ai/instances/   # Your instances
https://console.vast.ai/create/      # Search GPU offers
https://console.vast.ai/manage-keys/ # Create and manage API keys
https://cloud.vast.ai/billing/       # Billing

API

https://console.vast.ai/api/v0/instances/      # Instances endpoint
https://console.vast.ai/api/v0/asks/           # Offers search

实例端口

访问实例上公开的端口：

ssh://root@<ssh_host>:<ssh_port>               # SSH (from vastai ssh-url)

直接连接（创建时使用了 --direct）：实例 JSON 中的 <direct_port_end> 字段。

文档

在此获取完整的文档索引：https://docs.vast.ai/llms.txt