LogoTensorFusion 文档
LogoTensorFusion 文档
首页文档

快速开始

TensorFusion 概览在 Kubernetes 安装在虚拟机/服务器安装(K3S)Helm 快速安装在宿主机/虚拟机安装TensorFusion 架构

应用操作

创建 AI 应用注解使用场景配置自动扩缩容迁移现有应用最佳实践

自定义AI基础设施

生产级部署指南QoS 级别与计价云厂商集成(BYOC)管理许可证

维护与优化

组件更新配置告警GPU 热迁移预加载模型优化 GPU 效率

故障排除

问题处理手册链路追踪/性能分析查询指标和日志

参考

对比

与 NVIDIA vGPU 对比与 MIG/MPS 对比与趋动科技对比与 NVIDIA Run:ai 对比与 HAMi 的对比

GPUPool

API documentation for GPUPool

Resource Information

FieldValue
API Versiontensor-fusion.ai/v1
KindGPUPool
ScopeCluster

Spec

GPUPoolSpec defines the desired state of GPUPool.

PropertyTypeDescription
capacityConfigobject
componentConfigobjectCustomize system components for seamless onboarding.
nodeManagerConfigobject
qosConfigobjectDefine different QoS and their price.
schedulingConfigTemplatestring

Status

GPUPoolStatus defines the observed state of GPUPool.

PropertyTypeDescription
availableTFlops *anypattern: Regex
availableVRAM *anypattern: Regex
budgetExceededstringIf the budget is exceeded, the set value in comma separated string to indicate which period caused the exceeding. If this field is not empty, scheduler will not schedule new AI workloads and stop scaling-up check.
clusterstring
componentStatus *objectwhen updating any component version or config, pool controller will perform rolling update. the status will be updated periodically, default to 5s, progress will be 0-100. when the progress is 100, the component version or config is fully updated.
conditionsarray
lastCompactionTimestring<date-time>
notReadyNodes *integer<int32>
phase *string(default: Pending) Allowed values: Pending, Running, Updating, Destroying, Unknown
potentialSavingsPerMonthstring
provisioningPhasestring(default: None) Allowed values: None, Initializing, Provisioning, Completed
readyNodes *integer<int32>
runningAppsCntinteger<int32>
savedCostsPerMonthstring
totalGPUsinteger<int32>
totalNodesinteger<int32>
totalTFlops *anypattern: Regex
totalVRAM *anypattern: Regex
virtualAvailableTFlopsanypattern: Regex
virtualAvailableVRAManypattern: Regex
virtualTFlops *anypattern: Regex
virtualVRAM *anypattern: Regex

目录

Resource Information
Spec
Status