Seldon模型部署器Skill seldon-model-deployer

---

name: seldon-model-deployer description: 用于在Kubernetes上进行模型服务、A/B测试和金丝雀部署的Seldon Core部署技能。 allowed-tools:

• Read
• Write
• Bash
• Glob
• Grep

---

seldon-model-deployer

概述

用于在Kubernetes上进行模型服务、A/B测试、金丝雀部署和高级推理图的Seldon Core部署技能。

能力

• SeldonDeployment的创建和管理
• 多模型服务
• 流量分割（金丝雀/影子/A/B）
• 模型监控集成
• 自定义推理图
• 解释器部署（SHAP、Anchor）
• 请求日志记录和追踪
• 自动扩缩容配置

目标流程

• 带有金丝雀发布的模型部署流水线
• 用于机器学习模型的A/B测试框架
• 机器学习模型再训练流水线

工具和库

• Seldon Core
• Seldon Deploy
• Kubernetes
• Istio/Ambassador（入口）

输入模式

{
  "type": "object",
  "required": ["action"],
  "properties": {
    "action": {
      "type": "string",
      "enum": ["deploy", "update", "rollback", "delete", "status", "traffic-split"],
      "description": "要执行的Seldon操作"
    },
    "deploymentConfig": {
      "type": "object",
      "properties": {
        "name": { "type": "string" },
        "namespace": { "type": "string" },
        "modelUri": { "type": "string" },
        "implementation": { "type": "string" },
        "replicas": { "type": "integer" },
        "resources": {
          "type": "object",
          "properties": {
            "requests": { "type": "object" },
            "limits": { "type": "object" }
          }
        }
      }
    },
    "trafficConfig": {
      "type": "object",
      "properties": {
        "canaryPercent": { "type": "integer" },
        "shadowEnabled": { "type": "boolean" },
        "abTestEnabled": { "type": "boolean" }
      }
    },
    "explainerConfig": {
      "type": "object",
      "properties": {
        "type": { "type": "string", "enum": ["anchor_tabular", "anchor_text", "shap"] },
        "enabled": { "type": "boolean" }
      }
    }
  }
}

输出模式

{
  "type": "object",
  "required": ["status", "action"],
  "properties": {
    "status": {
      "type": "string",
      "enum": ["success", "error", "pending"]
    },
    "action": {
      "type": "string"
    },
    "deploymentName": {
      "type": "string"
    },
    "endpoint": {
      "type": "string"
    },
    "deploymentStatus": {
      "type": "string",
      "enum": ["creating", "available", "failed", "unknown"]
    },
    "replicas": {
      "type": "object",
      "properties": {
        "desired": { "type": "integer" },
        "ready": { "type": "integer" }
      }
    },
    "trafficSplit": {
      "type": "object"
    }
  }
}

使用示例

{
  kind: 'skill',
  title: '使用金丝雀部署模型',
  skill: {
    name: 'seldon-model-deployer',
    context: {
      action: 'deploy',
      deploymentConfig: {
        name: 'fraud-detector',
        namespace: 'ml-serving',
        modelUri: 'gs://models/fraud-v2',
        implementation: 'SKLEARN_SERVER',
        replicas: 3
      },
      trafficConfig: {
        canaryPercent: 10
      },
      explainerConfig: {
        type: 'shap',
        enabled: true
      }
    }
  }
}