Model Market

后端 API

模型市场没有专门的后端模块。它的目录是 src/data/modelCatalog.js 中的静态前端数据，因此没有需要文档化的后端目录 API。该页面只调用共享的集群和部署 API。

列出 Lens 管理的集群

列出 Lens 管理的集群，可选择包含尚未完成的草稿记录。

REST API
Python API

GET /api/cluster/clusters

请求：

查询参数：
- includeDrafts（boolean，可选）：是否包含尚未完成的草稿集群记录。

该端点没有请求体。

响应：

返回一个 ClustersResponse 对象，其中包含一个 ClusterRecord 对象数组 items。

{
  "items": [
    {
      "id": "cluster-123",
      "name": "lab-cluster",
      "description": "Shared Intel GPU cluster",
      "createdAt": "2026-09-20T10:15:00+00:00",
      "ready": true,
      "proxy": {
        "mode": "auto",
        "httpProxy": null,
        "httpsProxy": null,
        "noProxy": null
      },
      "llmDRef": "v0.9.0",
      "llmDBenchmarkRef": "main",
      "llmDRepoPath": "/opt/prism/checkouts/llm-d",
      "llmDBenchmarkRepoPath": "/opt/prism/checkouts/llm-d-benchmark",
      "deploymentSource": {
        "resolved_from": "cluster-registry",
        "ref": "v0.9.0"
      },
      "benchmarkSource": {
        "resolved_from": "cluster-registry",
        "ref": "main"
      },
      "draft": false
    }
  ]
}

import asyncio

from llm_d_lens.cluster import service

async def main() -> None:
    clusters = service.list_clusters(include_drafts=True)
    readiness = await asyncio.gather(*(service.cluster_is_ready(cluster) for cluster in clusters))

    for cluster, is_ready in zip(clusters, readiness):
        print(cluster.id, cluster.name, is_ready)

asyncio.run(main())

获取集群概览

获取 Lens 管理的集群的实时 Kubernetes 概览，包括节点、GPU、缓存镜像和组件健康状况摘要。

REST API
Python API

GET /api/cluster/overview

请求：

查询参数：
- clusterId（string，必需）：要查看的 Lens 管理集群 id。

该端点没有请求体。

响应：

返回一个 OverviewResponse，其中包含集群记录、获取时间戳以及一份 KubernetesSummary 快照。

{
  "cluster": {
    "id": "cluster-123",
    "name": "lab-cluster",
    "description": "Shared Intel GPU cluster",
    "createdAt": "2026-09-20T10:15:00+00:00",
    "ready": true,
    "proxy": {
      "mode": "auto",
      "httpProxy": null,
      "httpsProxy": null,
      "noProxy": null
    },
    "llmDRef": "v0.9.0",
    "llmDBenchmarkRef": "main",
    "llmDRepoPath": "/opt/prism/checkouts/llm-d",
    "llmDBenchmarkRepoPath": "/opt/prism/checkouts/llm-d-benchmark",
    "deploymentSource": {
      "resolved_from": "cluster-registry",
      "ref": "v0.9.0"
    },
    "benchmarkSource": {
      "resolved_from": "cluster-registry",
      "ref": "main"
    },
    "draft": false
  },
  "fetchedAt": "2026-09-23T09:30:00+00:00",
  "kubernetes": {
    "components": {
      "intelDevicePlugin": {
        "installed": true,
        "ready": 1,
        "desired": 1,
        "status": "ready"
      },
      "monitoring": {
        "installed": true,
        "ready": 5,
        "desired": 5,
        "status": "ready"
      }
    },
    "nodes": [
      {
        "name": "worker-1",
        "version": "v1.30.2",
        "osImage": "Ubuntu 22.04.5 LTS",
        "ready": true,
        "schedulingDisabled": false,
        "cpu": 16.0,
        "cpuUsagePercent": 25.0,
        "memoryBytes": 68719476736,
        "memoryUsagePercent": 50.0,
        "gpu": true,
        "gpuCount": 2,
        "gpuUsagePercent": 80.0,
        "vramBytes": 103079215104,
        "vramUsagePercent": 40.0,
        "role": "worker",
        "cachedImages": [
          "ghcr.io/llm-d/llm-d-xpu:v0.9.0"
        ],
        "gpus": [
          {
            "id": "gpu-0",
            "index": 0,
            "name": "GPU #0",
            "pciAddress": "0000:03:00.0",
            "computeUsagePercent": 80.0,
            "vramTotalBytes": 51539607552,
            "vramUsedBytes": 21474836480,
            "vramUsagePercent": 41.6,
            "vramReadThroughputBytes": 10000,
            "vramWriteThroughputBytes": 20000,
            "vramTotalThroughputBytes": 30000,
            "vramBandwidthPercent": 15.0
          }
        ],
        "diskBytes": 1099511627776,
        "diskUsedBytes": 274877906944,
        "diskUsagePercent": 25.0
      }
    ],
    "namespaces": [
      {
        "name": "standard-qwen3",
        "deployments": [
          {
            "name": "qwen3-vllm",
            "desired": 1,
            "ready": 1,
            "available": 1
          }
        ],
        "ready": 1
      }
    ],
    "hardware": {
      "nodes": 1,
      "cpuCores": 16.0,
      "cpuUsagePercent": 25.0,
      "memoryBytes": 68719476736,
      "memoryUsagePercent": 50.0,
      "gpuNodes": 1,
      "gpuCount": 2,
      "gpuUsagePercent": 80.0,
      "vramBytes": 103079215104,
      "vramUsagePercent": 40.0,
      "diskBytes": 1099511627776,
      "diskUsagePercent": 25.0,
      "allocatedGpuCount": 1,
      "availableGpuCount": 1,
      "usableGpuCount": 1,
      "configuredGpuLimit": null
    }
  }
}

import asyncio

from llm_d_lens.cluster import registry, service


async def main() -> None:
    cluster = registry.require_cluster("cluster-123")
    overview = await service.build_overview(cluster)
    ready = await service.cluster_is_ready(cluster)

    print(cluster.id, ready)
    print(overview["hardware"])


asyncio.run(main())

启动一次 Standard Kubernetes Service 部署运行

使用 Lens 的白名单 vLLM 契约启动一次 Standard Kubernetes Service 部署运行。

备注

**One-click deployment（一键部署）**页面已不再提供 Standard 模式，只能通过 **Manual（手动，Guide 模板）**或 **Agentic（智能体）**部署。该接口仍然保留给希望不使用 Guide 模板、直接调用白名单 vLLM 契约的 API 和 MCP 调用方；它仍是模型市场领域使用的部署契约，因此在此处记录。

REST API
Python API

POST /api/v1/deployments/standard-vllm-runs

请求：

JSON 请求体字段：
- model（string，必需）
- deployment_name（string，必需）
- description（string，可选）
- cluster_session_id（string，必需，36 个字符）
- replicas（integer，必需）
- tensor_parallel_size（integer，必需）
- storage_type（"local-cache" | "pvc" | "model-cache"，可选）
- storage_volume_id（string，可选）
- model_cache_path（string，可选）
- pvc_name（string，可选）
- max_model_len（integer，可选）
- gpu_memory_utilization（number，可选）
- max_num_seqs（integer，可选）
- max_num_batched_tokens（integer，可选）
- enable_prefix_caching（boolean，可选）
- enforce_eager（boolean，可选）
- vllm_arguments（array<object>，可选；每项使用 name 和 value 两个字符串键）
- use_cluster_proxy（boolean，可选）
- http_proxy（string，可选）
- https_proxy（string，可选）
- no_proxy（string，可选）

{
  "model": "Qwen/Qwen3-8B",
  "deployment_name": "qwen3-8b-standard",
  "description": "Model Market launch",
  "cluster_session_id": "123e4567-e89b-12d3-a456-426614174000",
  "replicas": 2,
  "tensor_parallel_size": 2,
  "storage_type": "local-cache",
  "model_cache_path": "/home/prism/.cache",
  "max_model_len": 8192,
  "gpu_memory_utilization": 0.85,
  "max_num_seqs": 128,
  "max_num_batched_tokens": 4096,
  "enable_prefix_caching": true,
  "enforce_eager": false,
  "vllm_arguments": [
    {
      "name": "enable-auto-tool-choice",
      "value": "true"
    }
  ],
  "use_cluster_proxy": true,
  "http_proxy": "http://proxy.example:911",
  "https_proxy": "http://proxy.example:911",
  "no_proxy": "localhost,.svc"
}

响应：

返回一个处于排队状态的 DeploymentRun。其中的 source_configurations 条目是根据白名单 Standard 请求生成的不可变 DeployableConfiguration。

{
  "id": "run-123",
  "status": "queued",
  "source_configurations": [
    {
      "schema_version": "deployable-configuration.v1",
      "type": "baseline",
      "format": "helm",
      "provider_ref": "baseline-vllm",
      "checksum": "sha256:1d1c9c5b7f4d",
      "content": {
        "model": {
          "name": "Qwen/Qwen3-8B"
        },
        "decode": {
          "replicaCount": 2,
          "tensorParallelSize": 2,
          "maxModelLen": 8192
        },
        "runtime": {
          "image": "ghcr.io/llm-d/llm-d-xpu:v0.9.0",
          "imageMode": "pull-existing",
          "modelSource": "auto-cache",
          "mountPath": "/home/prism/.cache",
          "environment": {
            "HTTP_PROXY": "http://proxy.example:911",
            "HTTPS_PROXY": "http://proxy.example:911",
            "NO_PROXY": "localhost,.svc,127.0.0.1,.cluster.local"
          }
        },
        "customParameters": [
          {
            "target": "decode",
            "kind": "argument",
            "name": "gpu-memory-utilization",
            "value": "0.85"
          },
          {
            "target": "decode",
            "kind": "argument",
            "name": "enable-prefix-caching",
            "value": "true"
          }
        ]
      },
      "provenance": {
        "source": "standard-kubernetes-service.v1",
        "deployment_name": "qwen3-8b-standard",
        "description": "Model Market launch",
        "storage_type": "local-cache",
        "namespace_policy": {
          "prefix": "standard-"
        }
      }
    }
  ],
  "cases": [],
  "failure_policy": "continue",
  "provenance": {
    "deployment_source": {
      "kind": "standard-kubernetes-service"
    },
    "cluster_session_id": "123e4567-e89b-12d3-a456-426614174000",
    "deployment_name": "qwen3-8b-standard",
    "description": "Model Market launch",
    "model_market": {
      "deployment_name": "qwen3-8b-standard",
      "description": "Model Market launch"
    }
  },
  "created_at": "2026-09-23T09:35:00+00:00",
  "started_at": null,
  "finished_at": null
}

import asyncio

from llm_d_lens.deploy.application import deployment_run_manager
from llm_d_lens.deploy.contracts import DeploymentRunCreateRequest
from llm_d_lens.deploy.standard_kubernetes_service import (
    StandardKubernetesServiceRequest,
    build_standard_kubernetes_service_configuration,
)


async def main() -> None:
    request = StandardKubernetesServiceRequest(
        model="Qwen/Qwen3-8B",
        deployment_name="qwen3-8b-standard",
        description="Model Market launch",
        cluster_session_id="123e4567-e89b-12d3-a456-426614174000",
        replicas=2,
        tensor_parallel_size=2,
        storage_type="local-cache",
        model_cache_path="/home/prism/.cache",
        max_model_len=8192,
        gpu_memory_utilization=0.85,
        max_num_seqs=128,
        max_num_batched_tokens=4096,
        enable_prefix_caching=True,
        vllm_arguments=[{"name": "enable-auto-tool-choice", "value": "true"}],
        use_cluster_proxy=True,
        http_proxy="http://proxy.example:911",
        https_proxy="http://proxy.example:911",
        no_proxy="localhost,.svc",
    )
    configuration = build_standard_kubernetes_service_configuration(request)
    run = await deployment_run_manager.start_run(
        DeploymentRunCreateRequest(
            configurations=[configuration],
            provenance={
                "deployment_source": {"kind": "standard-kubernetes-service"},
                "cluster_session_id": request.cluster_session_id,
                "deployment_name": request.deployment_name,
                "description": request.description.strip(),
                "model_market": {
                    "deployment_name": request.deployment_name,
                    "description": request.description.strip(),
                },
            },
        )
    )
    print(run.id, run.status)


asyncio.run(main())

模型市场并不暴露自己的目录 API；完整的后端相关模块页面请参见 Clusters 和 Deployments。

完整的请求/响应模式请参见 FastAPI 的 OpenAPI schema（使用 npm run docs:api 重新生成）。

后端 API​

列出 Lens 管理的集群​

获取集群概览​

启动一次 Standard Kubernetes Service 部署运行​

后端 API

列出 Lens 管理的集群

获取集群概览

启动一次 Standard Kubernetes Service 部署运行