Helm快速安装

快速安装

默认安装不启用Cluster Agent，也不依赖TensorFusion控制台。Chart会自动创建NVIDIA ProviderConfig、SchedulingConfigTemplate和默认TensorFusionCluster，无需再手工应用集群配置清单。

第一步，使用Helm命令一键安装TensorFusion。

默认通过nvidia.com/gpu.present=true发现NVIDIA GPU节点，该标签通常由GPU Feature Discovery、云厂商组件或集群初始化脚本设置。无GPU的集群不需要添加该标签。

helm repo add tensor-fusion https://nexusgpu.github.io/tensor-fusion --force-update
helm repo update tensor-fusion

helm upgrade --install tensor-fusion-sys tensor-fusion/tensor-fusion \
  --version 1.8.3 --namespace tensor-fusion-sys --create-namespace \
  --wait --timeout 10m \
  -f https://download.tensor-fusion.ai/values-cn.yaml

helm upgrade --install tensor-fusion-sys tensor-fusion/tensor-fusion \
  --version 1.8.3 --namespace tensor-fusion-sys --create-namespace \
  --wait --timeout 10m

helm upgrade --install tensor-fusion-sys tensor-fusion/tensor-fusion \
  --version 1.8.3 --namespace tensor-fusion-sys --create-namespace \
  --wait --timeout 10m \
  --set agent.enrollToken=xxx --set agent.agentId=xxx \
  --set agent.cloudEndpoint=wss://your-own.domain/_ws

第二步，验证TensorFusion控制面是否安装成功。

kubectl get deployment,pods -n tensor-fusion-sys
kubectl get providerconfig,tensorfusioncluster,gpupool
kubectl get pods -n tensor-fusion-sys -l tensor-fusion.ai/component=hypervisor

# controller和alert-manager应为Ready/Running
# 默认资源包括nvidia-provider、tensor-fusion和tensor-fusion-shared
# 无GPU节点时，最后一条命令应返回No resources found

无GPU的集群只验证控制面即可。没有节点命中nvidia.com/gpu.present=true时，不会创建GPUNode或Hypervisor Pod，默认TFC/GPUPool可能保持Updating，这是预期行为。

有GPU的集群还应看到hypervisor-<节点名称>进入Running，并确认tensor-fusion-shared为Running。随后可部署一个Pytorch Pod端到端验证TensorFusion远程vGPU：

# simple-pytorch.yaml
# kubectl apply -f simple-pytorch.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: pytorch-example
  namespace: default
  labels:
    app: pytorch-example
    tensor-fusion.ai/enabled: 'true'
spec:
  replicas: 1
  selector:
    matchLabels:
      app: pytorch-example
  template:
    metadata:
      labels:
        app: pytorch-example
        tensor-fusion.ai/enabled: 'true'
      annotations:
        tensor-fusion.ai/inject-container: python
        tensor-fusion.ai/gpu-count: '1'
        tensor-fusion.ai/gpupool: tensor-fusion-shared
        tensor-fusion.ai/is-local-gpu: 'false'
        tensor-fusion.ai/isolation: soft
        tensor-fusion.ai/tflops-limit: '20'
        tensor-fusion.ai/tflops-request: '10'
        tensor-fusion.ai/vendor: NVIDIA
        tensor-fusion.ai/vram-limit: 4Gi
        tensor-fusion.ai/vram-request: 4Gi
    spec:
      containers:
        - name: python
          image: docker.m.daocloud.io/pytorch/pytorch:2.6.0-cuda12.4-cudnn9-runtime
          command:
            - sh
            - '-c'
            - sleep 1d
      restartPolicy: Always
      terminationGracePeriodSeconds: 0
      dnsPolicy: ClusterFirst

执行以下命令验证GPU资源分配：

kubectl exec deploy/pytorch-example -- nvidia-smi

# 预期显存为4Gi，而不是显卡的总显存数量

执行以下脚本，可在虚拟GPU中运行Qwen3 0.6B，验证推理结果

pip config set global.index-url https://pypi.mirrors.ustc.edu.cn/simple
pip install modelscope packaging transformers accelerate

cat << EOF >> test-qwen.py
from modelscope import AutoModelForCausalLM, AutoTokenizer

model_name = "Qwen/Qwen3-0.6B"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="cuda:0"
)

prompt = "Give me a short introduction to large language model."
messages = [
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
    enable_thinking=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=32768
)
output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist()
try:
    # rindex finding 151668 (</think>)
    index = len(output_ids) - output_ids[::-1].index(151668)
except ValueError:
    index = 0

thinking_content = tokenizer.decode(output_ids[:index], skip_special_tokens=True).strip("\n")
content = tokenizer.decode(output_ids[index:], skip_special_tokens=True).strip("\n")

print("thinking content:", thinking_content)
print("content:", content)
EOF

python3 test-qwen.py

卸载TensorFusion

运行如下命令一键卸载所有组件

# 可指定 KUBECONFIG 环境变量
curl -sfL https://download.tensor-fusion.ai/uninstall.sh | sh -

常见问题

如果集群有GPU，但hypervisor Pod未显示，请检查GPU节点是否带有nvidia.com/gpu.present=true标签。无GPU节点时不应创建hypervisor。

kubectl get nodes --show-labels | grep nvidia.com/gpu.present=true

# 预期找到GPU节点输出：
# gpu-node-name   Ready   <none>   42h   v1.32.1 beta.kubernetes.io/arch=amd64,...,kubernetes.io/os=linux,nvidia.com/gpu.present=true

节点缺少默认标签时，可以补充标签：

kubectl label node <gpu-node-name> nvidia.com/gpu.present=true

如果集群使用值为true的自定义GPU标签，需要同时修改operator初始扫描selector和默认GPU资源池的节点选择器：

helm upgrade --install tensor-fusion-sys tensor-fusion/tensor-fusion \
  --version 1.8.3 --namespace tensor-fusion-sys --create-namespace \
  --set initialGpuNodeLabelSelector="your-own-gpu-label-key=true" \
  --set cluster.pool.nodeSelectorKey="your-own-gpu-label-key"

节点隔离模式由GPUPool的nodeManagerConfig.defaultIsolationMode和有序的isolationModeRules统一控制，无需给Node添加tensor-fusion.ai/isolationMode标签。

快速安装

卸载TensorFusion

常见问题

On this page