Files

42 lines
1.6 KiB
YAML
Raw Permalink Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# -*- coding: utf-8 -*-
# 模板「推理后端适配层」配置资产示例:ti-cl4(氯化车间/海绵钛,Template-Ti 一期)。
#
# 说明:
# - 这是「推理后端选择」配置点(PRD 5.6 ⑥ 部署底座):
# 切换后端(NVIDIA 5090 GPU ↔ 华为昇腾 NPU)**只改本文件 backend 字段**,
# 业务代码零改动(统一走 InferenceBackend 接口);
# - backend: gpu —— NVIDIA 5090(Triton/ONNX,vLLM/Triton OpenAI 兼容接口)
# npu —— 华为昇腾(ACL/CANN,MindIE/onnxruntime-ascend)
# - endpoint 为空时进入 dry-run 模式(适配层就绪,便于离线验证/联调);
# - 该配置由 Helm 渲染注入(deploy/k8s/helm/iaop/values.yaml -> ConfigMap)。
template: ti-cl4
version: 1.0.0
# ---- 推理后端选择(切换后端只改这一处) ----
backend: gpu # gpu | npu
inference:
# 与 Helm values.inference 保持同构,便于一键部署时直接覆盖
model: iaop-ti-cl4-v1
endpoint: http://iaop-inference.iaop.svc.cluster.local:8000/v1
timeout_seconds: 60
runtime: vllm # gpu: vllm|triton;npu: mindie|onnx-ascend
device: nvidia-5090 # gpu: nvidia-5090;npu: ascend-310p|ascend-910b
cann_version: "" # npu: CANN 工具链版本(如 8.0),gpu 忽略
max_tokens: 1024
temperature: 0.1
# ---- 资源配额(Helm 部署时以此为默认值) ----
resources:
requests:
cpu: "2"
memory: 8Gi
limits:
cpu: "8"
memory: 32Gi
# ---- 灰度发布策略(ArgoCD 滚动发布参数,PRD 5.6 配置点) ----
rollingUpdate:
maxUnavailable: 0
maxSurge: 1