一、命名空间
Namespace = QCE/TOKEN_HUB
二、监控指标
| 指标名 | 指标中文名 | 说明 | 单位 | 维度 | 统计规则 |
| CacheCreationToken | 每分钟写入缓存 token 数 | anthropics 协议下每分钟创建缓存的 token 数 | Count | apikeyid endpoint modelname uin |
[ 60s, sum ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| CacheHitRatio | 缓存命中率 | 请求缓存命中 token 数 / 输入总 token 数 | % | apikeyid endpoint modelname uin |
[ 60s, expr ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| CacheToken | 每分钟读缓存 token 数 | 单位时间内命中缓存的 token 数 | Count | apikeyid endpoint modelname uin |
[ 60s, sum ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| CanceledRatio | 超时错误率 | 因超时等原因触发的取消请求 / 总调用次数 | % | apikeyid endpoint modelname uin |
[ 60s, expr ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| ClientErrorRatio | 客户端错误率 | 4xx 请求数 / 总请求数 | % | apikeyid endpoint modelname uin |
[ 60s, expr ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| ErrorRatio | 错误率 | HTTP 状态码非200请求数 / 总请求数 | % | apikeyid endpoint modelname uin |
[ 60s, expr ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| ExplicitCacheToken | 读显式缓存 token 数 | 每分钟命中显式缓存的 token 数 | Count | apikeyid endpoint modelname uin |
[ 60s, sum ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| Http429Count | HTTP 429错误总数 | HTTP 429 限流错误请求总数 | Count | apikeyid endpoint modelname uin |
[ 60s, sum ] [ 300s, sum ] [ 3600s, sum ] [ 86400s, sum ] |
| InputAvgToken | 平均输入 Token | 每请求输入 Token 的平均值 | Tokens | apikeyid endpoint modelname uin |
[ 60s, expr ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| InputTotalToken | 总输入 Token 数 | 区间内总的输入 Token 数（包含 cache） | Count | apikeyid endpoint modelname uin |
[ 60s, sum ] [ 300s, sum ] [ 3600s, sum ] [ 86400s, sum ] |
| InputTpm | 每分钟输入 Token 数 | 1分钟内总的输入 Token 数（包含 cache，如果时间粒度是5分钟，是5分钟下每1分钟的平均值） | tokens/min | apikeyid endpoint modelname uin |
[ 60s, sum ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| ModelErrorCount | 调用失败次数 | 模型调用失败次数 | Count | apikeyid endpoint modelname uin |
[ 60s, sum ] [ 300s, sum ] [ 3600s, sum ] [ 86400s, sum ] |
| ModelInputAvgToken | 平均输入 Token | 每请求输入 Token 的平均值 | Tokens | apikeyid endpoint modelname uin |
[ 60s, expr ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| ModelInputTotalToken | 输入总 Token | 输入总 Token | Count | apikeyid endpoint modelname uin |
[ 60s, sum ] [ 300s, sum ] [ 3600s, sum ] [ 86400s, sum ] |
| ModelInputTpm | 输入 TPM | 输入 TPM | tokens/min | apikeyid endpoint modelname uin |
[ 60s, sum ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| ModelOutputAvgToken | 平均输出 Token | 每请求输出 Token 的平均值 | Tokens | apikeyid endpoint modelname uin |
[ 60s, expr ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| ModelOutputTotalToken | 输出总 Token | 输出总 Token | Count | apikeyid endpoint modelname uin |
[ 60s, sum ] [ 300s, sum ] [ 3600s, sum ] [ 86400s, sum ] |
| ModelOutputTpm | 输出 TPM | 输出 TPM | tokens/min | apikeyid endpoint modelname uin |
[ 60s, sum ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| ModelRpm | 请求数 | 每分钟请求数 | reqs/min | apikeyid endpoint modelname uin |
[ 60s, sum ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| ModelSuccessCount | 调用成功次数 | 模型调用成功次数 | Count | apikeyid endpoint modelname uin |
[ 60s, expr ] [ 300s, sum ] [ 3600s, sum ] [ 86400s, sum ] |
| ModelTotalRpm | 调用次数 | 模型调用次数 | Count | apikeyid endpoint modelname uin |
[ 60s, sum ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| ModelTotalToken | 总 Token 数 | 总 Token | Count | apikeyid endpoint modelname uin |
[ 60s, sum ] [ 300s, sum ] [ 3600s, sum ] [ 86400s, sum ] |
| ModelTpm | 总 TPM | 总 TPM | tokens/min | apikeyid endpoint modelname uin |
[ 60s, sum ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| ModelTpmQuota | 模型 TPM 配额 | 模型每分钟可用的 token 配额 | tokens/min | apikeyid endpoint modelname uin |
[ 60s, last ] [ 300s, last ] [ 3600s, last ] [ 86400s, last ] |
| ModelTpot | 每输出 Token 延时 | 非首 Token 延时 | ms | apikeyid endpoint modelname uin |
[ 60s, avg ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| ModelTtft | 首 Token 延时 | 首 Token 延时 | ms | apikeyid endpoint modelname uin |
[ 60s, avg ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| NonStreamLatency | 非流式接口延时 | 单位时间内针对非流式请求的 latency，端到端耗时 | ms | apikeyid endpoint modelname uin |
[ 60s, avg ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| OutputAvgToken | 平均输出 Token | 每请求输出 Token 的平均值 | Tokens | apikeyid endpoint modelname uin |
[ 60s, expr ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| OutputTotalToken | 总输出 Token 数 | 区间内总的输出 Token 数 | Count | apikeyid endpoint modelname uin |
[ 60s, sum ] [ 300s, sum ] [ 3600s, sum ] [ 86400s, sum ] |
| OutputTpm | 每分钟输出 Token 数 | 1分钟内总的输出 Token 数（如果时间粒度是5分钟，是5分钟下每1分钟的平均值） | tokens/min | apikeyid endpoint modelname uin |
[ 60s, sum ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| RateLimitedRatio | 429限流错误率 | HTTP429限流错误请求占比 | % | apikeyid endpoint modelname uin |
[ 60s, expr ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| Rpm | 每分钟请求数 | 单位时间内请求的数量 | reqs/min | apikeyid endpoint modelname uin |
[ 60s, sum ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| RpmOverRatio | RPM 限流率 | 限流次数 / 总调用次数 | % | apikeyid endpoint modelname uin |
[ 60s, expr ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| ServerErrorRatio | 服务端错误率 | 5xx 请求数 / 总请求数 | % | apikeyid endpoint modelname uin |
[ 60s, expr ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| SuccessRatio | 成功率 | 成功请求数 / 总请求数 | % | apikeyid endpoint modelname uin |
[ 60s, expr ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| TotalCacheCreationToken | 写入缓存总 Token 数 | anthropics 协议下区间内创建缓存的总 Token 数 | Count | apikeyid endpoint modelname uin |
[ 60s, sum ] [ 300s, sum ] [ 3600s, sum ] [ 86400s, sum ] |
| TotalCacheToken | 读缓存总 Token 数 | 区间内命中缓存的总 Token 数 | Count | apikeyid endpoint modelname uin |
[ 60s, sum ] [ 300s, sum ] [ 3600s, sum ] [ 86400s, sum ] |
| TotalExplicitCacheToken | 读显式缓存总 Token 数 | 区间内命中显式缓存的总 Token 数 | Count | apikeyid endpoint modelname uin |
[ 60s, sum ] [ 300s, sum ] [ 3600s, sum ] [ 86400s, sum ] |
| TotalToken | 总 Token 数 | 区间内总的 token 数 | Count | apikeyid endpoint modelname uin |
[ 60s, sum ] [ 300s, sum ] [ 3600s, sum ] [ 86400s, sum ] |
| Tpm | 每分钟总 Token 数 | 1分钟内总的 token 数，对应 TPM（如果时间粒度是5分钟，是5分钟下每1分钟的平均值） | tokens/min | apikeyid endpoint modelname uin |
[ 60s, sum ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| TpmQuotaUsageRatio | TPM 配额使用率 | 模型 TPM 配额使用占比 | % | apikeyid endpoint modelname uin |
[ 60s, expr ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| Tpot | 每输出 Token 延时 | 单位时间内每 token 输出延时平均值 | ms | apikeyid endpoint modelname uin |
[ 60s, avg ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
| Ttft | 首 Token 延时 | 单位时间内流式请求的 ttft 平均值 | ms | apikeyid endpoint modelname uin |
[ 60s, avg ] [ 300s, avg ] [ 3600s, avg ] [ 86400s, avg ] |
三、各维度对应参数总览
| 参数名称 | 维度名称 | 维度解释 | 格式 |
| Instances.N.Dimensions.0.Name | uin | 腾讯云账号 UIN 的维度名称 | 输入 String 类型维度名称：uin |
| Instances.N.Dimensions.0.Value | uin | 腾讯云账号 UIN | 输入腾讯云账号 UIN，例如：100012345678 |
| Instances.N.Dimensions.1.Name | modelname | 大模型 ID 的维度名称 | 输入 String 类型维度名称：modelname |
| Instances.N.Dimensions.1.Value | modelname | 大模型的具体模型 ID | 输入大模型 ID，例如：deepseek-v3 |
| Instances.N.Dimensions.2.Name | endpoint | 大模型服务接入点的维度名称 | 输入 String 类型维度名称：endpoint |
| Instances.N.Dimensions.2.Value | endpoint | 大模型服务的具体接入点 | 输入大模型服务接入点，例如：glm-5.2 |
| Instances.N.Dimensions.3.Name | apikeyid | API Key ID 的维度名称 | 输入 String 类型维度名称：apikeyid |
| Instances.N.Dimensions.3.Value | apikeyid | 调用大模型所使用的 API Key ID | 输入 API Key ID，例如：sk-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx |
四、入参说明
大模型服务平台 TokenHub 提供了获取以下六种维度组合的监控数据：账号级别、账号-模型级别、账号-接入点级别、账号-接入点-模型级别、账号-接入点-API Key 级别、账号-模型-接入点-API Key 级别。
查询「账号」维度的监控指标数据，入参取值如下：
&Namespace=QCE/TOKEN_HUB
&Instances.N.Dimensions.0.Name=uin
&Instances.N.Dimensions.0.Value=腾讯云账号 UIN
查询「账号-模型」维度的监控指标数据，入参取值如下：
&Namespace=QCE/TOKEN_HUB
&Instances.N.Dimensions.0.Name=uin
&Instances.N.Dimensions.0.Value=腾讯云账号 UIN
&Instances.N.Dimensions.1.Name=modelname
&Instances.N.Dimensions.1.Value=大模型的具体模型 id
查询「账号-接入点」维度的监控指标数据，入参取值如下：
&Namespace=QCE/TOKEN_HUB
&Instances.N.Dimensions.0.Name=uin
&Instances.N.Dimensions.0.Value=腾讯云账号 UIN
&Instances.N.Dimensions.1.Name=endpoint
&Instances.N.Dimensions.1.Value=大模型服务的具体接入点
查询「账号-接入点-模型」维度的监控指标数据，入参取值如下：
&Namespace=QCE/TOKEN_HUB
&Instances.N.Dimensions.0.Name=uin
&Instances.N.Dimensions.0.Value=腾讯云账号 UIN
&Instances.N.Dimensions.1.Name=modelname
&Instances.N.Dimensions.1.Value=大模型的具体模型 id
&Instances.N.Dimensions.2.Name=endpoint
&Instances.N.Dimensions.2.Value=大模型服务的具体接入点
查询「账号-接入点-API Key」维度的监控指标数据，入参取值如下：
&Namespace=QCE/TOKEN_HUB
&Instances.N.Dimensions.0.Name=uin
&Instances.N.Dimensions.0.Value=腾讯云账号 UIN
&Instances.N.Dimensions.1.Name=endpoint
&Instances.N.Dimensions.1.Value=大模型服务的具体接入点
&Instances.N.Dimensions.2.Name=apikeyid
&Instances.N.Dimensions.2.Value=调用大模型所使用的 API Key ID
查询「账号-模型-接入点-API Key」维度的监控指标数据，入参取值如下：（最细粒度组合，可精确定位到某个 API Key 在指定接入点调用某个模型的监控数据）
&Namespace=QCE/TOKEN_HUB
&Instances.N.Dimensions.0.Name=uin
&Instances.N.Dimensions.0.Value=腾讯云账号 UIN
&Instances.N.Dimensions.1.Name=modelname
&Instances.N.Dimensions.1.Value=大模型的具体模型 id
&Instances.N.Dimensions.2.Name=endpoint
&Instances.N.Dimensions.2.Value=大模型服务的具体接入点
&Instances.N.Dimensions.3.Name=apikeyid
&Instances.N.Dimensions.3.Value=调用大模型所使用的 API Key ID