Setup custom Prometheus metric with Kubernetes labels for TorchServe

Question

I'm trying to setup http_request_duration_seconds metric the same as provides prometheus-net

.NET implementation somehow adds dimensions like namespace, service, pod, and code that could be used in Grafana:

// ...
"targets": [
  {
    "datasource": {
      "type": "prometheus",
      "uid": "${datasource}"
    },
    "exemplar": true,
    "expr": "sum(rate(http_request_duration_seconds_count{namespace=\"$namespace\", service=~\"$service\", pod=~\"$pod\"}[$interval])) by (code)",
    "interval": "",
    "legendFormat": "{{ code }}",
    "queryType": "randomWalk",
    "refId": "A"
  }
],
"title": "Requests per second",
"type": "timeseries"
// ...

It seems like I could extract these values from context. I've read the article related to k8s and TorchServe and didn't find anything related to my problem.

A piece of code related to the metric:

http_request_duration_seconds_count = self.metrics.get_metric(
    metric_name='http_request_duration_seconds',
    metric_type=MetricTypes.HISTOGRAM)
start = time.monotonic()

request = data[0]
dataset = self.preprocess(request)
inferenced = self.inference(dataset)
result = self.postprocess(inferenced)

http_request_duration_seconds_count.add_or_update(
    value=time.monotonic() - start,
    # dimension_values=['namespace_name', 'service_service', 'pod_name']
)
emit_metrics(self.metrics.store)