integrations/vllm/prometheus_metadata.yaml

platforms:
  - type: GKE
    launch_stage: GA
    detections:
      - characteristic_metric:
          metric_type: prometheus.googleapis.com/vllm:prompt_tokens_total/counter
    exporter_metadata:
      name: vLLM Prometheus Exporter
      doc_url: https://docs.vllm.ai/en/v0.6.1/serving/metrics.html
      minimum_supported_version: v0.6.2
    default_metrics:
      - name: prometheus.googleapis.com/vllm:avg_generation_throughput_toks_per_s/gauge
        prometheus_name: avg_generation_throughput_toks_per_s
        kind: GAUGE
        value_type: DOUBLE
      - name: prometheus.googleapis.com/vllm:avg_prompt_throughput_toks_per_s/gauge
        prometheus_name: avg_prompt_throughput_toks_per_s
        kind: GAUGE
        value_type: DOUBLE
      - name: prometheus.googleapis.com/vllm:cache_config_info/gauge
        prometheus_name: cache_config_info
        kind: GAUGE
        value_type: DOUBLE
      - name: prometheus.googleapis.com/vllm:cache_config_info/unknown
        prometheus_name: cache_config_info
        kind: GAUGE
        value_type: DOUBLE
      - name: prometheus.googleapis.com/vllm:cache_config_info/unknown:counter
        prometheus_name: cache_config_info
        kind: CUMULATIVE
        value_type: DOUBLE
      - name: prometheus.googleapis.com/vllm:cpu_cache_usage_perc/gauge
        prometheus_name: cpu_cache_usage_perc
        kind: GAUGE
        value_type: DOUBLE
      - name: prometheus.googleapis.com/vllm:cpu_prefix_cache_hit_rate/gauge
        prometheus_name: cpu_prefix_cache_hit_rate
        kind: GAUGE
        value_type: DOUBLE
      - name: prometheus.googleapis.com/vllm:e2e_request_latency_seconds/histogram
        prometheus_name: e2e_request_latency_seconds
        kind: CUMULATIVE
        value_type: DISTRIBUTION
      - name: prometheus.googleapis.com/vllm:generation_tokens_total/counter
        prometheus_name: generation_tokens_total
        kind: CUMULATIVE
        value_type: DOUBLE
      - name: prometheus.googleapis.com/vllm:gpu_cache_usage_perc/gauge
        prometheus_name: gpu_cache_usage_perc
        kind: GAUGE
        value_type: DOUBLE
      - name: prometheus.googleapis.com/vllm:gpu_prefix_cache_hit_rate/gauge
        prometheus_name: gpu_prefix_cache_hit_rate
        kind: GAUGE
        value_type: DOUBLE
      - name: prometheus.googleapis.com/vllm:num_preemptions_total/counter
        prometheus_name: num_preemptions_total
        kind: CUMULATIVE
        value_type: DOUBLE
      - name: prometheus.googleapis.com/vllm:num_requests_running/gauge
        prometheus_name: num_requests_running
        kind: GAUGE
        value_type: DOUBLE
      - name: prometheus.googleapis.com/vllm:num_requests_swapped/gauge
        prometheus_name: num_requests_swapped
        kind: GAUGE
        value_type: DOUBLE
      - name: prometheus.googleapis.com/vllm:num_requests_waiting/gauge
        prometheus_name: num_requests_waiting
        kind: GAUGE
        value_type: DOUBLE
      - name: prometheus.googleapis.com/vllm:prompt_tokens_total/counter
        prometheus_name: prompt_tokens_total
        kind: CUMULATIVE
        value_type: DOUBLE
      - name: prometheus.googleapis.com/vllm:request_generation_tokens/histogram
        prometheus_name: request_generation_tokens
        kind: CUMULATIVE
        value_type: DISTRIBUTION
      - name: prometheus.googleapis.com/vllm:request_params_best_of/histogram
        prometheus_name: request_params_best_of
        kind: CUMULATIVE
        value_type: DISTRIBUTION
      - name: prometheus.googleapis.com/vllm:request_params_n/histogram
        prometheus_name: request_params_n
        kind: CUMULATIVE
        value_type: DISTRIBUTION
      - name: prometheus.googleapis.com/vllm:request_prompt_tokens/histogram
        prometheus_name: request_prompt_tokens
        kind: CUMULATIVE
        value_type: DISTRIBUTION
      - name: prometheus.googleapis.com/vllm:request_success_total/counter
        prometheus_name: request_success_total
        kind: CUMULATIVE
        value_type: DOUBLE
      - name: prometheus.googleapis.com/vllm:spec_decode_draft_acceptance_rate/gauge
        prometheus_name: spec_decode_draft_acceptance_rate
        kind: GAUGE
        value_type: DOUBLE
      - name: prometheus.googleapis.com/vllm:spec_decode_efficiency/gauge
        prometheus_name: spec_decode_efficiency
        kind: GAUGE
        value_type: DOUBLE
      - name: prometheus.googleapis.com/vllm:spec_decode_num_accepted_tokens_total/counter
        prometheus_name: spec_decode_num_accepted_tokens_total
        kind: CUMULATIVE
        value_type: DOUBLE
      - name: prometheus.googleapis.com/vllm:spec_decode_num_draft_tokens_total/counter
        prometheus_name: spec_decode_num_draft_tokens_total
        kind: CUMULATIVE
        value_type: DOUBLE
      - name: prometheus.googleapis.com/vllm:spec_decode_num_emitted_tokens_total/counter
        prometheus_name: spec_decode_num_emitted_tokens_total
        kind: CUMULATIVE
        value_type: DOUBLE
      - name: prometheus.googleapis.com/vllm:time_per_output_token_seconds/histogram
        prometheus_name: time_per_output_token_seconds
        kind: CUMULATIVE
        value_type: DISTRIBUTION
      - name: prometheus.googleapis.com/vllm:time_to_first_token_seconds/histogram
        prometheus_name: time_to_first_token_seconds
        kind: CUMULATIVE
        value_type: DISTRIBUTION
    install_documentation_url: https://cloud.google.com/stackdriver/docs/managed-prometheus/exporters/vllm