# HELP omniserve_responses_total HTTP responses by status class. # TYPE omniserve_responses_total counter omniserve_responses_total{class="2xx"} 154 omniserve_responses_total{class="3xx"} 0 omniserve_responses_total{class="4xx"} 46 omniserve_responses_total{class="5xx"} 2 # HELP omniserve_gpu_degraded Embedded model requested GPU layers but runs on CPU. # TYPE omniserve_gpu_degraded gauge omniserve_gpu_degraded 0 # HELP omniserve_vram_free_gib Free device memory, -1 when the driver is unreachable. # TYPE omniserve_vram_free_gib gauge omniserve_vram_free_gib 6.484 omniserve_vram_total_gib 31.843 # HELP omniserve_access_log_dropped_total Access log lines dropped because the ring was full. # TYPE omniserve_access_log_dropped_total counter omniserve_access_log_dropped_total 0 # HELP omniserve_admission_slots Configured and in-use admission slots. # TYPE omniserve_admission_slots gauge omniserve_admission_slots{state="total"} 1 omniserve_admission_slots{state="used"} 0 # HELP omniserve_admission_timeouts_total Requests rejected after waiting. # TYPE omniserve_admission_timeouts_total counter omniserve_admission_timeouts_total{tier="paid"} 0 omniserve_admission_timeouts_total{tier="sub"} 0 omniserve_admission_timeouts_total{tier="free"} 0 omniserve_admission_timeouts_total{tier="background"} 0 # HELP omniserve_queue_ms_max Worst observed admission wait. # TYPE omniserve_queue_ms_max gauge omniserve_queue_ms_max{tier="paid"} 0.0 omniserve_queue_ms_max{tier="free"} 0.0 omniserve_capacity_spend_rate_usd_hr{lane="tts"} 0.000 omniserve_capacity_instances{lane="tts"} 0 omniserve_capacity_spend_rate_usd_hr{lane="stt"} 0.000 omniserve_capacity_instances{lane="stt"} 0 omniserve_capacity_spend_rate_usd_hr{lane="image"} 0.000 omniserve_capacity_instances{lane="image"} 0 omniserve_capacity_spend_rate_usd_hr{lane="multimodal"} 0.000 omniserve_capacity_instances{lane="multimodal"} 0 # HELP omniserve_vram_headroom_mb Device memory the broker would grant now. # TYPE omniserve_vram_headroom_mb gauge omniserve_vram_headroom_mb 5615 omniserve_vram_leased_mb 0 omniserve_vram_reserved_mb 0 omniserve_vram_leases 0 # HELP omniserve_vram_lease_total Broker lease outcomes. # TYPE omniserve_vram_lease_total counter omniserve_vram_lease_total{outcome="granted"} 0 omniserve_vram_lease_total{outcome="partial"} 0 omniserve_vram_lease_total{outcome="denied"} 0 omniserve_vram_lease_total{outcome="released"} 0 omniserve_vram_lease_total{outcome="expired"} 0 # HELP omniserve_llm_spec_tokens_total Speculative draft tokens by outcome. # TYPE omniserve_llm_spec_tokens_total counter omniserve_llm_spec_tokens_total{outcome="drafted"} 0 omniserve_llm_spec_tokens_total{outcome="accepted"} 0 # HELP omniserve_llm_spec_calls_saved_total Model calls speculation avoided. # TYPE omniserve_llm_spec_calls_saved_total counter omniserve_llm_spec_calls_saved_total 0 omniserve_llm_spec_rounds_total 0 omniserve_llm_spec_draft_max 0 # HELP omniserve_llm_engine_steps_total Batched decode steps, sampled rows, and tokens. # TYPE omniserve_llm_engine_steps_total counter omniserve_llm_engine_steps_total 0 omniserve_llm_engine_rows_total 0 omniserve_llm_engine_tokens_total 0 # HELP omniserve_overflow_total Requests sent to a standing remote endpoint. # TYPE omniserve_overflow_total counter omniserve_overflow_total{cause="saturated"} 0 omniserve_overflow_total{cause="local_failed"} 0 # HELP omniserve_frontier_kept_local_total Busy-lane requests the frontier policy queued locally. # TYPE omniserve_frontier_kept_local_total counter omniserve_frontier_kept_local_total 8 omniserve_upstream_breaker_open{upstream="image_overflow"} 0 omniserve_upstream_breaker_opens_total{upstream="image_overflow"} 0 omniserve_upstream_breaker_rejects_total{upstream="image_overflow"} 0 omniserve_upstream_failures_total{upstream="image_overflow"} 0 # HELP omniserve_guard_blocks_total Requests refused by the content guard. # TYPE omniserve_guard_blocks_total counter omniserve_guard_blocks_total{category="minor_sexual"} 0 omniserve_guard_blocks_total{category="noncon_sexual"} 0 omniserve_guard_blocks_total{category="realperson_sexual"} 0 # HELP omniserve_guard_names_loaded Real-person names loaded for the guard. # TYPE omniserve_guard_names_loaded gauge omniserve_guard_names_loaded 0 # HELP omniserve_guard_judge_calls_total Judge model invocations. # TYPE omniserve_guard_judge_calls_total counter omniserve_guard_judge_calls_total 0 # HELP omniserve_guard_judge_cache_hits_total Judge verdict cache hits. # TYPE omniserve_guard_judge_cache_hits_total counter omniserve_guard_judge_cache_hits_total 0 # HELP omniserve_guard_judge_invoked_total Requests passing the judge gate. # TYPE omniserve_guard_judge_invoked_total counter omniserve_guard_judge_invoked_total 0