bench_latency

{“what”:“Unique numeric identifier for this measurement row.”,“why”:“Without it, rows cannot be distinguished or updated individually.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T00:02:00.833174818+00:00”}

{“what”:“UUID of the benchmark run this measurement belongs to.”,“why”:“Without it, measurements cannot be grouped by run.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T00:02:00.856979631+00:00”}

{“what”:“Timestamp when the measurement was recorded, in UTC.”,“why”:“Without it, time-based ordering and correlation is lost.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T00:02:00.868386159+00:00”}

{“what”:“Name of the inference provider (e.g., OpenAI).”,“why”:“Without it, cannot compare latency across providers.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T00:02:00.884298770+00:00”}

{“what”:“Model identifier as used by the provider (e.g., gpt-4o).”,“why”:“Without it, cannot compare latency across models.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T00:02:00.910463477+00:00”}

{“what”:“Benchmark phase name (e.g., prefill, decode).”,“why”:“Without it, latency metrics cannot be attributed to a stage.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T00:02:00.921379114+00:00”}

{“what”:“Number of tokens in the input prompt.”,“why”:“Without it, latency cannot be normalized per token.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T00:02:00.939854298+00:00”}

{“what”:“Time to first token in milliseconds.”,“why”:“Without it, perceived responsiveness is unknown.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T00:02:00.955337542+00:00”}

{“what”:“Decode throughput in tokens per second.”,“why”:“Without it, generation speed cannot be assessed.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T00:02:00.969143945+00:00”}

{“what”:“Time per output token in milliseconds.”,“why”:“Without it, per-token generation cost is unknown.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T00:02:00.997001395+00:00”}

{“what”:“Median inter-token latency in milliseconds.”,“why”:“Without it, variability in token generation is not captured.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T00:02:01.010669495+00:00”}

{“what”:“99th percentile inter-token latency in milliseconds.”,“why”:“Without it, tail latency outliers are invisible.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T00:02:01.024509100+00:00”}

{“what”:“Total request duration in milliseconds.”,“why”:“Without it, end-to-end latency cannot be compared.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T00:02:01.047238808+00:00”}

{“what”:“Number of tokens generated in the completion.”,“why”:“Without it, output length is unknown, affecting throughput.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T00:02:01.084281099+00:00”}

{“what”:“HTTP-like status code; 0 success, non-zero error.”,“why”:“Without it, failed requests would be indistinguishable.”,“by”:“fleet-annotator-worker”,“at”:“2026-09-23T00:02:01.100215011+00:00”}