Missing JSON fields

[Bslabe123]

PR #832 introduced a bug where the folllowing keys werent set on 'metrics', this PR fixes this.

    "benchmark_time": 44.91958165168762,
    "total_output_token": 58986,
    "output_tokens_per_min": 78788.80145062602,
    "total_input_tokens": 39871,
    "input_tokens_per_min": 53256.506673412514,
    "total_tokens": 98857,
    "tokens_per_min": 132045.30812403854,
    "avg_latency": 7.175992427666982,
    "avg_per_token_latency": 0.06020041991300422,
    "avg_per_output_token_latency": 0.35792783426739533,
    "avg_input_len": 66.45166666666667,
    "avg_output_len": 98.31

Example fixed output JSON:

{
  "metrics": {
    "num_prompts": 600,
    "request_rate": 20,
    "server_metrics": {
      "jetstream_slots_used_percentage": {
        "Mean": "0.80625",
        "Median": "1",
        "Std": "0.3875",
        "Min": "0.03125",
        "Max": "1",
        "P90": "1",
        "P99": "1"
      },
      "jetstream_prefill_backlog_size": {
        "Mean": "55.4",
        "Median": "59",
        "Std": "42.846703490467036",
        "Min": "0",
        "Max": "124",
        "P90": "103.2",
        "P99": "121.91999999999999"
      },
      "jetstream_time_to_first_token": {
        "Mean": "0.13120518984972915",
        "Median": "0.16634615384615387",
        "Min": "0.01",
        "Max": "0.25",
        "P90": "0.23326923076923076",
        "P99": "0.24832692307692308"
      },
      "jetstream_time_per_output_token": {
        "Mean": "0",
        "Median": "0.028433476394849787",
        "Min": "0.01",
        "Max": "2.5",
        "P90": "0.06770270270270269",
        "P99": "2.5"
      },
      "jetstream_time_per_request": {
        "Mean": "2.4746437307432325",
        "Median": "2.1858638743455496",
        "Min": "1",
        "Max": "10",
        "P90": "5.3442622950819665",
        "P99": "9.5344262295082"
      }
    },
    "benchmark_time": 44.91958165168762,
    "total_output_token": 58986,
    "output_tokens_per_min": 78788.80145062602,
    "total_input_tokens": 39871,
    "input_tokens_per_min": 53256.506673412514,
    "total_tokens": 98857,
    "tokens_per_min": 132045.30812403854,
    "avg_latency": 7.175992427666982,
    "avg_per_token_latency": 0.06020041991300422,
    "avg_per_output_token_latency": 0.35792783426739533,
    "avg_input_len": 66.45166666666667,
    "avg_output_len": 98.31
  },
  "dimensions": {
    "date": "20240925-163414",
    "backend": "jetstream",
    "model_id": "google/gemma-7b",
    "tokenizer_id": "google/gemma-7b"
  },
  "config": {
    "model": "google/gemma-7b",
    "model_server": "jetstream",
    "start_time": "2024-09-25T16:34:14.365Z"
  },
  "summary_stats": {
    "stats": {
      "ttft": {
        "Mean": "0.13120518984972915",
        "Median": "0.16634615384615387",
        "Min": "0.01",
        "Max": "0.25",
        "P90": "0.23326923076923076",
        "P99": "0.24832692307692308"
      },
      "tpot": {
        "Mean": "0",
        "Median": "0.028433476394849787",
        "Min": "0.01",
        "Max": "2.5",
        "P90": "0.06770270270270269",
        "P99": "2.5"
      },
      "request_latency": {
        "Mean": "2.4746437307432325",
        "Median": "2.1858638743455496",
        "Min": "1",
        "Max": "10",
        "P90": "5.3442622950819665",
        "P99": "9.5344262295082"
      }
    }
  }
}

[achandrasekar]

Change looks good overall. But why are ttft, tpot and request latency from model server separately captured in summary stats? For latency, tpot and ttft we need to rely on the benchmarking tool's stats as opposed to model server's since that is the one which provides client's observable stats.