Add files using upload-large-folder tool

experiments/kv-cache-fp8/capacity/hammer/N0008/aiperf_stdout.log

@@ -0,0 +1,225 @@
+m │ 229,946.70 │ 130,635.95 │  80,401.80 │     379.43 │ 35,740.41 │
+│      Effective Decode Throughput Per User │      58.47 │   0.00 │  32,813.06 │     229.08 │      83.89 │      68.83 │    146.22 │
+│                         (tokens/sec/user) │            │        │            │            │            │            │           │
+│     Effective Prefill Throughput Per User │  15,775.27 │   0.00 │ 132,048.85 │ 111,797.56 │  63,994.09 │       0.00 │ 29,424.47 │
+│                         (tokens/sec/user) │            │        │            │            │            │            │           │
+│                 Tokens In Flight (tokens) │ 166,795.79 │   0.00 │ 271,358.41 │ 263,693.79 │ 231,076.54 │ 170,662.29 │ 49,648.28 │
+│         Effective Latency (CO-aware) (ms) │   7,590.51 │ 744.71 │  30,524.44 │  27,608.61 │  20,500.16 │   3,519.81 │  7,683.13 │
+└───────────────────────────────────────────┴────────────┴────────┴────────────┴────────────┴────────────┴────────────┴───────────┘
+                                              NVIDIA AIPerf | LLM Metrics: Active                                               
+┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┓
+┃                                 Metric ┃       avg ┃      min ┃        max ┃        p99 ┃        p90 ┃       p50 ┃       std ┃
+┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━┩
+│  Active Decode Throughput (tokens/sec) │    311.63 │     0.00 │ 131,252.23 │   1,211.39 │     503.14 │    280.05 │    717.59 │
+│ Active Prefill Throughput (tokens/sec) │ 69,134.63 │ 3,161.03 │ 229,946.70 │ 161,713.59 │ 113,039.85 │ 66,238.68 │ 33,182.33 │
+│      Active Decode Throughput Per User │     58.69 │     0.00 │  32,813.06 │     229.63 │      83.92 │     68.96 │    146.46 │
+│                      (tokens/sec/user) │           │          │            │            │            │           │           │
+│     Active Prefill Throughput Per User │ 55,860.35 │ 3,006.51 │ 132,048.85 │ 124,351.10 │  96,782.23 │ 52,626.90 │ 28,751.29 │
+│                      (tokens/sec/user) │           │          │            │            │            │           │           │
+│   Active Total Throughput (tokens/sec) │ 19,836.75 │    62.64 │ 229,946.70 │ 130,635.95 │  80,401.80 │    379.43 │ 35,741.86 │
+└────────────────────────────────────────┴───────────┴──────────┴────────────┴────────────┴────────────┴───────────┴───────────┘
+                                                NVIDIA AIPerf | LLM Metrics: Usage                                                
+┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┓
+┃                                  Metric ┃           avg ┃      min ┃       max ┃       p99 ┃       p90 ┃       p50 ┃       std ┃
+┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━┩
+│            Usage Prompt Tokens (tokens) │     26,460.34 │ 8,103.00 │ 59,589.00 │ 55,921.25 │ 44,064.50 │ 24,594.50 │ 12,309.78 │
+│ Usage Prompt Cache Read Tokens (tokens) │     25,854.85 │ 7,744.00 │ 59,456.00 │ 55,658.24 │ 43,430.40 │ 24,160.00 │ 12,344.32 │
+│        Usage Completion Tokens (tokens) │        421.71 │    28.00 │  1,990.00 │  1,651.00 │  1,138.00 │    164.50 │    460.71 │
+│             Usage Total Tokens (tokens) │     26,882.05 │ 8,188.00 │ 60,066.00 │ 56,244.50 │ 44,627.00 │ 24,941.00 │ 12,438.90 │
+│      Total Usage Prompt Tokens (tokens) │ 17,887,191.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│    Total Usage Prompt Cache Read Tokens │ 17,271,040.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│                                (tokens) │               │          │           │           │           │           │           │
+│       Overall Usage Prompt Cache Read % │         96.56 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│                                     (%) │               │          │           │           │           │           │           │
+│  Total Usage Completion Tokens (tokens) │    285,078.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│       Total Usage Total Tokens (tokens) │ 18,172,269.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+└─────────────────────────────────────────┴───────────────┴──────────┴───────────┴───────────┴───────────┴───────────┴───────────┘
+                                                    NVIDIA AIPerf | LLM Metrics                                                     
+┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┓
+┃                                        Metric ┃       avg ┃      min ┃       max ┃       p99 ┃       p90 ┃       p50 ┃       std ┃
+┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━┩
+│                      Time to First Token (ms) │    551.55 │   304.69 │  3,189.82 │  2,702.84 │    793.01 │    429.83 │    345.27 │
+│                     Time to Second Token (ms) │     39.74 │     0.00 │    612.17 │    416.83 │     49.18 │     23.69 │     77.15 │
+│               Time to First Output Token (ms) │  3,095.22 │   351.50 │ 25,785.74 │ 21,981.24 │ 10,905.10 │    765.60 │  5,076.13 │
+│                          Request Latency (ms) │  7,573.56 │   743.63 │ 30,522.45 │ 27,607.30 │ 20,473.26 │  3,517.85 │  7,686.34 │
+│                      Inter Token Latency (ms) │     16.58 │     5.89 │     40.62 │     32.43 │     19.92 │     16.28 │      4.00 │
+│              Output Token Throughput Per User │     63.15 │    24.62 │    169.66 │     87.25 │     78.87 │     61.44 │     12.97 │
+│                             (tokens/sec/user) │           │          │           │           │           │           │           │
+│ E2E Output Token Throughput (tokens/sec/user) │     51.36 │    17.98 │     80.45 │     71.73 │     62.72 │     53.04 │     10.51 │
+│               Output Sequence Length (tokens) │    421.71 │    28.00 │  1,990.00 │  1,651.00 │  1,138.00 │    164.50 │    460.71 │
+│                Input Sequence Length (tokens) │ 26,460.34 │ 8,103.00 │ 59,589.00 │ 55,921.25 │ 44,064.50 │ 24,594.50 │ 12,309.78 │
+│          Output Token Throughput (tokens/sec) │    311.16 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│           Input Token Throughput (tokens/sec) │ 19,524.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│             Request Throughput (requests/sec) │      0.74 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│                      Request Count (requests) │    676.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+└───────────────────────────────────────────────┴───────────┴──────────┴───────────┴───────────┴───────────┴───────────┴───────────┘
+
+
+No GPU telemetry data collected during the benchmarking run.
+
+╭─────────────────────────────── Output Sequence Length Mismatch Warning ───────────────────────────────╮
+│  567 of 676 requests (83.9%) have output length differing from requested by more than the threshold.  │
+│  Threshold (tokens): min(requested x 5%, 50)                                                          │
+│  Average mismatch: -66.1%                                                                             │
+│                                                                                                       │
+│  Why: Server hit EOS token before reaching requested output length.                                   │
+│                                                                                                       │
+│  Fix Options:                                                                                         │
+│    - --extra-inputs ignore_eos:true - Generate until max_tokens (vLLM, TensorRT-LLM)                  │
+│    - --extra-inputs min_tokens:<N> - Set minimum output length (vLLM, TensorRT-LLM, SGLang)           │
+│    - --use-server-token-count - Use server-reported token counts if tokenizer mismatch suspected      │
+│                                                                                                       │
+│  Diagnostics:                                                                                         │
+│    - Review profile_export.jsonl -> osl_mismatch_diff_pct for per-request values                      │
+│    - Adjust: AIPERF_METRICS_OSL_MISMATCH_PCT_THRESHOLD=5                                              │
+│    - Adjust: AIPERF_METRICS_OSL_MISMATCH_MAX_TOKEN_THRESHOLD=50                                       │
+╰───────────────────────────────────────────────────────────────────────────────────────────────────────╯
+
+CLI Command: aiperf profile --model 'glm-5.2' --url 'http://127.0.0.1:8391' --api-key '<redacted>' --endpoint-type 'chat' --streaming --tokenizer 'builtin' 
+--custom-dataset-type 'dag_jsonl' --input-file 'data/capacity/corpus.jsonl' --extra-inputs '{"cache_salt": "1f22c4b2d0a78d727b84f211cf20ef5d"}' --concurrency 8 
+--warmup-duration 120 --benchmark-duration 900 --benchmark-grace-period 300 --request-count 1000000 --random-seed 0 --use-server-token-count --artifact-dir 
+'results/capacity/sweep-hammer-jul20/N0008' --export-level 'raw' --export-http-trace --slice-duration 60 --server-metrics-formats 'json' 'csv' 'jsonl' --ui 'simple'
+Benchmark Duration: 916.14 sec
+CSV Export: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0008/profile_export_aiperf.csv
+JSON Export: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0008/profile_export_aiperf.json
+Raw Records: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0008/profile_export_raw.jsonl
+Server Metrics CSV Export: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0008/server_metrics_export.csv
+Server Metrics JSON Export: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0008/server_metrics_export.json
+Timeslice CSV Export: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0008/profile_export_aiperf_timeslices.csv
+Timeslice JSON Export: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0008/profile_export_aiperf_timeslices.json
+Log File: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0008/logs/aiperf.log
+
+
+Warmup:   0% |          | [00:00<?]
+Warmup:   0% |          | [00:00<?]
+                                   
+
+Warmup:   2% |▏         | [00:01<01:57]
+Warmup:   2% |▏         | [00:01<01:57]
+Warmup:   3% |▎         | [00:03<01:55]
+Warmup:   3% |▎         | [00:03<01:55]
+Warmup:   5% |▌         | [00:05<01:53]
+Warmup:   5% |▌         | [00:05<01:53]
+Warmup:   7% |▋         | [00:07<01:51]
+Warmup:   7% |▋         | [00:07<01:51]
+Warmup:   8% |▊         | [00:09<01:49]
+Warmup:   8% |▊         | [00:09<01:49]
+Warmup:  10% |█         | [00:11<01:47]
+Warmup:  10% |█         | [00:11<01:47]
+Warmup:  12% |█▏        | [00:13<01:45]
+Warmup:  12% |█▏        | [00:13<01:45]
+Warmup:  13% |█▎        | [00:15<01:43]
+Warmup:  13% |█▎        | [00:15<01:43]
+Warmup:  15% |█▌        | [00:17<01:41]
+Warmup:  15% |█▌        | [00:17<01:41]
+Warmup:  17% |█▋        | [00:19<01:39]
+Warmup:  17% |█▋        | [00:19<01:39]
+Warmup:  18% |█▊        | [00:21<01:37]
+Warmup:  18% |█▊        | [00:21<01:37]
+Warmup:  20% |██        | [00:23<01:35]
+Warmup:  20% |██        | [00:23<01:35]
+Warmup:  22% |██▏       | [00:25<01:33]
+Warmup:  22% |██▏       | [00:25<01:33]
+Warmup:  23% |██▎       | [00:27<01:31]
+Warmup:  23% |██▎       | [00:27<01:31]
+Warmup:  25% |██▌       | [00:29<01:29]
+Warmup:  25% |██▌       | [00:29<01:29]
+Warmup:  27% |██▋       | [00:31<01:27]
+Warmup:  27% |██▋       | [00:31<01:27]
+Warmup:  28% |██▊       | [00:33<01:25]
+Warmup:  28% |██▊       | [00:33<01:25]
+Warmup:  30% |███       | [00:35<01:23]
+Warmup:  30% |███       | [00:35<01:23]
+Warmup:  32% |███▏      | [00:37<01:21]
+Warmup:  32% |███▏      | [00:37<01:21]
+Warmup:  33% |███▎      | [00:39<01:19]
+Warmup:  33% |███▎      | [00:39<01:19]
+Warmup:  35% |███▌      | [00:41<01:17]
+Warmup:  35% |███▌      | [00:41<01:17]
+Warmup:  37% |███▋      | [00:43<01:15]
+Warmup:  37% |███▋      | [00:43<01:15]
+Warmup:  38% |███▊      | [00:45<01:13]
+Warmup:  38% |███▊      | [00:45<01:13]
+Warmup:  40% |████      | [00:47<01:11]
+Warmup:  40% |████      | [00:47<01:11]
+Warmup:  42% |████▏     | [00:49<01:09]
+Warmup:  42% |████▏     | [00:49<01:09]
+Warmup:  43% |████▎     | [00:51<01:07]
+Warmup:  43% |████▎     | [00:51<01:07]
+Warmup:  45% |████▌     | [00:53<01:05]
+Warmup:  45% |████▌     | [00:53<01:05]
+Warmup:  47% |████▋     | [00:55<01:03]
+Warmup:  47% |████▋     | [00:55<01:03]
+Warmup:  48% |████▊     | [00:57<01:01]
+Warmup:  48% |████▊     | [00:57<01:01]
+Warmup:  50% |█████     | [00:59<00:59]
+Warmup:  50% |█████     | [00:59<00:59]
+Warmup:  52% |█████▏    | [01:01<00:57]
+Warmup:  52% |█████▏    | [01:01<00:57]
+Warmup:  53% |█████▎    | [01:03<00:55]
+Warmup:  53% |█████▎    | [01:03<00:55]
+Warmup:  55% |█████▌    | [01:05<00:53]
+Warmup:  55% |█████▌    | [01:05<00:53]
+Warmup:  57% |█████▋    | [01:07<00:51]
+Warmup:  57% |█████▋    | [01:07<00:51]
+Warmup:  58% |█████▊    | [01:09<00:49]
+Warmup:  58% |█████▊    | [01:09<00:49]
+Warmup:  60% |██████    | [01:11<00:47]
+Warmup:  60% |██████    | [01:11<00:47]
+Warmup:  62% |██████▏   | [01:13<00:45]
+Warmup:  62% |██████▏   | [01:13<00:45]
+Warmup:  63% |██████▎   | [01:15<00:43]
+Warmup:  63% |██████▎   | [01:15<00:43]
+Warmup:  65% |██████▌   | [01:17<00:41]
+Warmup:  65% |██████▌   | [01:17<00:41]
+Warmup:  67% |██████▋   | [01:19<00:39]
+Warmup:  67% |██████▋   | [01:19<00:39]
+Warmup:  68% |██████▊   | [01:21<00:37]
+Warmup:  68% |██████▊   | [01:21<00:37]
+Warmup:  70% |███████   | [01:23<00:35]
+Warmup:  70% |███████   | [01:23<00:35]
+Warmup:  72% |███████▏  | [01:25<00:33]
+Warmup:  72% |███████▏  | [01:25<00:33]
+Warmup:  73% |███████▎  | [01:27<00:31]
+Warmup:  73% |███████▎  | [01:27<00:31]
+Warmup:  75% |███████▌  | [01:29<00:29]
+Warmup:  75% |███████▌  | [01:29<00:29]
+Warmup:  77% |███████▋  | [01:31<00:27]
+Warmup:  77% |███████▋  | [01:31<00:27]
+Warmup:  78% |███████▊  | [01:33<00:25]
+Warmup:  78% |███████▊  | [01:33<00:25]
+Warmup:  80% |████████  | [01:35<00:23]
+Warmup:  80% |████████  | [01:35<00:23]
+Warmup:  82% |████████▏ | [01:37<00:21]
+Warmup:  82% |████████▏ | [01:37<00:21]
+Warmup:  83% |████████▎ | [01:39<00:19]
+Warmup:  83% |████████▎ | [01:39<00:19]
+Warmup:  85% |████████▌ | [01:41<00:17]
+Warmup:  85% |████████▌ | [01:41<00:17]
+Warmup:  87% |████████▋ | [01:43<00:15]
+Warmup:  87% |████████▋ | [01:43<00:15]
+Warmup:  88% |████████▊ | [01:45<00:13]
+Warmup:  88% |████████▊ | [01:45<00:13]
+Warmup:  90% |█████████ | [01:47<00:11]
+Warmup:  90% |█████████ | [01:47<00:11]
+Warmup:  92% |█████████▏| [01:49<00:09]
+Warmup:  92% |█████████▏| [01:49<00:09]
+Warmup:  93% |█████████▎| [01:51<00:07]
+Warmup:  93% |█████████▎| [01:51<00:07]
+Warmup:  95% |█████████▌| [01:53<00:05]
+Warmup:  95% |█████████▌| [01:53<00:05]
+Warmup:  97% |█████████▋| [01:55<00:03]
+Warmup:  97% |█████████▋| [01:55<00:03]
+Warmup:  98% |█████████▊| [01:57<00:01]
+Warmup:  98% |█████████▊| [01:57<00:01]
+Warmup Grace Period: 0/136 |          |   0% [00:00<?]
+Warmup Grace Period: 132/136 |█████████▋|  97% [00:00<00:00]
+Warmup Grace Period: 135/136 |█████████▉|  99% [00:06<00:00]
+Warmup Grace Period: 135/136 |█████████▉|  99% [00:06<00:00]
+Processing Records: 0/136 |          |   0% [00:00<?]
+Profiling: 0/1,000,000 |          |   0% [00:00<?]
+Profiling Grace Period: 0/676 |          |   0% [00:00<?]
+Profiling Grace Period: 670/676 |█████████▉|  99% [00:00<00:00]
+Processing Records: 0/676 |          |   0% [00:00<?]
+Processing Records: 676/676 |██████████| 100% [00:01<00:00]
+Processing Records: 676/676 |██████████| 100% [00:01<00:00]

experiments/kv-cache-fp8/capacity/hammer/N0008/manifest.json

@@ -0,0 +1,111 @@
+{
+  "concurrency": 8,
+  "mode": "hammer",
+  "cache_salt": "1f22c4b2d0a78d727b84f211cf20ef5d",
+  "command": [
+    "aiperf",
+    "profile",
+    "--model",
+    "glm-5.2",
+    "--url",
+    "http://127.0.0.1:8391",
+    "--api-key",
+    "<key>",
+    "--endpoint-type",
+    "chat",
+    "--streaming",
+    "--tokenizer",
+    "builtin",
+    "--custom-dataset-type",
+    "dag_jsonl",
+    "--input-file",
+    "data/capacity/corpus.jsonl",
+    "--extra-inputs",
+    "{\"cache_salt\": \"1f22c4b2d0a78d727b84f211cf20ef5d\"}",
+    "--concurrency",
+    "8",
+    "--warmup-duration",
+    "120",
+    "--benchmark-duration",
+    "900",
+    "--benchmark-grace-period",
+    "300",
+    "--request-count",
+    "1000000",
+    "--random-seed",
+    "0",
+    "--use-server-token-count",
+    "--artifact-dir",
+    "results/capacity/sweep-hammer-jul20/N0008",
+    "--export-level",
+    "raw",
+    "--export-http-trace",
+    "--slice-duration",
+    "60",
+    "--server-metrics-formats",
+    "json",
+    "csv",
+    "jsonl",
+    "--ui",
+    "simple"
+  ],
+  "dataset_path": "data/capacity/corpus.jsonl",
+  "signature": {
+    "dataset_sha256": "0327d658a267715aebdab45b67b6e0d324616a0faeec8a75d09b8defcd0b2ac2",
+    "url": "http://127.0.0.1:8391",
+    "model": "glm-5.2",
+    "duration": 900,
+    "warmup": 120,
+    "grace": 300,
+    "seed": 0,
+    "request_ceiling": 1000000,
+    "python": "3.12.3",
+    "aiperf": {
+      "version": "0.11.0",
+      "rev": "73d0524f98f30199ca659a8ab586454a823c887d"
+    },
+    "server": {
+      "version": "0.5.15.post1",
+      "model_path": "/home/jl_fs/models/GLM-5.2-FP8",
+      "served_model_name": "glm-5.2",
+      "tp_size": 8,
+      "dp_size": 8,
+      "enable_cache_report": true,
+      "speculative_algorithm": "EAGLE",
+      "max_running_requests": 48,
+      "context_length": 262144,
+      "kv_cache_dtype": "fp8_e4m3",
+      "mem_fraction_static": 0.872,
+      "scheduler_pids": [
+        82149,
+        82150,
+        82151,
+        82152,
+        82153,
+        82154,
+        82155,
+        82156
+      ],
+      "max_total_num_tokens": 234176,
+      "max_model_len": 262144,
+      "model_id": "glm-5.2"
+    },
+    "evaluator_sha256": "3e48dfede43cbf47244f560547eda95d29c671d4fa23b203a5d220a14edf1c80",
+    "verifier_sha256": "6f84bee84368a72ea1445d0961185a0d5d2fd6ba8f8ec4c3fefac619b46a32aa",
+    "runner_sha256": "aa3e1a0f66bcfac6663b5d2e59a37cae9d7264dda6b4c13130a3edf403e77104"
+  },
+  "started_utc": "2026-07-20T14:14:35Z",
+  "loadavg_start": [
+    0.1611328125,
+    0.6181640625,
+    0.68798828125
+  ],
+  "status": "ok",
+  "elapsed_s": 1065.7,
+  "loadavg_end": [
+    2.091796875,
+    1.7939453125,
+    1.40966796875
+  ],
+  "returncode": 0
+}

experiments/kv-cache-fp8/capacity/hammer/N0008/profile_export_aiperf.csv

@@ -0,0 +1,68 @@
+Metric,avg,min,max,sum,p1,p5,p10,p25,p50,p75,p90,p95,p99,std
+Active Decode Throughput (tokens/sec),311.63,0.00,131252.23,,,,,,280.05,,503.14,545.08,1211.39,717.59
+Active Decode Throughput Per User (tokens/sec/user),58.69,0.00,32813.06,,,,,,68.96,,83.92,90.31,229.63,146.46
+Active Prefill Throughput (tokens/sec),69134.63,3161.03,229946.70,,,,,,66238.68,,113039.85,124902.65,161713.59,33182.33
+Active Prefill Throughput Per User (tokens/sec/user),55860.35,3006.51,132048.85,,,,,,52626.90,,96782.23,107059.88,124351.10,28751.29
+Active Total Throughput (tokens/sec),19836.75,62.64,229946.70,,,,,,379.43,,80401.80,98288.82,130635.95,35741.86
+Credit-to-Start Latency (ms),1.58,0.83,7.56,1069.26,0.94,1.05,1.10,1.20,1.35,1.56,2.03,3.33,5.59,0.82
+E2E Output Token Throughput (tokens/sec/user),51.36,17.98,80.45,34716.20,21.18,29.92,37.34,46.29,53.04,57.85,62.72,66.38,71.73,10.51
+Effective Concurrency,5.60,0.00,8.00,,,,,,6.00,,7.00,8.00,8.00,1.49
+Effective Decode Concurrency,5.19,0.00,8.00,,,,,,5.00,,7.00,7.00,8.00,1.55
+Effective Decode Throughput (tokens/sec),310.43,0.00,131252.23,,,,,,277.02,,503.06,544.81,1210.22,716.46
+Effective Decode Throughput Per User (tokens/sec/user),58.47,0.00,32813.06,,,,,,68.83,,83.89,90.26,229.08,146.22
+Effective Latency (CO-aware) (ms),7590.51,744.71,30524.44,5131186.80,934.68,1195.10,1369.73,1868.88,3519.81,12332.03,20500.16,22650.66,27608.61,7683.13
+Effective Prefill Concurrency,0.41,0.00,8.00,,,,,,0.00,,1.00,2.00,3.00,0.82
+Effective Prefill Throughput (tokens/sec),19524.00,0.00,229946.70,,,,,,0.00,,80157.46,98007.84,130393.58,35770.73
+Effective Prefill Throughput Per User (tokens/sec/user),15775.27,0.00,132048.85,,,,,,0.00,,63994.09,85262.32,111797.56,29424.47
+Effective Total Throughput (tokens/sec),19834.42,0.00,229946.70,,,,,,379.43,,80401.80,98288.82,130635.95,35740.41
+HTTP Blocked (ms),0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00
+HTTP Chunks Received,222.93,15.00,1032.00,150698.00,21.00,27.00,32.00,44.00,88.50,367.50,607.50,704.25,837.50,242.07
+HTTP Chunks Sent,1.00,1.00,1.00,676.00,1.00,1.00,1.00,1.00,1.00,1.00,1.00,1.00,1.00,0.00
+HTTP Connecting (ms),0.05,0.00,1.81,31.18,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,1.43,0.24
+HTTP Connection Overhead (ms),0.05,0.00,1.81,31.18,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,1.43,0.24
+HTTP Connection Reused (ratio),0.96,0.00,1.00,652.00,0.00,1.00,1.00,1.00,1.00,1.00,1.00,1.00,1.00,0.19
+HTTP Data Received (KB),55.67,4.31,255.56,37634.77,5.57,7.33,8.59,11.67,22.83,94.88,149.36,172.29,209.47,59.40
+HTTP Data Sent (KB),106.25,36.83,232.43,71825.63,37.81,42.02,46.63,67.58,98.57,141.34,172.81,193.03,223.18,47.65
+HTTP DNS Lookup (ms),0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00
+HTTP Duration (excl. conn) (ms),7588.79,743.51,30522.14,5130018.73,932.15,1193.46,1368.22,1867.51,3518.55,12330.21,20498.63,22648.86,27607.18,7683.12
+HTTP Receiving (ms),7050.28,218.36,29753.20,4765991.84,509.82,723.00,887.34,1360.90,2766.10,11843.13,19889.52,22169.86,27179.05,7681.21
+HTTP Sending (ms),0.32,0.18,2.83,219.06,0.19,0.20,0.20,0.22,0.25,0.30,0.39,0.50,2.17,0.32
+HTTP Total Time (ms),7588.83,744.28,30522.14,5130049.91,932.60,1193.46,1368.22,1867.51,3518.55,12330.53,20498.63,22648.86,27607.18,7683.10
+HTTP Waiting (TTFB) (ms),538.18,304.29,3187.36,363807.83,328.45,342.40,354.07,375.99,419.59,625.62,764.17,907.32,2700.46,340.92
+Input Sequence Length (tokens),26460.34,8103.00,59589.00,17887191.00,8347.00,9649.50,10918.00,16504.50,24594.50,35760.50,44064.50,47562.50,55921.25,12309.78
+Inter Chunk Latency (ms),31.09,0.00,1177.04,4746882.33,0.00,0.59,1.09,23.40,25.39,26.60,46.94,62.12,301.77,47.41
+Inter Token Latency (ms),16.58,5.89,40.62,11206.65,11.46,12.07,12.68,13.83,16.28,17.90,19.92,23.95,32.43,4.00
+OSL Mismatch Diff % (%),-66.06,-97.57,0.00,-44654.85,-97.23,-95.38,-94.63,-92.81,-85.75,-42.08,0.00,0.00,0.00,35.52
+Output Sequence Length (tokens),421.71,28.00,1990.00,285078.00,38.00,52.00,61.00,83.00,164.50,699.75,1138.00,1345.75,1651.00,460.71
+Output Token Count (tokens),421.71,28.00,1990.00,285078.00,38.00,52.00,61.00,83.00,164.50,699.75,1138.00,1345.75,1651.00,460.71
+Output Token Throughput Per User (tokens/sec/user),63.15,24.62,169.66,42688.72,30.84,41.75,50.20,55.87,61.44,72.30,78.87,82.83,87.25,12.97
+Prefill Throughput Per User (tokens/sec/user),55629.38,2852.00,132048.85,37605459.06,3661.50,14786.76,21339.38,31620.28,52781.87,76390.86,96799.70,107052.74,124219.16,28822.23
+Request Latency (ms),7573.56,743.63,30522.45,5119729.37,932.27,1180.88,1353.27,1861.63,3517.85,12330.42,20473.26,22649.07,27607.30,7686.34
+Time to First Output Token (ms),3095.22,351.50,25785.74,1847847.50,368.70,384.07,396.18,437.88,765.60,2675.07,10905.10,15879.63,21981.24,5076.13
+Time to First Token (ms),551.55,304.69,3189.82,372847.04,329.25,345.82,360.62,385.07,429.83,638.36,793.01,942.64,2702.84,345.27
+Time to Second Token (ms),39.74,0.00,612.17,26864.61,0.00,0.13,1.04,20.56,23.69,26.56,49.18,93.84,416.83,77.15
+Tokens In Flight (tokens),166795.79,0.00,271358.41,,,,,,170662.29,,231076.54,245748.87,263693.79,49648.28
+Usage Completion Tokens (tokens),421.71,28.00,1990.00,285078.00,38.00,52.00,61.00,83.00,164.50,699.75,1138.00,1345.75,1651.00,460.71
+Usage Prompt Cache Read Tokens (tokens),25854.85,7744.00,59456.00,17271040.00,7744.00,8790.40,10412.80,15872.00,24160.00,35136.00,43430.40,47168.00,55658.24,12344.32
+Usage Prompt Tokens (tokens),26460.34,8103.00,59589.00,17887191.00,8347.00,9649.50,10918.00,16504.50,24594.50,35760.50,44064.50,47562.50,55921.25,12309.78
+Usage Total Tokens (tokens),26882.05,8188.00,60066.00,18172269.00,8420.25,9724.50,11011.50,16748.75,24941.00,36258.50,44627.00,48540.75,56244.50,12438.90
+
+Metric,Value
+Benchmark Duration (sec),916.14
+Input Token Throughput (tokens/sec),19524.00
+OSL Mismatch Count,567.00
+Output Token Throughput (tokens/sec),311.16
+Overall Usage Prompt Cache Read % (%),96.56
+Request Count,676.00
+Request Throughput (requests/sec),0.74
+Total Input Sequence Length (tokens),17887191.00
+Total Output Sequence Length (tokens),285078.00
+Total Output Tokens (tokens),285078.00
+Total Token Throughput (tokens/sec),19835.16
+Total Usage Completion Tokens (tokens),285078.00
+Total Usage Prompt Cache Read Tokens (tokens),17271040.00
+Total Usage Prompt Tokens (tokens),17887191.00
+Total Usage Total Tokens (tokens),18172269.00
+
+
+Endpoint,GPU_Index,GPU_Name,GPU_UUID,Metric,avg,min,max,sum,p1,p5,p10,p25,p50,p75,p90,p95,p99,std

experiments/kv-cache-fp8/capacity/hammer/N0008/profile_export_aiperf.json

@@ -0,0 +1,1726 @@
+{
+  "schema_version": "1.4",
+  "aiperf_version": "0.11.0",
+  "benchmark_id": "10070654e5b2",
+  "request_throughput": {
+    "unit": "requests/sec",
+    "avg": 0.737858802511654
+  },
+  "request_latency": {
+    "unit": "ms",
+    "avg": 7573.564152266272,
+    "p1": 932.2663037499999,
+    "p5": 1180.880011,
+    "p10": 1353.2668844999998,
+    "p25": 1861.63210575,
+    "p50": 3517.8491725,
+    "p75": 12330.418869,
+    "p90": 20473.2568205,
+    "p95": 22649.06756625,
+    "p99": 27607.2966035,
+    "min": 743.628568,
+    "max": 30522.449643,
+    "std": 7686.337522041795,
+    "count": 676,
+    "sum": 5119729.366932
+  },
+  "request_count": {
+    "unit": "requests",
+    "avg": 676.0
+  },
+  "time_to_first_token": {
+    "unit": "ms",
+    "avg": 551.5488734763313,
+    "p1": 329.24822975,
+    "p5": 345.8161925,
+    "p10": 360.616137,
+    "p25": 385.0733515,
+    "p50": 429.833394,
+    "p75": 638.35825925,
+    "p90": 793.0084734999999,
+    "p95": 942.6356475,
+    "p99": 2702.8425939999997,
+    "min": 304.685214,
+    "max": 3189.817549,
+    "std": 345.265270971378,
+    "count": 676,
+    "sum": 372847.03846999997
+  },
+  "time_to_second_token": {
+    "unit": "ms",
+    "avg": 39.74054270857988,
+    "p1": 0.0,
+    "p5": 0.13244775,
+    "p10": 1.040598,
+    "p25": 20.5568375,
+    "p50": 23.6939605,
+    "p75": 26.5636285,
+    "p90": 49.181386499999995,
+    "p95": 93.84136699999999,
+    "p99": 416.82876274999995,
+    "min": 0.0,
+    "max": 612.1721759999999,
+    "std": 77.14620759314485,
+    "count": 676,
+    "sum": 26864.606871
+  },
+  "inter_token_latency": {
+    "unit": "ms",
+    "avg": 16.5778884469506,
+    "p1": 11.461740016971829,
+    "p5": 12.072380020454544,
+    "p10": 12.679817842941176,
+    "p25": 13.831408074456519,
+    "p50": 16.2766705818603,
+    "p75": 17.89734092739617,
+    "p90": 19.92128951846785,
+    "p95": 23.9527383575659,
+    "p99": 32.430385601026735,
+    "min": 5.894173297297296,
+    "max": 40.623499458333335,
+    "std": 3.9984107791048533,
+    "count": 676,
+    "sum": 11206.652590138605
+  },
+  "output_token_throughput": {
+    "unit": "tokens/sec",
+    "avg": 311.1646622816824
+  },
+  "output_token_throughput_per_user": {
+    "unit": "tokens/sec/user",
+    "avg": 63.14899650189225,
+    "p1": 30.83546029482411,
+    "p5": 41.749042128479715,
+    "p10": 50.19760903448288,
+    "p25": 55.874222042158145,
+    "p50": 61.43766800864626,
+    "p75": 72.2992288593425,
+    "p90": 78.86558140810732,
+    "p95": 82.833712215892,
+    "p99": 87.24711295327137,
+    "min": 24.61629385291336,
+    "max": 169.6590767798663,
+    "std": 12.965071154779212,
+    "count": 676,
+    "sum": 42688.72163527916
+  },
+  "output_sequence_length": {
+    "unit": "tokens",
+    "avg": 421.7130177514793,
+    "p1": 38.0,
+    "p5": 52.0,
+    "p10": 61.0,
+    "p25": 83.0,
+    "p50": 164.5,
+    "p75": 699.75,
+    "p90": 1138.0,
+    "p95": 1345.75,
+    "p99": 1651.0,
+    "min": 28.0,
+    "max": 1990.0,
+    "std": 460.7059002410634,
+    "count": 676,
+    "sum": 285078.0
+  },
+  "input_sequence_length": {
+    "unit": "tokens",
+    "avg": 26460.34171597633,
+    "p1": 8347.0,
+    "p5": 9649.5,
+    "p10": 10918.0,
+    "p25": 16504.5,
+    "p50": 24594.5,
+    "p75": 35760.5,
+    "p90": 44064.5,
+    "p95": 47562.5,
+    "p99": 55921.25,
+    "min": 8103.0,
+    "max": 59589.0,
+    "std": 12309.779652807998,
+    "count": 676,
+    "sum": 17887191.0
+  },
+  "output_token_count": {
+    "unit": "tokens",
+    "avg": 421.7130177514793,
+    "p1": 38.0,
+    "p5": 52.0,
+    "p10": 61.0,
+    "p25": 83.0,
+    "p50": 164.5,
+    "p75": 699.75,
+    "p90": 1138.0,
+    "p95": 1345.75,
+    "p99": 1651.0,
+    "min": 28.0,
+    "max": 1990.0,
+    "std": 460.7059002410634,
+    "count": 676,
+    "sum": 285078.0
+  },
+  "inter_chunk_latency": {
+    "unit": "ms",
+    "avg": 31.088771406148485,
+    "p1": 0.0,
+    "p5": 0.5891495000000001,
+    "p10": 1.0867227000000004,
+    "p25": 23.3965835,
+    "p50": 25.388783999999998,
+    "p75": 26.59858475,
+    "p90": 46.93948680000009,
+    "p95": 62.12123289999998,
+    "p99": 301.77113008000003,
+    "min": 0.0,
+    "max": 1177.040022,
+    "std": 47.408039918892676,
+    "count": 152688,
+    "sum": 4746882.328462
+  },
+  "total_output_tokens": {
+    "unit": "tokens",
+    "avg": 285078.0
+  },
+  "benchmark_duration": {
+    "unit": "sec",
+    "avg": 916.142313216
+  },
+  "total_isl": {
+    "unit": "tokens",
+    "avg": 17887191.0
+  },
+  "total_osl": {
+    "unit": "tokens",
+    "avg": 285078.0
+  },
+  "telemetry_data": {
+    "summary": {
+      "endpoints_configured": [],
+      "endpoints_successful": [],
+      "start_time": "2026-07-20T14:16:58.899671",
+      "end_time": "2026-07-20T14:32:17.612839"
+    },
+    "endpoints": {},
+    "error_summary": []
+  },
+  "input_config": {
+    "models": {
+      "items": [
+        {
+          "name": "glm-5.2"
+        }
+      ]
+    },
+    "endpoint": {
+      "urls": [
+        "http://127.0.0.1:8391"
+      ],
+      "type": "chat",
+      "api_key": "<redacted>",
+      "streaming": true,
+      "use_server_token_count": true,
+      "extra": {
+        "cache_salt": "1f22c4b2d0a78d727b84f211cf20ef5d"
+      }
+    },
+    "datasets": [
+      {
+        "name": "main",
+        "type": "file",
+        "path": "data/capacity/corpus.jsonl",
+        "format": "dag_jsonl",
+        "entries": 1000000,
+        "random_seed": 0
+      }
+    ],
+    "phases": [
+      {
+        "name": "warmup",
+        "type": "concurrency",
+        "exclude_from_results": true,
+        "duration": 120.0,
+        "concurrency": 8
+      },
+      {
+        "name": "profiling",
+        "type": "concurrency",
+        "requests": 1000000,
+        "duration": 900.0,
+        "concurrency": 8,
+        "grace_period": 300.0
+      }
+    ],
+    "artifacts": {
+      "dir": "/home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0008",
+      "records": [
+        "jsonl"
+      ],
+      "raw": true,
+      "trace": true,
+      "slice_duration": 60.0,
+      "auto_plot": false,
+      "plot_required": false
+    },
+    "tokenizer": {
+      "name": "builtin"
+    },
+    "gpu_telemetry": {
+      "enabled": true
+    },
+    "server_metrics": {
+      "enabled": true,
+      "urls": [],
+      "formats": [
+        "json",
+        "csv",
+        "jsonl"
+      ]
+    },
+    "network_latency": {
+      "enabled": false
+    },
+    "otel": {},
+    "mlflow": {},
+    "wandb": {},
+    "runtime": {
+      "ui": "simple"
+    }
+  },
+  "run_info": {
+    "benchmark_id": "10070654e5b2",
+    "random_seed": 0,
+    "trial": 0,
+    "cli_command": "aiperf profile --model 'glm-5.2' --url 'http://127.0.0.1:8391' --api-key '<redacted>' --endpoint-type 'chat' --streaming --tokenizer 'builtin' --custom-dataset-type 'dag_jsonl' --input-file 'data/capacity/corpus.jsonl' --extra-inputs '{\"cache_salt\": \"1f22c4b2d0a78d727b84f211cf20ef5d\"}' --concurrency 8 --warmup-duration 120 --benchmark-duration 900 --benchmark-grace-period 300 --request-count 1000000 --random-seed 0 --use-server-token-count --artifact-dir 'results/capacity/sweep-hammer-jul20/N0008' --export-level 'raw' --export-http-trace --slice-duration 60 --server-metrics-formats 'json' 'csv' 'jsonl' --ui 'simple'"
+  },
+  "was_cancelled": false,
+  "error_summary": [],
+  "start_time": "2026-07-20T14:16:58.899671",
+  "end_time": "2026-07-20T14:32:15.064116",
+  "warmup_metrics": {
+    "output_token_count": {
+      "unit": "tokens",
+      "avg": 239.77205882352942,
+      "p1": 44.35,
+      "p5": 51.0,
+      "p10": 56.0,
+      "p25": 67.75,
+      "p50": 92.5,
+      "p75": 194.25,
+      "p90": 633.5,
+      "p95": 1024.5,
+      "p99": 1699.8500000000004,
+      "min": 36.0,
+      "max": 1879.0,
+      "std": 358.27263211857667,
+      "count": 136,
+      "sum": 32609.0
+    },
+    "http_req_dns_lookup": {
+      "unit": "ms",
+      "avg": 0.0,
+      "p1": 0.0,
+      "p5": 0.0,
+      "p10": 0.0,
+      "p25": 0.0,
+      "p50": 0.0,
+      "p75": 0.0,
+      "p90": 0.0,
+      "p95": 0.0,
+      "p99": 0.0,
+      "min": 0.0,
+      "max": 0.0,
+      "std": 0.0,
+      "count": 136,
+      "sum": 0.0
+    },
+    "input_sequence_length": {
+      "unit": "tokens",
+      "avg": 16740.242647058825,
+      "p1": 8472.05,
+      "p5": 8759.0,
+      "p10": 10125.0,
+      "p25": 12083.5,
+      "p50": 16190.5,
+      "p75": 21177.5,
+      "p90": 23923.5,
+      "p95": 26179.75,
+      "p99": 28319.100000000002,
+      "min": 8381.0,
+      "max": 29262.0,
+      "std": 5472.739105431795,
+      "count": 136,
+      "sum": 2276673.0
+    },
+    "time_to_first_token": {
+      "unit": "ms",
+      "avg": 752.6342808897059,
+      "p1": 319.72456705,
+      "p5": 343.87514275,
+      "p10": 358.30414099999996,
+      "p25": 381.64800775,
+      "p50": 553.069273,
+      "p75": 692.92237425,
+      "p90": 973.2809785,
+      "p95": 2934.448953,
+      "p99": 4063.6535584499998,
+      "min": 313.628898,
+      "max": 4849.065368,
+      "std": 780.6537767534721,
+      "count": 136,
+      "sum": 102358.26220099999
+    },
+    "http_req_sending": {
+      "unit": "ms",
+      "avg": 0.4972723455882353,
+      "p1": 0.1858059,
+      "p5": 0.19756574999999998,
+      "p10": 0.20473149999999998,
+      "p25": 0.21462725,
+      "p50": 0.242648,
+      "p75": 0.297638,
+      "p90": 0.547976,
+      "p95": 2.97434225,
+      "p99": 3.9805568000000053,
+      "min": 0.176445,
+      "max": 4.5152529999999995,
+      "std": 0.8253129370215951,
+      "count": 136,
+      "sum": 67.62903899999999
+    },
+    "http_req_data_sent": {
+      "unit": "KB",
+      "avg": 69.41786822150735,
+      "p1": 38.358642578125,
+      "p5": 39.43798828125,
+      "p10": 43.75830078125,
+      "p25": 50.7314453125,
+      "p50": 66.3115234375,
+      "p75": 85.499267578125,
+      "p90": 99.0673828125,
+      "p95": 108.28955078125,
+      "p99": 117.22666015625,
+      "min": 37.912109375,
+      "max": 120.96484375,
+      "std": 21.426324198857593,
+      "count": 136,
+      "sum": 9440.830078125
+    },
+    "request_latency": {
+      "unit": "ms",
+      "avg": 5011.392510088234,
+      "p1": 1074.74691815,
+      "p5": 1245.1420544999999,
+      "p10": 1478.3020265,
+      "p25": 1877.3755585,
+      "p50": 2723.5574659999997,
+      "p75": 6077.088917749999,
+      "p90": 10539.693227,
+      "p95": 18781.298834999998,
+      "p99": 31928.14163205008,
+      "min": 1039.929525,
+      "max": 37020.719005,
+      "std": 5958.839062398037,
+      "count": 136,
+      "sum": 681549.381372
+    },
+    "http_req_chunks_sent": {
+      "unit": "count",
+      "avg": 1.0,
+      "p1": 1.0,
+      "p5": 1.0,
+      "p10": 1.0,
+      "p25": 1.0,
+      "p50": 1.0,
+      "p75": 1.0,
+      "p90": 1.0,
+      "p95": 1.0,
+      "p99": 1.0,
+      "min": 1.0,
+      "max": 1.0,
+      "std": 0.0,
+      "count": 136,
+      "sum": 136.0
+    },
+    "http_req_data_received": {
+      "unit": "KB",
+      "avg": 32.064015107996326,
+      "p1": 6.05966796875,
+      "p5": 7.015380859375,
+      "p10": 7.8310546875,
+      "p25": 9.517578125,
+      "p50": 12.89697265625,
+      "p75": 27.33251953125,
+      "p90": 84.33349609375,
+      "p95": 132.310546875,
+      "p99": 217.2578125,
+      "min": 5.3134765625,
+      "max": 236.75390625,
+      "std": 46.11630362320757,
+      "count": 136,
+      "sum": 4360.7060546875
+    },
+    "output_sequence_length": {
+      "unit": "tokens",
+      "avg": 239.77205882352942,
+      "p1": 44.35,
+      "p5": 51.0,
+      "p10": 56.0,
+      "p25": 67.75,
+      "p50": 92.5,
+      "p75": 194.25,
+      "p90": 633.5,
+      "p95": 1024.5,
+      "p99": 1699.8500000000004,
+      "min": 36.0,
+      "max": 1879.0,
+      "std": 358.27263211857667,
+      "count": 136,
+      "sum": 32609.0
+    },
+    "time_to_first_output_token": {
+      "unit": "ms",
+      "avg": 2851.7498229848484,
+      "p1": 367.76618673,
+      "p5": 379.90407495,
+      "p10": 385.2546019,
+      "p25": 644.27188975,
+      "p50": 1200.44009,
+      "p75": 2959.9133092499997,
+      "p90": 5946.4682372000025,
+      "p95": 10531.99668249999,
+      "p99": 21558.10134016,
+      "min": 357.76890299999997,
+      "max": 34672.82316,
+      "std": 4681.075323633009,
+      "count": 132,
+      "sum": 376430.97663399996
+    },
+    "http_req_duration": {
+      "unit": "ms",
+      "avg": 5029.641734941176,
+      "p1": 1079.04369315,
+      "p5": 1250.3358027499999,
+      "p10": 1481.6827644999998,
+      "p25": 1912.5112405,
+      "p50": 2727.4463615,
+      "p75": 6076.8993025,
+      "p90": 10550.435273,
+      "p95": 18794.07998575,
+      "p99": 31938.803667950076,
+      "min": 1039.69888,
+      "max": 37020.591443,
+      "std": 5954.622570724064,
+      "count": 136,
+      "sum": 684031.2759519999
+    },
+    "http_req_connection_reused": {
+      "unit": "ratio",
+      "avg": 0.9191176470588235,
+      "p1": 0.0,
+      "p5": 0.0,
+      "p10": 1.0,
+      "p25": 1.0,
+      "p50": 1.0,
+      "p75": 1.0,
+      "p90": 1.0,
+      "p95": 1.0,
+      "p99": 1.0,
+      "min": 0.0,
+      "max": 1.0,
+      "std": 0.2726543561432229,
+      "count": 136,
+      "sum": 125.0
+    },
+    "http_req_blocked": {
+      "unit": "ms",
+      "avg": 0.0,
+      "p1": 0.0,
+      "p5": 0.0,
+      "p10": 0.0,
+      "p25": 0.0,
+      "p50": 0.0,
+      "p75": 0.0,
+      "p90": 0.0,
+      "p95": 0.0,
+      "p99": 0.0,
+      "min": 0.0,
+      "max": 0.0,
+      "std": 0.0,
+      "count": 136,
+      "sum": 0.0
+    },
+    "usage_prompt_tokens": {
+      "unit": "tokens",
+      "avg": 16740.242647058825,
+      "p1": 8472.05,
+      "p5": 8759.0,
+      "p10": 10125.0,
+      "p25": 12083.5,
+      "p50": 16190.5,
+      "p75": 21177.5,
+      "p90": 23923.5,
+      "p95": 26179.75,
+      "p99": 28319.100000000002,
+      "min": 8381.0,
+      "max": 29262.0,
+      "std": 5472.739105431795,
+      "count": 136,
+      "sum": 2276673.0
+    },
+    "http_req_receiving": {
+      "unit": "ms",
+      "avg": 4285.294017419117,
+      "p1": 686.93976445,
+      "p5": 750.3942925,
+      "p10": 876.3018575,
+      "p25": 1347.7771695,
+      "p50": 2111.8468395,
+      "p75": 3711.5455829999996,
+      "p90": 10188.067774,
+      "p95": 18380.4553685,
+      "p99": 30843.07874610007,
+      "min": 592.366901,
+      "max": 36390.135363,
+      "std": 5947.4641747963,
+      "count": 136,
+      "sum": 582799.9863689999
+    },
+    "usage_completion_tokens": {
+      "unit": "tokens",
+      "avg": 239.77205882352942,
+      "p1": 44.35,
+      "p5": 51.0,
+      "p10": 56.0,
+      "p25": 67.75,
+      "p50": 92.5,
+      "p75": 194.25,
+      "p90": 633.5,
+      "p95": 1024.5,
+      "p99": 1699.8500000000004,
+      "min": 36.0,
+      "max": 1879.0,
+      "std": 358.27263211857667,
+      "count": 136,
+      "sum": 32609.0
+    },
+    "http_req_connecting": {
+      "unit": "ms",
+      "avg": 0.18173928676470585,
+      "p1": 0.0,
+      "p5": 0.0,
+      "p10": 0.0,
+      "p25": 0.0,
+      "p50": 0.0,
+      "p75": 0.0,
+      "p90": 0.0,
+      "p95": 2.21050025,
+      "p99": 2.6565537500000005,
+      "min": 0.0,
+      "max": 3.776115,
+      "std": 0.6473690723335254,
+      "count": 136,
+      "sum": 24.716542999999998
+    },
+    "request_count": {
+      "unit": "requests",
+      "avg": 136.0
+    },
+    "http_req_waiting": {
+      "unit": "ms",
+      "avg": 743.8504451764707,
+      "p1": 319.33812185,
+      "p5": 335.6974605,
+      "p10": 348.31226,
+      "p25": 377.33595825,
+      "p50": 541.9367075,
+      "p75": 682.8577329999999,
+      "p90": 958.8941924999999,
+      "p95": 2929.9344014999997,
+      "p99": 4060.13787,
+      "min": 313.294267,
+      "max": 4845.354590999999,
+      "std": 781.3684840311562,
+      "count": 136,
+      "sum": 101163.660544
+    },
+    "usage_total_tokens": {
+      "unit": "tokens",
+      "avg": 16980.014705882353,
+      "p1": 8534.8,
+      "p5": 8818.0,
+      "p10": 10211.0,
+      "p25": 12157.25,
+      "p50": 16318.5,
+      "p75": 21541.25,
+      "p90": 24804.5,
+      "p95": 26651.5,
+      "p99": 28444.350000000002,
+      "min": 8439.0,
+      "max": 30930.0,
+      "std": 5616.246486998671,
+      "count": 136,
+      "sum": 2309282.0
+    },
+    "time_to_second_token": {
+      "unit": "ms",
+      "avg": 117.59320249999999,
+      "p1": 0.0,
+      "p5": 0.087128,
+      "p10": 0.8168734999999999,
+      "p25": 19.23249575,
+      "p50": 24.107184,
+      "p75": 30.056447499999997,
+      "p90": 309.97833149999997,
+      "p95": 641.811462,
+      "p99": 1705.665427400003,
+      "min": 0.0,
+      "max": 1932.420517,
+      "std": 317.1135227766937,
+      "count": 136,
+      "sum": 15992.67554
+    },
+    "http_req_chunks_received": {
+      "unit": "count",
+      "avg": 125.91176470588235,
+      "p1": 22.35,
+      "p5": 25.0,
+      "p10": 29.0,
+      "p25": 36.0,
+      "p50": 49.0,
+      "p75": 107.5,
+      "p90": 335.5,
+      "p95": 545.25,
+      "p99": 878.0000000000001,
+      "min": 18.0,
+      "max": 899.0,
+      "std": 184.9903602650783,
+      "count": 136,
+      "sum": 17124.0
+    },
+    "prefill_throughput_per_user": {
+      "unit": "tokens/sec/user",
+      "avg": 33538.981899540915,
+      "p1": 2802.3777274293693,
+      "p5": 3556.1511535182913,
+      "p10": 12350.113501290707,
+      "p25": 17070.86643065596,
+      "p50": 28830.622750802282,
+      "p75": 49800.33547872987,
+      "p90": 61211.0529810102,
+      "p95": 67456.57651127188,
+      "p99": 77848.58485263704,
+      "min": 2428.4584190250284,
+      "max": 81496.99468343651,
+      "std": 19525.96369310937,
+      "count": 136,
+      "sum": 4561301.538337564
+    },
+    "e2e_output_token_throughput": {
+      "unit": "tokens/sec/user",
+      "avg": 42.145954328053264,
+      "p1": 11.551396540508302,
+      "p5": 16.058165819700662,
+      "p10": 24.89844958935006,
+      "p25": 31.260298188695643,
+      "p50": 40.59301098855299,
+      "p75": 52.2666088162327,
+      "p90": 63.18495034905247,
+      "p95": 69.57774483696801,
+      "p99": 75.10304189404336,
+      "min": 10.408008525136616,
+      "max": 90.42262551778853,
+      "std": 15.341398449547805,
+      "count": 136,
+      "sum": 5731.849788615244
+    },
+    "osl_mismatch_diff_pct": {
+      "unit": "%",
+      "avg": -81.41729796974377,
+      "p1": -96.11220207321418,
+      "p5": -95.64658410544766,
+      "p10": -95.15477695212178,
+      "p25": -93.94539648920124,
+      "p50": -91.83987673443833,
+      "p75": -84.05738626492922,
+      "p90": -42.64497333418585,
+      "p95": -21.71376544933127,
+      "p99": 0.0,
+      "min": -96.68813247470102,
+      "max": 0.0,
+      "std": 24.067824075293863,
+      "count": 136,
+      "sum": -11072.752523885152
+    },
+    "inter_token_latency": {
+      "unit": "ms",
+      "avg": 20.184033970438747,
+      "p1": 11.572004384649121,
+      "p5": 12.329982490957446,
+      "p10": 12.787635950763217,
+      "p25": 14.386378485757389,
+      "p50": 18.592902382035174,
+      "p75": 23.750563830136365,
+      "p90": 30.76660652540107,
+      "p95": 32.66437963007179,
+      "p99": 43.03067045594598,
+      "min": 10.885774661341854,
+      "max": 45.7391756491228,
+      "std": 7.0615651200068354,
+      "count": 136,
+      "sum": 2745.0286199796697
+    },
+    "http_req_connection_overhead": {
+      "unit": "ms",
+      "avg": 0.18173928676470585,
+      "p1": 0.0,
+      "p5": 0.0,
+      "p10": 0.0,
+      "p25": 0.0,
+      "p50": 0.0,
+      "p75": 0.0,
+      "p90": 0.0,
+      "p95": 2.21050025,
+      "p99": 2.6565537500000005,
+      "min": 0.0,
+      "max": 3.776115,
+      "std": 0.6473690723335254,
+      "count": 136,
+      "sum": 24.716542999999998
+    },
+    "http_req_total": {
+      "unit": "ms",
+      "avg": 5029.823474227941,
+      "p1": 1079.04369315,
+      "p5": 1250.3358027499999,
+      "p10": 1481.6827644999998,
+      "p25": 1912.5112405,
+      "p50": 2727.4463615,
+      "p75": 6077.48639175,
+      "p90": 10550.435273,
+      "p95": 18794.07998575,
+      "p99": 31938.803667950076,
+      "min": 1042.31034,
+      "max": 37020.591443,
+      "std": 5954.6392057396015,
+      "count": 136,
+      "sum": 684055.992495
+    },
+    "osl_mismatch_count": {
+      "unit": "requests",
+      "avg": 129.0
+    },
+    "output_token_throughput_per_user": {
+      "unit": "tokens/sec/user",
+      "avg": 55.02010766171066,
+      "p1": 23.35278754955469,
+      "p5": 30.61668633028668,
+      "p10": 32.50714046725825,
+      "p25": 42.104590564817876,
+      "p50": 53.784944390904,
+      "p75": 69.51519377181677,
+      "p90": 78.20403411692514,
+      "p95": 81.10781757033855,
+      "p99": 86.42197912034528,
+      "min": 21.863096258473522,
+      "max": 91.86300755896166,
+      "std": 16.725668662911218,
+      "count": 136,
+      "sum": 7482.73464199265
+    },
+    "usage_prompt_cache_read_tokens": {
+      "unit": "tokens",
+      "avg": 16122.625954198473,
+      "p1": 7744.0,
+      "p5": 8480.0,
+      "p10": 8832.0,
+      "p25": 11584.0,
+      "p50": 15296.0,
+      "p75": 20544.0,
+      "p90": 23616.0,
+      "p95": 25216.0,
+      "p99": 28076.8,
+      "min": 7744.0,
+      "max": 28416.0,
+      "std": 5422.3694946421365,
+      "count": 131,
+      "sum": 2112064.0
+    },
+    "inter_chunk_latency": {
+      "unit": "ms",
+      "avg": 33.19527276312471,
+      "p1": 0.0,
+      "p5": 0.10740799999999999,
+      "p10": 0.8920714,
+      "p25": 21.759397,
+      "p50": 24.646787,
+      "p75": 26.4873405,
+      "p90": 47.914179000000026,
+      "p95": 62.719750699999935,
+      "p99": 312.25494125999995,
+      "min": 0.0,
+      "max": 1932.420517,
+      "std": 63.867168727736136,
+      "count": 17448,
+      "sum": 579191.119171
+    },
+    "total_output_tokens": {
+      "unit": "tokens",
+      "avg": 32609.0
+    },
+    "total_usage_prompt_cache_read_tokens": {
+      "unit": "tokens",
+      "avg": 2112064.0
+    },
+    "total_usage_prompt_tokens": {
+      "unit": "tokens",
+      "avg": 2276673.0
+    },
+    "total_usage_completion_tokens": {
+      "unit": "tokens",
+      "avg": 32609.0
+    },
+    "total_isl": {
+      "unit": "tokens",
+      "avg": 2276673.0
+    },
+    "total_osl": {
+      "unit": "tokens",
+      "avg": 32609.0
+    },
+    "total_usage_total_tokens": {
+      "unit": "tokens",
+      "avg": 2309282.0
+    },
+    "overall_usage_prompt_cache_read_pct": {
+      "unit": "%",
+      "avg": 92.76975657022331
+    },
+    "benchmark_duration": {
+      "unit": "sec",
+      "avg": 136.26778828800002
+    },
+    "request_throughput": {
+      "unit": "requests/sec",
+      "avg": 0.9978949337473823
+    },
+    "total_token_throughput": {
+      "unit": "tokens/sec",
+      "avg": 16944.270649956048
+    },
+    "output_token_throughput": {
+      "unit": "tokens/sec",
+      "avg": 239.26732275417933
+    },
+    "input_token_throughput": {
+      "unit": "tokens/sec",
+      "avg": 16705.00332720187
+    },
+    "effective_concurrency": {
+      "unit": "requests",
+      "avg": 5.019202293589894,
+      "p50": 5.0,
+      "p90": 8.0,
+      "p95": 8.0,
+      "p99": 8.0,
+      "min": 0.0,
+      "max": 8.0,
+      "std": 2.1713611337379772
+    },
+    "effective_decode_throughput": {
+      "unit": "tokens/sec",
+      "avg": 238.26879558694176,
+      "p50": 172.88277034897084,
+      "p90": 492.34730867757344,
+      "p95": 553.9263862812562,
+      "p99": 919.6301134145355,
+      "min": 0.0,
+      "max": 100457.01770453555,
+      "std": 686.1943110422081
+    },
+    "effective_prefill_throughput": {
+      "unit": "tokens/sec",
+      "avg": 16705.003344239416,
+      "p50": 0.0,
+      "p90": 60570.302967557494,
+      "p95": 71899.79835119536,
+      "p99": 97434.76346614833,
+      "min": 0.0,
+      "max": 128264.32602223808,
+      "std": 26735.78962246377
+    },
+    "effective_decode_concurrency": {
+      "unit": "requests",
+      "avg": 4.268152363952053,
+      "p50": 4.0,
+      "p90": 7.0,
+      "p95": 8.0,
+      "p99": 8.0,
+      "min": 0.0,
+      "max": 8.0,
+      "std": 2.041035239220169
+    },
+    "effective_prefill_concurrency": {
+      "unit": "requests",
+      "avg": 0.7510499296378411,
+      "p50": 0.0,
+      "p90": 2.0,
+      "p95": 4.0,
+      "p99": 8.0,
+      "min": 0.0,
+      "max": 8.0,
+      "std": 1.5543101451295402
+    },
+    "effective_total_throughput": {
+      "unit": "tokens/sec",
+      "avg": 16943.272772302807,
+      "p50": 322.5267586198214,
+      "p90": 60900.36892178564,
+      "p95": 71931.14136306154,
+      "p99": 97611.3887229846,
+      "min": 0.0,
+      "max": 128440.12023235495,
+      "std": 26709.98199135904
+    },
+    "effective_decode_throughput_per_user": {
+      "unit": "tokens/sec/user",
+      "avg": 57.00919772461379,
+      "p50": 49.64107916609465,
+      "p90": 90.62224053974492,
+      "p95": 112.13758247548627,
+      "p99": 184.24042793724794,
+      "min": 0.0,
+      "max": 33485.67256817852,
+      "std": 168.26177289902688
+    },
+    "effective_prefill_throughput_per_user": {
+      "unit": "tokens/sec/user",
+      "avg": 11438.219610655426,
+      "p50": 0.0,
+      "p90": 48666.55140995103,
+      "p95": 59825.784866239555,
+      "p99": 72630.16291795287,
+      "min": 0.0,
+      "max": 81496.94951531618,
+      "std": 19925.161835514875
+    },
+    "tokens_in_flight": {
+      "unit": "tokens",
+      "avg": 93588.07150949888,
+      "p50": 97634.6666666665,
+      "p90": 130484.86028915553,
+      "p95": 134951.07706657276,
+      "p99": 155894.0782820748,
+      "min": 0.0,
+      "max": 156934.50854530174,
+      "std": 31166.907646808675
+    },
+    "active_decode_throughput": {
+      "unit": "tokens/sec",
+      "avg": 243.52887258798575,
+      "p50": 184.3567904121298,
+      "p90": 493.78208606766447,
+      "p95": 554.4398635342595,
+      "p99": 939.605883037496,
+      "min": 0.0,
+      "max": 100457.01770453555,
+      "std": 692.8033806123002
+    },
+    "active_prefill_throughput": {
+      "unit": "tokens/sec",
+      "avg": 49371.743795234855,
+      "p50": 49712.6413172078,
+      "p90": 74011.20006847722,
+      "p95": 81496.94951531618,
+      "p99": 126280.45738144607,
+      "min": 2826.1686613754514,
+      "max": 128264.32602223808,
+      "std": 22355.96664135998
+    },
+    "active_decode_throughput_per_user": {
+      "unit": "tokens/sec/user",
+      "avg": 58.26774594978326,
+      "p50": 51.65452876105438,
+      "p90": 90.81718632982195,
+      "p95": 112.16160159460493,
+      "p99": 194.6328406855248,
+      "min": 0.0,
+      "max": 33485.67256817852,
+      "std": 169.89324365944378
+    },
+    "active_prefill_throughput_per_user": {
+      "unit": "tokens/sec/user",
+      "avg": 33805.730921068716,
+      "p50": 29283.961086519088,
+      "p90": 61884.4194864873,
+      "p95": 67846.39920982516,
+      "p99": 79937.76733602778,
+      "min": 2813.201170237322,
+      "max": 81496.94951531618,
+      "std": 20426.0041480925
+    },
+    "active_total_throughput": {
+      "unit": "tokens/sec",
+      "avg": 16945.648190309254,
+      "p50": 322.5267586198214,
+      "p90": 60900.36892178564,
+      "p95": 71931.14136306154,
+      "p99": 97611.3887229846,
+      "min": 31.34301186616326,
+      "max": 128440.12023235495,
+      "std": 26711.10079518685
+    },
+    "credit_to_start_latency": {
+      "unit": "ms",
+      "avg": 1.6261082352941174,
+      "p1": 1.0134016,
+      "p5": 1.133888,
+      "p10": 1.16672,
+      "p25": 1.22368,
+      "p50": 1.36192,
+      "p75": 1.516096,
+      "p90": 2.5011200000000002,
+      "p95": 3.845184,
+      "p99": 4.448140800000002,
+      "min": 0.927488,
+      "max": 5.98016,
+      "std": 0.8193703805084321,
+      "count": 136,
+      "sum": 221.15071999999998
+    },
+    "effective_latency": {
+      "unit": "ms",
+      "avg": 5031.416438588235,
+      "p1": 1080.4216448,
+      "p5": 1252.227904,
+      "p10": 1483.177984,
+      "p25": 1913.9290879999999,
+      "p50": 2728.993792,
+      "p75": 6079.020288,
+      "p90": 10552.518784,
+      "p95": 18795.782976,
+      "p99": 31940.222054400074,
+      "min": 1042.368,
+      "max": 37022.035456,
+      "std": 5954.648235145611,
+      "count": 136,
+      "sum": 684272.6356479999
+    }
+  },
+  "http_req_dns_lookup": {
+    "unit": "ms",
+    "avg": 0.0,
+    "p1": 0.0,
+    "p5": 0.0,
+    "p10": 0.0,
+    "p25": 0.0,
+    "p50": 0.0,
+    "p75": 0.0,
+    "p90": 0.0,
+    "p95": 0.0,
+    "p99": 0.0,
+    "min": 0.0,
+    "max": 0.0,
+    "std": 0.0,
+    "count": 676,
+    "sum": 0.0
+  },
+  "http_req_sending": {
+    "unit": "ms",
+    "avg": 0.32405490828402367,
+    "p1": 0.1929695,
+    "p5": 0.19984074999999998,
+    "p10": 0.203729,
+    "p25": 0.21566375,
+    "p50": 0.24769149999999998,
+    "p75": 0.29615674999999997,
+    "p90": 0.3942495,
+    "p95": 0.49635724999999997,
+    "p99": 2.172116,
+    "min": 0.184547,
+    "max": 2.832894,
+    "std": 0.3223752381276766,
+    "count": 676,
+    "sum": 219.061118
+  },
+  "http_req_data_sent": {
+    "unit": "KB",
+    "avg": 106.25093900240384,
+    "p1": 37.8125,
+    "p5": 42.019287109375,
+    "p10": 46.62939453125,
+    "p25": 67.578857421875,
+    "p50": 98.57470703125,
+    "p75": 141.337646484375,
+    "p90": 172.810546875,
+    "p95": 193.025634765625,
+    "p99": 223.180908203125,
+    "min": 36.833984375,
+    "max": 232.431640625,
+    "std": 47.65063795604463,
+    "count": 676,
+    "sum": 71825.634765625
+  },
+  "http_req_chunks_sent": {
+    "unit": "count",
+    "avg": 1.0,
+    "p1": 1.0,
+    "p5": 1.0,
+    "p10": 1.0,
+    "p25": 1.0,
+    "p50": 1.0,
+    "p75": 1.0,
+    "p90": 1.0,
+    "p95": 1.0,
+    "p99": 1.0,
+    "min": 1.0,
+    "max": 1.0,
+    "std": 0.0,
+    "count": 676,
+    "sum": 676.0
+  },
+  "http_req_data_received": {
+    "unit": "KB",
+    "avg": 55.67273888221154,
+    "p1": 5.572021484375,
+    "p5": 7.329833984375,
+    "p10": 8.587890625,
+    "p25": 11.673828125,
+    "p50": 22.826171875,
+    "p75": 94.876708984375,
+    "p90": 149.357421875,
+    "p95": 172.2939453125,
+    "p99": 209.474853515625,
+    "min": 4.310546875,
+    "max": 255.55859375,
+    "std": 59.39577980427408,
+    "count": 676,
+    "sum": 37634.771484375
+  },
+  "time_to_first_output_token": {
+    "unit": "ms",
+    "avg": 3095.2219346767165,
+    "p1": 368.70130908,
+    "p5": 384.06765939999997,
+    "p10": 396.17521,
+    "p25": 437.87580699999995,
+    "p50": 765.6034589999999,
+    "p75": 2675.070749,
+    "p90": 10905.097361999993,
+    "p95": 15879.62832419998,
+    "p99": 21981.241319919965,
+    "min": 351.49622899999997,
+    "max": 25785.741812,
+    "std": 5076.126782004231,
+    "count": 597,
+    "sum": 1847847.495002
+  },
+  "http_req_duration": {
+    "unit": "ms",
+    "avg": 7588.785109588757,
+    "p1": 932.1455639999999,
+    "p5": 1193.4641482499999,
+    "p10": 1368.2155735,
+    "p25": 1867.5132342499999,
+    "p50": 3518.5519084999996,
+    "p75": 12330.2090655,
+    "p90": 20498.632491,
+    "p95": 22648.8628685,
+    "p99": 27607.178034,
+    "min": 743.512224,
+    "max": 30522.138736999997,
+    "std": 7683.124718461235,
+    "count": 676,
+    "sum": 5130018.734081999
+  },
+  "http_req_connection_reused": {
+    "unit": "ratio",
+    "avg": 0.9644970414201184,
+    "p1": 0.0,
+    "p5": 1.0,
+    "p10": 1.0,
+    "p25": 1.0,
+    "p50": 1.0,
+    "p75": 1.0,
+    "p90": 1.0,
+    "p95": 1.0,
+    "p99": 1.0,
+    "min": 0.0,
+    "max": 1.0,
+    "std": 0.18504728723209338,
+    "count": 676,
+    "sum": 652.0
+  },
+  "http_req_blocked": {
+    "unit": "ms",
+    "avg": 0.0,
+    "p1": 0.0,
+    "p5": 0.0,
+    "p10": 0.0,
+    "p25": 0.0,
+    "p50": 0.0,
+    "p75": 0.0,
+    "p90": 0.0,
+    "p95": 0.0,
+    "p99": 0.0,
+    "min": 0.0,
+    "max": 0.0,
+    "std": 0.0,
+    "count": 676,
+    "sum": 0.0
+  },
+  "usage_prompt_tokens": {
+    "unit": "tokens",
+    "avg": 26460.34171597633,
+    "p1": 8347.0,
+    "p5": 9649.5,
+    "p10": 10918.0,
+    "p25": 16504.5,
+    "p50": 24594.5,
+    "p75": 35760.5,
+    "p90": 44064.5,
+    "p95": 47562.5,
+    "p99": 55921.25,
+    "min": 8103.0,
+    "max": 59589.0,
+    "std": 12309.779652807998,
+    "count": 676,
+    "sum": 17887191.0
+  },
+  "http_req_receiving": {
+    "unit": "ms",
+    "avg": 7050.283790856509,
+    "p1": 509.82421875,
+    "p5": 722.997499,
+    "p10": 887.3421374999999,
+    "p25": 1360.8973065,
+    "p50": 2766.0955375,
+    "p75": 11843.126752999999,
+    "p90": 19889.5180715,
+    "p95": 22169.85721,
+    "p99": 27179.05023875,
+    "min": 218.36265799999998,
+    "max": 29753.203737,
+    "std": 7681.213273701946,
+    "count": 676,
+    "sum": 4765991.842619
+  },
+  "usage_completion_tokens": {
+    "unit": "tokens",
+    "avg": 421.7130177514793,
+    "p1": 38.0,
+    "p5": 52.0,
+    "p10": 61.0,
+    "p25": 83.0,
+    "p50": 164.5,
+    "p75": 699.75,
+    "p90": 1138.0,
+    "p95": 1345.75,
+    "p99": 1651.0,
+    "min": 28.0,
+    "max": 1990.0,
+    "std": 460.7059002410634,
+    "count": 676,
+    "sum": 285078.0
+  },
+  "http_req_connecting": {
+    "unit": "ms",
+    "avg": 0.0461207899408284,
+    "p1": 0.0,
+    "p5": 0.0,
+    "p10": 0.0,
+    "p25": 0.0,
+    "p50": 0.0,
+    "p75": 0.0,
+    "p90": 0.0,
+    "p95": 0.0,
+    "p99": 1.42599375,
+    "min": 0.0,
+    "max": 1.806948,
+    "std": 0.24440212168255043,
+    "count": 676,
+    "sum": 31.177653999999997
+  },
+  "http_req_waiting": {
+    "unit": "ms",
+    "avg": 538.1772638239645,
+    "p1": 328.44534999999996,
+    "p5": 342.396342,
+    "p10": 354.0692275,
+    "p25": 375.98796175,
+    "p50": 419.590978,
+    "p75": 625.6224575,
+    "p90": 764.171016,
+    "p95": 907.32358825,
+    "p99": 2700.461703,
+    "min": 304.286257,
+    "max": 3187.355082,
+    "std": 340.91516286029275,
+    "count": 676,
+    "sum": 363807.83034499997
+  },
+  "usage_total_tokens": {
+    "unit": "tokens",
+    "avg": 26882.05473372781,
+    "p1": 8420.25,
+    "p5": 9724.5,
+    "p10": 11011.5,
+    "p25": 16748.75,
+    "p50": 24941.0,
+    "p75": 36258.5,
+    "p90": 44627.0,
+    "p95": 48540.75,
+    "p99": 56244.5,
+    "min": 8188.0,
+    "max": 60066.0,
+    "std": 12438.899232997308,
+    "count": 676,
+    "sum": 18172269.0
+  },
+  "http_req_chunks_received": {
+    "unit": "count",
+    "avg": 222.92603550295857,
+    "p1": 21.0,
+    "p5": 27.0,
+    "p10": 32.0,
+    "p25": 44.0,
+    "p50": 88.5,
+    "p75": 367.5,
+    "p90": 607.5,
+    "p95": 704.25,
+    "p99": 837.5,
+    "min": 15.0,
+    "max": 1032.0,
+    "std": 242.0664864832916,
+    "count": 676,
+    "sum": 150698.0
+  },
+  "prefill_throughput_per_user": {
+    "unit": "tokens/sec/user",
+    "avg": 55629.37730280787,
+    "p1": 3661.4958380021776,
+    "p5": 14786.756624493777,
+    "p10": 21339.382762793262,
+    "p25": 31620.28122056382,
+    "p50": 52781.87274662545,
+    "p75": 76390.86205347342,
+    "p90": 96799.69524736652,
+    "p95": 107052.73788029251,
+    "p99": 124219.15923818217,
+    "min": 2852.000016996057,
+    "max": 132048.84780878513,
+    "std": 28822.231302942964,
+    "count": 676,
+    "sum": 37605459.05669812
+  },
+  "e2e_output_token_throughput": {
+    "unit": "tokens/sec/user",
+    "avg": 51.35533242616132,
+    "p1": 21.176504804192316,
+    "p5": 29.917414476408478,
+    "p10": 37.341606914346585,
+    "p25": 46.29374222341738,
+    "p50": 53.043970969087226,
+    "p75": 57.84777090797985,
+    "p90": 62.72436156503562,
+    "p95": 66.3837296537102,
+    "p99": 71.7289324349675,
+    "min": 17.976746164020202,
+    "max": 80.45477889281663,
+    "std": 10.506196388185163,
+    "count": 676,
+    "sum": 34716.20472008505
+  },
+  "osl_mismatch_diff_pct": {
+    "unit": "%",
+    "avg": -66.05746489953965,
+    "p1": -97.22669426556271,
+    "p5": -95.38160046167164,
+    "p10": -94.63160614552899,
+    "p25": -92.80882633717917,
+    "p50": -85.75349732149336,
+    "p75": -42.08018566869742,
+    "p90": 0.0,
+    "p95": 0.0,
+    "p99": 0.0,
+    "min": -97.57498404594767,
+    "max": 0.0,
+    "std": 35.5227961223412,
+    "count": 676,
+    "sum": -44654.846272088806
+  },
+  "http_req_connection_overhead": {
+    "unit": "ms",
+    "avg": 0.0461207899408284,
+    "p1": 0.0,
+    "p5": 0.0,
+    "p10": 0.0,
+    "p25": 0.0,
+    "p50": 0.0,
+    "p75": 0.0,
+    "p90": 0.0,
+    "p95": 0.0,
+    "p99": 1.42599375,
+    "min": 0.0,
+    "max": 1.806948,
+    "std": 0.24440212168255043,
+    "count": 676,
+    "sum": 31.177653999999997
+  },
+  "http_req_total": {
+    "unit": "ms",
+    "avg": 7588.831230378698,
+    "p1": 932.5973009999999,
+    "p5": 1193.4641482499999,
+    "p10": 1368.2155735,
+    "p25": 1867.5132342499999,
+    "p50": 3518.5519084999996,
+    "p75": 12330.525196999999,
+    "p90": 20498.632491,
+    "p95": 22648.8628685,
+    "p99": 27607.178034,
+    "min": 744.27739,
+    "max": 30522.138736999997,
+    "std": 7683.098542813112,
+    "count": 676,
+    "sum": 5130049.911735999
+  },
+  "osl_mismatch_count": {
+    "unit": "requests",
+    "avg": 567.0
+  },
+  "usage_prompt_cache_read_tokens": {
+    "unit": "tokens",
+    "avg": 25854.850299401198,
+    "p1": 7744.0,
+    "p5": 8790.4,
+    "p10": 10412.8,
+    "p25": 15872.0,
+    "p50": 24160.0,
+    "p75": 35136.0,
+    "p90": 43430.40000000001,
+    "p95": 47168.0,
+    "p99": 55658.240000000005,
+    "min": 7744.0,
+    "max": 59456.0,
+    "std": 12344.32246505665,
+    "count": 668,
+    "sum": 17271040.0
+  },
+  "total_usage_prompt_cache_read_tokens": {
+    "unit": "tokens",
+    "avg": 17271040.0
+  },
+  "total_usage_prompt_tokens": {
+    "unit": "tokens",
+    "avg": 17887191.0
+  },
+  "total_usage_completion_tokens": {
+    "unit": "tokens",
+    "avg": 285078.0
+  },
+  "total_usage_total_tokens": {
+    "unit": "tokens",
+    "avg": 18172269.0
+  },
+  "overall_usage_prompt_cache_read_pct": {
+    "unit": "%",
+    "avg": 96.55535069760255
+  },
+  "total_token_throughput": {
+    "unit": "tokens/sec",
+    "avg": 19835.160714881145
+  },
+  "input_token_throughput": {
+    "unit": "tokens/sec",
+    "avg": 19523.99605259946
+  },
+  "effective_concurrency": {
+    "unit": "requests",
+    "avg": 5.599559737183051,
+    "p50": 6.0,
+    "p90": 7.0,
+    "p95": 8.0,
+    "p99": 8.0,
+    "min": 0.0,
+    "max": 8.0,
+    "std": 1.4918262593084803
+  },
+  "effective_decode_throughput": {
+    "unit": "tokens/sec",
+    "avg": 310.4267936105399,
+    "p50": 277.01867795366957,
+    "p90": 503.06199703540585,
+    "p95": 544.8118980583299,
+    "p99": 1210.2174285467254,
+    "min": 0.0,
+    "max": 131252.22598194997,
+    "std": 716.4642939507476
+  },
+  "effective_prefill_throughput": {
+    "unit": "tokens/sec",
+    "avg": 19523.996051171656,
+    "p50": 0.0,
+    "p90": 80157.46257862731,
+    "p95": 98007.839562881,
+    "p99": 130393.5769899856,
+    "min": 0.0,
+    "max": 229946.70148518644,
+    "std": 35770.73055936496
+  },
+  "effective_decode_concurrency": {
+    "unit": "requests",
+    "avg": 5.192594545630038,
+    "p50": 5.0,
+    "p90": 7.0,
+    "p95": 7.0,
+    "p99": 8.0,
+    "min": 0.0,
+    "max": 8.0,
+    "std": 1.547710943280249
+  },
+  "effective_prefill_concurrency": {
+    "unit": "requests",
+    "avg": 0.4069651915530127,
+    "p50": 0.0,
+    "p90": 1.0,
+    "p95": 2.0,
+    "p99": 3.0,
+    "min": 0.0,
+    "max": 8.0,
+    "std": 0.818925411144775
+  },
+  "effective_total_throughput": {
+    "unit": "tokens/sec",
+    "avg": 19834.42285462804,
+    "p50": 379.431613502167,
+    "p90": 80401.80043864233,
+    "p95": 98288.81895484749,
+    "p99": 130635.95328606645,
+    "min": 0.0,
+    "max": 229946.70148518635,
+    "std": 35740.405300536884
+  },
+  "effective_decode_throughput_per_user": {
+    "unit": "tokens/sec/user",
+    "avg": 58.467390904891026,
+    "p50": 68.82969553226984,
+    "p90": 83.89320212815139,
+    "p95": 90.26449757036438,
+    "p99": 229.0756823080769,
+    "min": 0.0,
+    "max": 32813.05649548749,
+    "std": 146.21776689216983
+  },
+  "effective_prefill_throughput_per_user": {
+    "unit": "tokens/sec/user",
+    "avg": 15775.266919438609,
+    "p50": 0.0,
+    "p90": 63994.09176496421,
+    "p95": 85262.31775703348,
+    "p99": 111797.559625635,
+    "min": 0.0,
+    "max": 132048.85250915372,
+    "std": 29424.47185798128
+  },
+  "tokens_in_flight": {
+    "unit": "tokens",
+    "avg": 166795.7899371402,
+    "p50": 170662.29057100174,
+    "p90": 231076.5400655876,
+    "p95": 245748.87365878126,
+    "p99": 263693.7897202315,
+    "min": 0.0,
+    "max": 271358.4102345822,
+    "std": 49648.27608404763
+  },
+  "active_decode_throughput": {
+    "unit": "tokens/sec",
+    "avg": 311.63323563158866,
+    "p50": 280.0463614976609,
+    "p90": 503.13720064878135,
+    "p95": 545.0786800524462,
+    "p99": 1211.3854873705634,
+    "min": 0.0,
+    "max": 131252.22598194997,
+    "std": 717.5932601629097
+  },
+  "active_prefill_throughput": {
+    "unit": "tokens/sec",
+    "avg": 69134.62730833299,
+    "p50": 66238.67784332343,
+    "p90": 113039.84693070392,
+    "p95": 124902.6503148783,
+    "p99": 161713.59223088663,
+    "min": 3161.0290782330126,
+    "max": 229946.70148518644,
+    "std": 33182.3331651739
+  },
+  "active_decode_throughput_per_user": {
+    "unit": "tokens/sec/user",
+    "avg": 58.69461844678047,
+    "p50": 68.96270857458622,
+    "p90": 83.91716427719079,
+    "p95": 90.31279311864056,
+    "p99": 229.63228874144752,
+    "min": 0.0,
+    "max": 32813.05649548749,
+    "std": 146.45609611684083
+  },
+  "active_prefill_throughput_per_user": {
+    "unit": "tokens/sec/user",
+    "avg": 55860.347251986524,
+    "p50": 52626.89608475234,
+    "p90": 96782.2337203919,
+    "p95": 107059.87680097185,
+    "p99": 124351.10127761125,
+    "min": 3006.5146357024723,
+    "max": 132048.85250915372,
+    "std": 28751.289603639198
+  },
+  "active_total_throughput": {
+    "unit": "tokens/sec",
+    "avg": 19836.750883863242,
+    "p50": 379.431613502167,
+    "p90": 80401.80043864233,
+    "p95": 98288.81895484749,
+    "p99": 130635.95328606645,
+    "min": 62.64035932625538,
+    "max": 229946.70148518635,
+    "std": 35741.85669852253
+  },
+  "credit_to_start_latency": {
+    "unit": "ms",
+    "avg": 1.5817516686390534,
+    "p1": 0.940416,
+    "p5": 1.049344,
+    "p10": 1.099264,
+    "p25": 1.203776,
+    "p50": 1.3523200000000002,
+    "p75": 1.5611519999999999,
+    "p90": 2.027776,
+    "p95": 3.330944,
+    "p99": 5.589888,
+    "min": 0.827392,
+    "max": 7.5648,
+    "std": 0.8233047656534758,
+    "count": 676,
+    "sum": 1069.264128
+  },
+  "effective_latency": {
+    "unit": "ms",
+    "avg": 7590.513019455621,
+    "p1": 934.675712,
+    "p5": 1195.097472,
+    "p10": 1369.7276160000001,
+    "p25": 1868.8777599999999,
+    "p50": 3519.809408,
+    "p75": 12332.026496,
+    "p90": 20500.159488,
+    "p95": 22650.66048,
+    "p99": 27608.60544,
+    "min": 744.70528,
+    "max": 30524.443904,
+    "std": 7683.130048789198,
+    "count": 676,
+    "sum": 5131186.801152
+  }
+}

experiments/kv-cache-fp8/capacity/hammer/N0008/profile_export_console.txt

@@ -0,0 +1,87 @@
+
+
+                                              NVIDIA AIPerf | LLM Metrics: Effective                                               
+┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━┓
+┃                                    Metric ┃        avg ┃    min ┃        max ┃        p99 ┃        p90 ┃        p50 ┃       std ┃
+┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━┩
+│          Effective Concurrency (requests) │       5.60 │   0.00 │       8.00 │       8.00 │       7.00 │       6.00 │      1.49 │
+│  Effective Decode Throughput (tokens/sec) │     310.43 │   0.00 │ 131,252.23 │   1,210.22 │     503.06 │     277.02 │    716.46 │
+│ Effective Prefill Throughput (tokens/sec) │  19,524.00 │   0.00 │ 229,946.70 │ 130,393.58 │  80,157.46 │       0.00 │ 35,770.73 │
+│   Effective Decode Concurrency (requests) │       5.19 │   0.00 │       8.00 │       8.00 │       7.00 │       5.00 │      1.55 │
+│  Effective Prefill Concurrency (requests) │       0.41 │   0.00 │       8.00 │       3.00 │       1.00 │       0.00 │      0.82 │
+│   Effective Total Throughput (tokens/sec) │  19,834.42 │   0.00 │ 229,946.70 │ 130,635.95 │  80,401.80 │     379.43 │ 35,740.41 │
+│      Effective Decode Throughput Per User │      58.47 │   0.00 │  32,813.06 │     229.08 │      83.89 │      68.83 │    146.22 │
+│                         (tokens/sec/user) │            │        │            │            │            │            │           │
+│     Effective Prefill Throughput Per User │  15,775.27 │   0.00 │ 132,048.85 │ 111,797.56 │  63,994.09 │       0.00 │ 29,424.47 │
+│                         (tokens/sec/user) │            │        │            │            │            │            │           │
+│                 Tokens In Flight (tokens) │ 166,795.79 │   0.00 │ 271,358.41 │ 263,693.79 │ 231,076.54 │ 170,662.29 │ 49,648.28 │
+│         Effective Latency (CO-aware) (ms) │   7,590.51 │ 744.71 │  30,524.44 │  27,608.61 │  20,500.16 │   3,519.81 │  7,683.13 │
+└───────────────────────────────────────────┴────────────┴────────┴────────────┴────────────┴────────────┴────────────┴───────────┘
+                                              NVIDIA AIPerf | LLM Metrics: Active                                               
+┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┓
+┃                                 Metric ┃       avg ┃      min ┃        max ┃        p99 ┃        p90 ┃       p50 ┃       std ┃
+┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━┩
+│  Active Decode Throughput (tokens/sec) │    311.63 │     0.00 │ 131,252.23 │   1,211.39 │     503.14 │    280.05 │    717.59 │
+│ Active Prefill Throughput (tokens/sec) │ 69,134.63 │ 3,161.03 │ 229,946.70 │ 161,713.59 │ 113,039.85 │ 66,238.68 │ 33,182.33 │
+│      Active Decode Throughput Per User │     58.69 │     0.00 │  32,813.06 │     229.63 │      83.92 │     68.96 │    146.46 │
+│                      (tokens/sec/user) │           │          │            │            │            │           │           │
+│     Active Prefill Throughput Per User │ 55,860.35 │ 3,006.51 │ 132,048.85 │ 124,351.10 │  96,782.23 │ 52,626.90 │ 28,751.29 │
+│                      (tokens/sec/user) │           │          │            │            │            │           │           │
+│   Active Total Throughput (tokens/sec) │ 19,836.75 │    62.64 │ 229,946.70 │ 130,635.95 │  80,401.80 │    379.43 │ 35,741.86 │
+└────────────────────────────────────────┴───────────┴──────────┴────────────┴────────────┴────────────┴───────────┴───────────┘
+                                                NVIDIA AIPerf | LLM Metrics: Usage                                                
+┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┓
+┃                                  Metric ┃           avg ┃      min ┃       max ┃       p99 ┃       p90 ┃       p50 ┃       std ┃
+┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━┩
+│            Usage Prompt Tokens (tokens) │     26,460.34 │ 8,103.00 │ 59,589.00 │ 55,921.25 │ 44,064.50 │ 24,594.50 │ 12,309.78 │
+│ Usage Prompt Cache Read Tokens (tokens) │     25,854.85 │ 7,744.00 │ 59,456.00 │ 55,658.24 │ 43,430.40 │ 24,160.00 │ 12,344.32 │
+│        Usage Completion Tokens (tokens) │        421.71 │    28.00 │  1,990.00 │  1,651.00 │  1,138.00 │    164.50 │    460.71 │
+│             Usage Total Tokens (tokens) │     26,882.05 │ 8,188.00 │ 60,066.00 │ 56,244.50 │ 44,627.00 │ 24,941.00 │ 12,438.90 │
+│      Total Usage Prompt Tokens (tokens) │ 17,887,191.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│    Total Usage Prompt Cache Read Tokens │ 17,271,040.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│                                (tokens) │               │          │           │           │           │           │           │
+│       Overall Usage Prompt Cache Read % │         96.56 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│                                     (%) │               │          │           │           │           │           │           │
+│  Total Usage Completion Tokens (tokens) │    285,078.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│       Total Usage Total Tokens (tokens) │ 18,172,269.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+└─────────────────────────────────────────┴───────────────┴──────────┴───────────┴───────────┴───────────┴───────────┴───────────┘
+                                                    NVIDIA AIPerf | LLM Metrics                                                     
+┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┓
+┃                                        Metric ┃       avg ┃      min ┃       max ┃       p99 ┃       p90 ┃       p50 ┃       std ┃
+┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━┩
+│                      Time to First Token (ms) │    551.55 │   304.69 │  3,189.82 │  2,702.84 │    793.01 │    429.83 │    345.27 │
+│                     Time to Second Token (ms) │     39.74 │     0.00 │    612.17 │    416.83 │     49.18 │     23.69 │     77.15 │
+│               Time to First Output Token (ms) │  3,095.22 │   351.50 │ 25,785.74 │ 21,981.24 │ 10,905.10 │    765.60 │  5,076.13 │
+│                          Request Latency (ms) │  7,573.56 │   743.63 │ 30,522.45 │ 27,607.30 │ 20,473.26 │  3,517.85 │  7,686.34 │
+│                      Inter Token Latency (ms) │     16.58 │     5.89 │     40.62 │     32.43 │     19.92 │     16.28 │      4.00 │
+│              Output Token Throughput Per User │     63.15 │    24.62 │    169.66 │     87.25 │     78.87 │     61.44 │     12.97 │
+│                             (tokens/sec/user) │           │          │           │           │           │           │           │
+│ E2E Output Token Throughput (tokens/sec/user) │     51.36 │    17.98 │     80.45 │     71.73 │     62.72 │     53.04 │     10.51 │
+│               Output Sequence Length (tokens) │    421.71 │    28.00 │  1,990.00 │  1,651.00 │  1,138.00 │    164.50 │    460.71 │
+│                Input Sequence Length (tokens) │ 26,460.34 │ 8,103.00 │ 59,589.00 │ 55,921.25 │ 44,064.50 │ 24,594.50 │ 12,309.78 │
+│          Output Token Throughput (tokens/sec) │    311.16 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│           Input Token Throughput (tokens/sec) │ 19,524.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│             Request Throughput (requests/sec) │      0.74 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│                      Request Count (requests) │    676.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+└───────────────────────────────────────────────┴───────────┴──────────┴───────────┴───────────┴───────────┴───────────┴───────────┘
+
+
+No GPU telemetry data collected during the benchmarking run.
+
+╭─────────────────────────────── Output Sequence Length Mismatch Warning ───────────────────────────────╮
+│  567 of 676 requests (83.9%) have output length differing from requested by more than the threshold.  │
+│  Threshold (tokens): min(requested x 5%, 50)                                                          │
+│  Average mismatch: -66.1%                                                                             │
+│                                                                                                       │
+│  Why: Server hit EOS token before reaching requested output length.                                   │
+│                                                                                                       │
+│  Fix Options:                                                                                         │
+│    - --extra-inputs ignore_eos:true - Generate until max_tokens (vLLM, TensorRT-LLM)                  │
+│    - --extra-inputs min_tokens:<N> - Set minimum output length (vLLM, TensorRT-LLM, SGLang)           │
+│    - --use-server-token-count - Use server-reported token counts if tokenizer mismatch suspected      │
+│                                                                                                       │
+│  Diagnostics:                                                                                         │
+│    - Review profile_export.jsonl -> osl_mismatch_diff_pct for per-request values                      │
+│    - Adjust: AIPERF_METRICS_OSL_MISMATCH_PCT_THRESHOLD=5                                              │
+│    - Adjust: AIPERF_METRICS_OSL_MISMATCH_MAX_TOKEN_THRESHOLD=50                                       │
+╰───────────────────────────────────────────────────────────────────────────────────────────────────────╯

experiments/kv-cache-fp8/capacity/hammer/N0008/summary.json

@@ -0,0 +1,25 @@
+{
+  "sent": 676,
+  "records": 676,
+  "valid": 676,
+  "valid_frac": 1.0,
+  "p25_speed_tok_s": 55.9,
+  "p50_speed_tok_s": 61.4,
+  "p95_ttft_ms": 949.0,
+  "p50_ttft_ms": 430.0,
+  "cache_read_ratio": 0.9656,
+  "max_turn_index": 49,
+  "conversations": 21,
+  "recycled_record_frac": 0.0,
+  "inflight_mean": 7.1,
+  "inflight_p95": 10.0,
+  "foreign_p50": 1.0,
+  "foreign_p95": 3.0,
+  "active_sessions_mean": 7.5,
+  "active_sessions_p95": 8.0,
+  "tiers": {
+    "20tok": true,
+    "60tok": false,
+    "180tok": false
+  }
+}

experiments/kv-cache-fp8/capacity/hammer/N0008/verify_cache_salt.log

@@ -0,0 +1 @@
+records_checked=676 mismatches=0

experiments/kv-cache-fp8/capacity/hammer/N0008/verify_pin.log

@@ -0,0 +1 @@
+mode=pinned records_checked=676 mismatches=0

experiments/kv-cache-fp8/capacity/hammer/N0016/aiperf_stdout.log

@@ -0,0 +1,230 @@
+m │ 149,438.85 │  90,235.30 │     606.76 │ 40,016.37 │
+│      Effective Decode Throughput Per User │      42.60 │   0.00 │  13,960.98 │     173.86 │      65.01 │      38.19 │     85.17 │
+│                         (tokens/sec/user) │            │        │            │            │            │            │           │
+│     Effective Prefill Throughput Per User │  20,089.12 │   0.00 │ 141,468.65 │ 109,360.34 │  65,049.87 │       0.00 │ 29,067.61 │
+│                         (tokens/sec/user) │            │        │            │            │            │            │           │
+│                 Tokens In Flight (tokens) │ 356,248.93 │   0.00 │ 505,096.39 │ 492,375.69 │ 453,099.60 │ 362,879.28 │ 77,305.99 │
+│         Effective Latency (CO-aware) (ms) │  10,817.06 │ 935.75 │  52,221.87 │  40,892.60 │  27,953.63 │   5,275.09 │ 10,897.93 │
+└───────────────────────────────────────────┴────────────┴────────┴────────────┴────────────┴────────────┴────────────┴───────────┘
+                                              NVIDIA AIPerf | LLM Metrics: Active                                               
+┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┓
+┃                                 Metric ┃       avg ┃      min ┃        max ┃        p99 ┃        p90 ┃       p50 ┃       std ┃
+┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━┩
+│  Active Decode Throughput (tokens/sec) │    505.97 │     0.00 │ 136,616.40 │   2,087.53 │     821.68 │    452.53 │    980.46 │
+│ Active Prefill Throughput (tokens/sec) │ 67,997.99 │ 2,294.05 │ 211,712.97 │ 163,691.60 │ 114,519.99 │ 62,853.71 │ 34,187.76 │
+│      Active Decode Throughput Per User │     42.74 │     0.00 │  13,960.98 │     174.18 │      65.02 │     38.49 │     85.28 │
+│                      (tokens/sec/user) │           │          │            │            │            │           │           │
+│     Active Prefill Throughput Per User │ 48,627.32 │ 2,294.05 │ 141,468.65 │ 118,005.17 │  85,023.03 │ 44,712.08 │ 25,641.26 │
+│                      (tokens/sec/user) │           │          │            │            │            │           │           │
+│   Active Total Throughput (tokens/sec) │ 28,596.57 │    57.17 │ 212,194.50 │ 149,438.85 │  90,235.30 │    606.76 │ 40,016.60 │
+└────────────────────────────────────────┴───────────┴──────────┴────────────┴────────────┴────────────┴───────────┴───────────┘
+                                                NVIDIA AIPerf | LLM Metrics: Usage                                                
+┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┓
+┃                                  Metric ┃           avg ┃      min ┃       max ┃       p99 ┃       p90 ┃       p50 ┃       std ┃
+┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━┩
+│            Usage Prompt Tokens (tokens) │     24,722.81 │ 8,100.00 │ 64,531.00 │ 57,856.37 │ 42,932.30 │ 22,658.00 │ 12,184.92 │
+│ Usage Prompt Cache Read Tokens (tokens) │     23,983.75 │ 7,744.00 │ 64,384.00 │ 56,792.32 │ 42,163.20 │ 21,696.00 │ 12,131.37 │
+│        Usage Completion Tokens (tokens) │        444.81 │    37.00 │  2,260.00 │  1,814.20 │  1,178.50 │    191.00 │    480.39 │
+│             Usage Total Tokens (tokens) │     25,167.62 │ 8,187.00 │ 65,628.00 │ 58,307.57 │ 43,592.30 │ 23,067.50 │ 12,363.77 │
+│      Total Usage Prompt Tokens (tokens) │ 25,711,718.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│    Total Usage Prompt Cache Read Tokens │ 24,751,232.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│                                (tokens) │               │          │           │           ���           │           │           │
+│       Overall Usage Prompt Cache Read % │         96.26 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│                                     (%) │               │          │           │           │           │           │           │
+│  Total Usage Completion Tokens (tokens) │    462,605.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│       Total Usage Total Tokens (tokens) │ 26,174,323.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+└─────────────────────────────────────────┴───────────────┴──────────┴───────────┴───────────┴───────────┴───────────┴───────────┘
+                                                    NVIDIA AIPerf | LLM Metrics                                                     
+┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┓
+┃                                        Metric ┃       avg ┃      min ┃       max ┃       p99 ┃       p90 ┃       p50 ┃       std ┃
+┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━┩
+│                      Time to First Token (ms) │    628.30 │   325.98 │  3,592.81 │  3,582.36 │    900.76 │    539.64 │    422.70 │
+│                     Time to Second Token (ms) │     62.88 │     0.00 │  1,867.07 │    604.22 │     75.17 │     28.46 │    134.29 │
+│               Time to First Output Token (ms) │  4,987.19 │   356.06 │ 38,854.91 │ 31,966.11 │ 17,187.52 │  1,329.97 │  7,503.36 │
+│                          Request Latency (ms) │ 10,793.22 │   934.78 │ 52,220.37 │ 40,891.13 │ 27,950.81 │  5,243.05 │ 10,904.72 │
+│                      Inter Token Latency (ms) │     23.30 │    11.93 │     58.21 │     46.99 │     30.81 │     22.12 │      6.31 ��
+│              Output Token Throughput Per User │     45.51 │    17.18 │     83.82 │     71.13 │     60.38 │     45.20 │     10.33 │
+│                             (tokens/sec/user) │           │          │           │           │           │           │           │
+│ E2E Output Token Throughput (tokens/sec/user) │     38.67 │    11.91 │     59.10 │     54.16 │     47.42 │     40.46 │      8.38 │
+│               Output Sequence Length (tokens) │    444.81 │    37.00 │  2,260.00 │  1,814.20 │  1,178.50 │    191.00 │    480.39 │
+│                Input Sequence Length (tokens) │ 24,722.81 │ 8,100.00 │ 64,531.00 │ 57,856.37 │ 42,932.30 │ 22,658.00 │ 12,184.92 │
+│          Output Token Throughput (tokens/sec) │    505.42 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│           Input Token Throughput (tokens/sec) │ 28,091.61 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│             Request Throughput (requests/sec) │      1.14 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│                      Request Count (requests) │  1,040.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+└───────────────────────────────────────────────┴───────────┴──────────┴───────────┴───────────┴───────────┴───────────┴───────────┘
+
+
+No GPU telemetry data collected during the benchmarking run.
+
+╭──────────────────────────────── Output Sequence Length Mismatch Warning ────────────────────────────────╮
+│  895 of 1,040 requests (86.1%) have output length differing from requested by more than the threshold.  │
+│  Threshold (tokens): min(requested x 5%, 50)                                                            │
+│  Average mismatch: -65.7%                                                                               │
+│                                                                                                         │
+│  Why: Server hit EOS token before reaching requested output length.                                     │
+│                                                                                                         │
+│  Fix Options:                                                                                           │
+│    - --extra-inputs ignore_eos:true - Generate until max_tokens (vLLM, TensorRT-LLM)                    │
+│    - --extra-inputs min_tokens:<N> - Set minimum output length (vLLM, TensorRT-LLM, SGLang)             │
+│    - --use-server-token-count - Use server-reported token counts if tokenizer mismatch suspected        │
+│                                                                                                         │
+│  Diagnostics:                                                                                           │
+│    - Review profile_export.jsonl -> osl_mismatch_diff_pct for per-request values                        │
+│    - Adjust: AIPERF_METRICS_OSL_MISMATCH_PCT_THRESHOLD=5                                                │
+│    - Adjust: AIPERF_METRICS_OSL_MISMATCH_MAX_TOKEN_THRESHOLD=50                                         │
+╰─────────────────────────────────────────────────────────────────────────────────────────────────────────╯
+
+CLI Command: aiperf profile --model 'glm-5.2' --url 'http://127.0.0.1:8391' --api-key '<redacted>' --endpoint-type 'chat' --streaming --tokenizer 'builtin' --custom-dataset-type 
+'dag_jsonl' --input-file 'data/capacity/corpus.jsonl' --extra-inputs '{"cache_salt": "166ddc184ab466ba8f8b9f654d3f6dde"}' --concurrency 16 --warmup-duration 120 --benchmark-duration 
+900 --benchmark-grace-period 300 --request-count 1000000 --random-seed 0 --use-server-token-count --artifact-dir 'results/capacity/sweep-hammer-jul20/N0016' --export-level 'raw' 
+--export-http-trace --slice-duration 60 --server-metrics-formats 'json' 'csv' 'jsonl' --ui 'simple'
+Benchmark Duration: 915.26 sec
+CSV Export: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0016/profile_export_aiperf.csv
+JSON Export: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0016/profile_export_aiperf.json
+Raw Records: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0016/profile_export_raw.jsonl
+Server Metrics CSV Export: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0016/server_metrics_export.csv
+Server Metrics JSON Export: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0016/server_metrics_export.json
+Timeslice CSV Export: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0016/profile_export_aiperf_timeslices.csv
+Timeslice JSON Export: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0016/profile_export_aiperf_timeslices.json
+Log File: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0016/logs/aiperf.log
+
+
+Warmup:   0% |          | [00:00<?]
+Warmup:   0% |          | [00:00<?]
+                                   
+
+Warmup:   2% |▏         | [00:01<01:57]
+Warmup:   2% |▏         | [00:01<01:57]
+Warmup:   3% |▎         | [00:03<01:55]
+Warmup:   3% |▎         | [00:03<01:55]
+Warmup:   5% |▌         | [00:05<01:53]
+Warmup:   5% |▌         | [00:05<01:53]
+Warmup:   7% |▋         | [00:07<01:51]
+Warmup:   7% |▋         | [00:07<01:51]
+Warmup:   8% |▊         | [00:09<01:49]
+Warmup:   8% |▊         | [00:09<01:49]
+Warmup:  10% |█         | [00:11<01:47]
+Warmup:  10% |█         | [00:11<01:47]
+Warmup:  12% |█▏        | [00:13<01:45]
+Warmup:  12% |█▏        | [00:13<01:45]
+Warmup:  13% |█▎        | [00:15<01:43]
+Warmup:  13% |█▎        | [00:15<01:43]
+Warmup:  15% |█▌        | [00:17<01:41]
+Warmup:  15% |█▌        | [00:17<01:41]
+Warmup:  17% |█▋        | [00:19<01:39]
+Warmup:  17% |█▋        | [00:19<01:39]
+Warmup:  18% |█▊        | [00:21<01:37]
+Warmup:  18% |█▊        | [00:21<01:37]
+Warmup:  20% |██        | [00:23<01:35]
+Warmup:  20% |██        | [00:23<01:35]
+Warmup:  22% |██▏       | [00:25<01:33]
+Warmup:  22% |██▏       | [00:25<01:33]
+Warmup:  23% |██▎       | [00:27<01:31]
+Warmup:  23% |██▎       | [00:27<01:31]
+Warmup:  25% |██▌       | [00:29<01:29]
+Warmup:  25% |██▌       | [00:29<01:29]
+Warmup:  27% |██▋       | [00:31<01:27]
+Warmup:  27% |██▋       | [00:31<01:27]
+Warmup:  28% |██▊       | [00:33<01:25]
+Warmup:  28% |██▊       | [00:33<01:25]
+Warmup:  30% |███       | [00:35<01:23]
+Warmup:  30% |███       | [00:35<01:23]
+Warmup:  32% |███▏      | [00:37<01:21]
+Warmup:  32% |███▏      | [00:37<01:21]
+Warmup:  33% |███▎      | [00:39<01:19]
+Warmup:  33% |███▎      | [00:39<01:19]
+Warmup:  35% |███▌      | [00:41<01:17]
+Warmup:  35% |███▌      | [00:41<01:17]
+Warmup:  37% |███▋      | [00:43<01:15]
+Warmup:  37% |███▋      | [00:43<01:15]
+Warmup:  38% |███▊      | [00:45<01:13]
+Warmup:  38% |███▊      | [00:45<01:13]
+Warmup:  40% |████      | [00:47<01:11]
+Warmup:  40% |████      | [00:47<01:11]
+Warmup:  42% |████▏     | [00:49<01:09]
+Warmup:  42% |████▏     | [00:49<01:09]
+Warmup:  43% |████▎     | [00:51<01:07]
+Warmup:  43% |████▎     | [00:51<01:07]
+Warmup:  45% |████▌     | [00:53<01:05]
+Warmup:  45% |████▌     | [00:53<01:05]
+Warmup:  47% |████▋     | [00:55<01:03]
+Warmup:  47% |████▋     | [00:55<01:03]
+Warmup:  48% |████▊     | [00:57<01:01]
+Warmup:  48% |████▊     | [00:57<01:01]
+Warmup:  50% |█████     | [00:59<00:59]
+Warmup:  50% |█████     | [00:59<00:59]
+Warmup:  52% |█████▏    | [01:01<00:57]
+Warmup:  52% |█████▏    | [01:01<00:57]
+Warmup:  53% |█████▎    | [01:03<00:55]
+Warmup:  53% |█████▎    | [01:03<00:55]
+Warmup:  55% |█████▌    | [01:05<00:53]
+Warmup:  55% |█████▌    | [01:05<00:53]
+Warmup:  57% |█████▋    | [01:07<00:51]
+Warmup:  57% |█████▋    | [01:07<00:51]
+Warmup:  58% |█████▊    | [01:09<00:49]
+Warmup:  58% |█████▊    | [01:09<00:49]
+Warmup:  60% |██████    | [01:11<00:47]
+Warmup:  60% |██████    | [01:11<00:47]
+Warmup:  62% |██████▏   | [01:13<00:45]
+Warmup:  62% |██████▏   | [01:13<00:45]
+Warmup:  63% |██████▎   | [01:15<00:43]
+Warmup:  63% |██████▎   | [01:15<00:43]
+Warmup:  65% |██████▌   | [01:17<00:41]
+Warmup:  65% |██████▌   | [01:17<00:41]
+Warmup:  67% |██████▋   | [01:19<00:39]
+Warmup:  67% |██████▋   | [01:19<00:39]
+Warmup:  68% |██████▊   | [01:21<00:37]
+Warmup:  68% |██████▊   | [01:21<00:37]
+Warmup:  70% |███████   | [01:23<00:35]
+Warmup:  70% |███████   | [01:23<00:35]
+Warmup:  72% |███████▏  | [01:25<00:33]
+Warmup:  72% |███████▏  | [01:25<00:33]
+Warmup:  73% |███████▎  | [01:27<00:31]
+Warmup:  73% |███████▎  | [01:27<00:31]
+Warmup:  75% |███████▌  | [01:29<00:29]
+Warmup:  75% |███████▌  | [01:29<00:29]
+Warmup:  77% |███████▋  | [01:31<00:27]
+Warmup:  77% |███████▋  | [01:31<00:27]
+Warmup:  78% |███████▊  | [01:33<00:25]
+Warmup:  78% |███████▊  | [01:33<00:25]
+Warmup:  80% |████████  | [01:35<00:23]
+Warmup:  80% |████████  | [01:35<00:23]
+Warmup:  82% |████████▏ | [01:37<00:21]
+Warmup:  82% |████████▏ | [01:37<00:21]
+Warmup:  83% |████████▎ | [01:39<00:19]
+Warmup:  83% |████████▎ | [01:39<00:19]
+Warmup:  85% |████████▌ | [01:41<00:17]
+Warmup:  85% |████████▌ | [01:41<00:17]
+Warmup:  87% |████████▋ | [01:43<00:15]
+Warmup:  87% |████████▋ | [01:43<00:15]
+Warmup:  88% |████████▊ | [01:45<00:13]
+Warmup:  88% |████████▊ | [01:45<00:13]
+Warmup:  90% |█████████ | [01:47<00:11]
+Warmup:  90% |█████████ | [01:47<00:11]
+Warmup:  92% |█████████▏| [01:49<00:09]
+Warmup:  92% |█████████▏| [01:49<00:09]
+Warmup:  93% |█████████▎| [01:51<00:07]
+Warmup:  93% |█████████▎| [01:51<00:07]
+Warmup:  95% |█████████▌| [01:53<00:05]
+Warmup:  95% |█████████▌| [01:53<00:05]
+Warmup:  97% |█████████▋| [01:55<00:03]
+Warmup:  97% |█████████▋| [01:55<00:03]
+Warmup:  98% |█████████▊| [01:57<00:01]
+Warmup:  98% |█████████▊| [01:57<00:01]
+Warmup Grace Period: 0/241 |          |   0% [00:00<?]
+Warmup Grace Period: 0/241 |          |   0% [00:00<?]
+                                                      
+
+Warmup Grace Period: 231/241 |█████████▌|  96% [00:00<00:00]
+Warmup Grace Period: 237/241 |█████████▊|  98% [00:02<00:00]
+Warmup Grace Period: 237/241 |█████████▊|  98% [00:02<00:00]
+Warmup Grace Period: 240/241 |█████████▉| 100% [00:12<00:00]
+Processing Records: 0/241 |          |   0% [00:00<?]
+Profiling: 0/1,000,000 |          |   0% [00:00<?]
+Profiling Grace Period: 0/1,040 |          |   0% [00:00<?]
+Profiling Grace Period: 1,027/1,040 |█████████▉|  99% [00:00<00:00]
+Profiling Grace Period: 1,038/1,040 |█████████▉| 100% [00:12<00:00]
+Profiling Grace Period: 1,038/1,040 |█████████▉| 100% [00:12<00:00]
+Processing Records: 0/1,040 |          |   0% [00:00<?]
+Processing Records: 1,040/1,040 |██████████| 100% [00:00<00:00]

experiments/kv-cache-fp8/capacity/hammer/N0016/manifest.json

@@ -0,0 +1,111 @@
+{
+  "concurrency": 16,
+  "mode": "hammer",
+  "cache_salt": "166ddc184ab466ba8f8b9f654d3f6dde",
+  "command": [
+    "aiperf",
+    "profile",
+    "--model",
+    "glm-5.2",
+    "--url",
+    "http://127.0.0.1:8391",
+    "--api-key",
+    "<key>",
+    "--endpoint-type",
+    "chat",
+    "--streaming",
+    "--tokenizer",
+    "builtin",
+    "--custom-dataset-type",
+    "dag_jsonl",
+    "--input-file",
+    "data/capacity/corpus.jsonl",
+    "--extra-inputs",
+    "{\"cache_salt\": \"166ddc184ab466ba8f8b9f654d3f6dde\"}",
+    "--concurrency",
+    "16",
+    "--warmup-duration",
+    "120",
+    "--benchmark-duration",
+    "900",
+    "--benchmark-grace-period",
+    "300",
+    "--request-count",
+    "1000000",
+    "--random-seed",
+    "0",
+    "--use-server-token-count",
+    "--artifact-dir",
+    "results/capacity/sweep-hammer-jul20/N0016",
+    "--export-level",
+    "raw",
+    "--export-http-trace",
+    "--slice-duration",
+    "60",
+    "--server-metrics-formats",
+    "json",
+    "csv",
+    "jsonl",
+    "--ui",
+    "simple"
+  ],
+  "dataset_path": "data/capacity/corpus.jsonl",
+  "signature": {
+    "dataset_sha256": "0327d658a267715aebdab45b67b6e0d324616a0faeec8a75d09b8defcd0b2ac2",
+    "url": "http://127.0.0.1:8391",
+    "model": "glm-5.2",
+    "duration": 900,
+    "warmup": 120,
+    "grace": 300,
+    "seed": 0,
+    "request_ceiling": 1000000,
+    "python": "3.12.3",
+    "aiperf": {
+      "version": "0.11.0",
+      "rev": "73d0524f98f30199ca659a8ab586454a823c887d"
+    },
+    "server": {
+      "version": "0.5.15.post1",
+      "model_path": "/home/jl_fs/models/GLM-5.2-FP8",
+      "served_model_name": "glm-5.2",
+      "tp_size": 8,
+      "dp_size": 8,
+      "enable_cache_report": true,
+      "speculative_algorithm": "EAGLE",
+      "max_running_requests": 48,
+      "context_length": 262144,
+      "kv_cache_dtype": "fp8_e4m3",
+      "mem_fraction_static": 0.872,
+      "scheduler_pids": [
+        82149,
+        82150,
+        82151,
+        82152,
+        82153,
+        82154,
+        82155,
+        82156
+      ],
+      "max_total_num_tokens": 234176,
+      "max_model_len": 262144,
+      "model_id": "glm-5.2"
+    },
+    "evaluator_sha256": "3e48dfede43cbf47244f560547eda95d29c671d4fa23b203a5d220a14edf1c80",
+    "verifier_sha256": "6f84bee84368a72ea1445d0961185a0d5d2fd6ba8f8ec4c3fefac619b46a32aa",
+    "runner_sha256": "aa3e1a0f66bcfac6663b5d2e59a37cae9d7264dda6b4c13130a3edf403e77104"
+  },
+  "started_utc": "2026-07-20T14:32:33Z",
+  "loadavg_start": [
+    1.84912109375,
+    1.75439453125,
+    1.40234375
+  ],
+  "status": "ok",
+  "elapsed_s": 1062.1,
+  "loadavg_end": [
+    1.59619140625,
+    1.9072265625,
+    1.7919921875
+  ],
+  "returncode": 0
+}

experiments/kv-cache-fp8/capacity/hammer/N0016/profile_export_aiperf.csv

@@ -0,0 +1,68 @@
+Metric,avg,min,max,sum,p1,p5,p10,p25,p50,p75,p90,p95,p99,std
+Active Decode Throughput (tokens/sec),505.97,0.00,136616.40,,,,,,452.53,,821.68,899.05,2087.53,980.46
+Active Decode Throughput Per User (tokens/sec/user),42.74,0.00,13960.98,,,,,,38.49,,65.02,68.23,174.18,85.28
+Active Prefill Throughput (tokens/sec),67997.99,2294.05,211712.97,,,,,,62853.71,,114519.99,134481.28,163691.60,34187.76
+Active Prefill Throughput Per User (tokens/sec/user),48627.32,2294.05,141468.65,,,,,,44712.08,,85023.03,99464.70,118005.17,25641.26
+Active Total Throughput (tokens/sec),28596.57,57.17,212194.50,,,,,,606.76,,90235.30,111132.48,149438.85,40016.60
+Credit-to-Start Latency (ms),1.54,0.83,14.53,1600.76,0.90,1.04,1.10,1.20,1.35,1.55,1.89,2.74,5.61,0.85
+E2E Output Token Throughput (tokens/sec/user),38.67,11.91,59.10,40215.33,14.27,20.51,27.22,34.69,40.46,44.41,47.42,48.89,54.16,8.38
+Effective Concurrency,12.29,0.00,16.00,,,,,,12.00,,15.00,16.00,16.00,1.82
+Effective Decode Concurrency,11.58,0.00,16.00,,,,,,12.00,,14.00,15.00,16.00,2.02
+Effective Decode Throughput (tokens/sec),504.29,0.00,136616.40,,,,,,446.40,,821.38,898.34,2081.03,979.26
+Effective Decode Throughput Per User (tokens/sec/user),42.60,0.00,13960.98,,,,,,38.19,,65.01,68.19,173.86,85.17
+Effective Latency (CO-aware) (ms),10817.06,935.75,52221.87,11249744.62,1160.33,1583.70,1872.93,2778.06,5275.09,17247.15,27953.63,33244.65,40892.60,10897.93
+Effective Prefill Concurrency,0.71,0.00,16.00,,,,,,0.00,,2.00,3.00,5.00,1.38
+Effective Prefill Throughput (tokens/sec),28091.61,0.00,211712.97,,,,,,0.00,,89733.77,110602.03,148988.77,40048.66
+Effective Prefill Throughput Per User (tokens/sec/user),20089.12,0.00,141468.65,,,,,,0.00,,65049.87,80961.87,109360.34,29067.61
+Effective Total Throughput (tokens/sec),28595.90,0.00,212194.50,,,,,,606.76,,90235.30,111132.48,149438.85,40016.37
+HTTP Blocked (ms),0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00
+HTTP Chunks Received,238.75,19.00,1195.00,248305.00,21.39,27.00,33.00,49.00,104.50,393.00,639.00,767.20,961.93,257.06
+HTTP Chunks Sent,1.00,1.00,1.00,1040.00,1.00,1.00,1.00,1.00,1.00,1.00,1.00,1.00,1.00,0.00
+HTTP Connecting (ms),0.08,0.00,8.32,79.17,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,1.59,0.42
+HTTP Connection Overhead (ms),0.08,0.00,8.32,79.17,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,1.59,0.42
+HTTP Connection Reused (ratio),0.95,0.00,1.00,989.00,0.00,1.00,1.00,1.00,1.00,1.00,1.00,1.00,1.00,0.22
+HTTP Data Received (KB),58.57,5.31,290.32,60914.20,5.82,7.50,8.84,12.83,26.23,95.88,153.00,185.29,233.36,61.81
+HTTP Data Sent (KB),102.13,36.83,258.05,106216.69,37.59,40.62,44.77,63.24,94.45,130.84,174.12,196.87,239.31,49.10
+HTTP DNS Lookup (ms),0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00
+HTTP Duration (excl. conn) (ms),10815.39,934.69,52220.24,11248002.90,1158.35,1582.20,1871.68,2776.51,5273.44,17245.56,27950.50,33243.41,40890.98,10897.95
+HTTP Receiving (ms),10207.19,556.01,51774.45,10615482.20,691.86,1053.83,1306.66,2201.89,4530.18,16586.30,27173.62,32563.81,40360.42,10908.79
+HTTP Sending (ms),0.35,0.14,9.16,363.53,0.19,0.20,0.20,0.22,0.25,0.30,0.39,0.64,2.23,0.47
+HTTP Total Time (ms),10815.46,934.69,52220.24,11248082.06,1158.35,1582.20,1871.68,2776.51,5273.44,17245.56,27950.66,33243.41,40890.98,10897.93
+HTTP Waiting (TTFB) (ms),607.84,325.62,3592.01,632157.17,331.86,352.32,365.48,395.63,525.31,683.72,850.63,946.66,3580.41,413.13
+Input Sequence Length (tokens),24722.81,8100.00,64531.00,25711718.00,8265.39,9116.45,10190.50,15107.75,22658.00,32484.25,42932.30,47748.05,57856.37,12184.92
+Inter Chunk Latency (ms),42.68,0.00,2462.17,10571521.56,0.30,0.96,1.43,27.89,29.67,31.11,60.27,94.72,337.13,73.51
+Inter Token Latency (ms),23.30,11.93,58.21,24227.01,14.06,15.34,16.56,19.98,22.12,25.08,30.81,35.44,46.99,6.31
+OSL Mismatch Diff % (%),-65.66,-97.83,0.00,-68291.55,-96.57,-95.48,-94.66,-92.00,-84.13,-42.05,0.00,0.00,0.00,34.41
+Output Sequence Length (tokens),444.81,37.00,2260.00,462605.00,41.00,52.95,63.00,91.00,191.00,719.25,1178.50,1441.40,1814.20,480.39
+Output Token Count (tokens),444.81,37.00,2260.00,462605.00,41.00,52.95,63.00,91.00,191.00,719.25,1178.50,1441.40,1814.20,480.39
+Output Token Throughput Per User (tokens/sec/user),45.51,17.18,83.82,47326.28,21.28,28.22,32.45,39.87,45.20,50.06,60.38,65.17,71.13,10.33
+Prefill Throughput Per User (tokens/sec/user),47005.63,2261.55,141468.65,48885850.10,2776.03,12061.97,15786.47,26469.88,42607.35,63832.66,84377.65,98446.66,117078.30,26442.50
+Request Latency (ms),10793.22,934.78,52220.37,11224953.19,1141.98,1549.86,1860.21,2754.53,5243.05,17222.39,27950.81,33243.53,40891.13,10904.72
+Time to First Output Token (ms),4987.19,356.06,38854.91,4638084.23,380.32,403.54,426.51,638.12,1329.97,5279.21,17187.52,23047.72,31966.11,7503.36
+Time to First Token (ms),628.30,325.98,3592.81,653431.64,332.26,355.81,373.13,405.23,539.64,706.24,900.76,1045.85,3582.36,422.70
+Time to Second Token (ms),62.88,0.00,1867.07,65400.28,0.00,0.13,0.88,24.24,28.46,34.33,75.17,321.16,604.22,134.29
+Tokens In Flight (tokens),356248.93,0.00,505096.39,,,,,,362879.28,,453099.60,466969.64,492375.69,77305.99
+Usage Completion Tokens (tokens),444.81,37.00,2260.00,462605.00,41.00,52.95,63.00,91.00,191.00,719.25,1178.50,1441.40,1814.20,480.39
+Usage Prompt Cache Read Tokens (tokens),23983.75,7744.00,64384.00,24751232.00,7744.00,8512.00,9792.00,14208.00,21696.00,30928.00,42163.20,47126.40,56792.32,12131.37
+Usage Prompt Tokens (tokens),24722.81,8100.00,64531.00,25711718.00,8265.39,9116.45,10190.50,15107.75,22658.00,32484.25,42932.30,47748.05,57856.37,12184.92
+Usage Total Tokens (tokens),25167.62,8187.00,65628.00,26174323.00,8334.24,9195.70,10307.10,15575.25,23067.50,32788.25,43592.30,48338.90,58307.57,12363.77
+
+Metric,Value
+Benchmark Duration (sec),915.26
+Input Token Throughput (tokens/sec),28091.61
+OSL Mismatch Count,895.00
+Output Token Throughput (tokens/sec),505.42
+Overall Usage Prompt Cache Read % (%),96.26
+Request Count,1040.00
+Request Throughput (requests/sec),1.14
+Total Input Sequence Length (tokens),25711718.00
+Total Output Sequence Length (tokens),462605.00
+Total Output Tokens (tokens),462605.00
+Total Token Throughput (tokens/sec),28597.03
+Total Usage Completion Tokens (tokens),462605.00
+Total Usage Prompt Cache Read Tokens (tokens),24751232.00
+Total Usage Prompt Tokens (tokens),25711718.00
+Total Usage Total Tokens (tokens),26174323.00
+
+
+Endpoint,GPU_Index,GPU_Name,GPU_UUID,Metric,avg,min,max,sum,p1,p5,p10,p25,p50,p75,p90,p95,p99,std

experiments/kv-cache-fp8/capacity/hammer/N0016/profile_export_aiperf.json

@@ -0,0 +1,1726 @@
+{
+  "schema_version": "1.4",
+  "aiperf_version": "0.11.0",
+  "benchmark_id": "56f23bb8ed81",
+  "request_throughput": {
+    "unit": "requests/sec",
+    "avg": 1.136262983345431
+  },
+  "request_latency": {
+    "unit": "ms",
+    "avg": 10793.224225109614,
+    "p1": 1141.98407015,
+    "p5": 1549.8619056,
+    "p10": 1860.2086708,
+    "p25": 2754.528586,
+    "p50": 5243.0522525,
+    "p75": 17222.3855325,
+    "p90": 27950.810698600002,
+    "p95": 33243.52511225,
+    "p99": 40891.125925589986,
+    "min": 934.782882,
+    "max": 52220.365977999994,
+    "std": 10904.716164613603,
+    "count": 1040,
+    "sum": 11224953.194114
+  },
+  "request_count": {
+    "unit": "requests",
+    "avg": 1040.0
+  },
+  "time_to_first_token": {
+    "unit": "ms",
+    "avg": 628.2996517653846,
+    "p1": 332.2576325,
+    "p5": 355.8116639,
+    "p10": 373.13471489999995,
+    "p25": 405.23078899999996,
+    "p50": 539.6389075,
+    "p75": 706.24387,
+    "p90": 900.7553698,
+    "p95": 1045.8483894499998,
+    "p99": 3582.36235321,
+    "min": 325.98450099999997,
+    "max": 3592.8137469999997,
+    "std": 422.6998401990308,
+    "count": 1040,
+    "sum": 653431.637836
+  },
+  "time_to_second_token": {
+    "unit": "ms",
+    "avg": 62.884887448076924,
+    "p1": 0.0,
+    "p5": 0.13307245,
+    "p10": 0.8833489999999999,
+    "p25": 24.243421249999997,
+    "p50": 28.464256,
+    "p75": 34.32528875,
+    "p90": 75.16749820000013,
+    "p95": 321.16060939999994,
+    "p99": 604.2179416299999,
+    "min": 0.0,
+    "max": 1867.067664,
+    "std": 134.28805226997417,
+    "count": 1040,
+    "sum": 65400.282946
+  },
+  "inter_token_latency": {
+    "unit": "ms",
+    "avg": 23.29519948009281,
+    "p1": 14.059164732378786,
+    "p5": 15.34363918102564,
+    "p10": 16.56125794742324,
+    "p25": 19.97533661869472,
+    "p50": 22.123084328985506,
+    "p75": 25.0803984457504,
+    "p90": 30.812709618500797,
+    "p95": 35.44146257638888,
+    "p99": 46.99310185662844,
+    "min": 11.930356964285714,
+    "max": 58.21243352380952,
+    "std": 6.314833798961468,
+    "count": 1040,
+    "sum": 24227.007459296517
+  },
+  "output_token_throughput": {
+    "unit": "tokens/sec",
+    "avg": 505.42397827933956
+  },
+  "output_token_throughput_per_user": {
+    "unit": "tokens/sec/user",
+    "avg": 45.50604109107585,
+    "p1": 21.283160786034745,
+    "p5": 28.21556302750276,
+    "p10": 32.45414675960236,
+    "p25": 39.87177504566211,
+    "p50": 45.20166539525998,
+    "p75": 50.061735083011484,
+    "p90": 60.38193460198154,
+    "p95": 65.17359179307451,
+    "p99": 71.12798712192578,
+    "min": 17.178460673543032,
+    "max": 83.81978871156696,
+    "std": 10.332972753017701,
+    "count": 1040,
+    "sum": 47326.28273471889
+  },
+  "output_sequence_length": {
+    "unit": "tokens",
+    "avg": 444.8125,
+    "p1": 41.0,
+    "p5": 52.95,
+    "p10": 63.0,
+    "p25": 91.0,
+    "p50": 191.0,
+    "p75": 719.25,
+    "p90": 1178.5,
+    "p95": 1441.3999999999996,
+    "p99": 1814.199999999998,
+    "min": 37.0,
+    "max": 2260.0,
+    "std": 480.39091297278605,
+    "count": 1040,
+    "sum": 462605.0
+  },
+  "input_sequence_length": {
+    "unit": "tokens",
+    "avg": 24722.80576923077,
+    "p1": 8265.39,
+    "p5": 9116.45,
+    "p10": 10190.5,
+    "p25": 15107.75,
+    "p50": 22658.0,
+    "p75": 32484.25,
+    "p90": 42932.3,
+    "p95": 47748.049999999996,
+    "p99": 57856.36999999999,
+    "min": 8100.0,
+    "max": 64531.0,
+    "std": 12184.919747090813,
+    "count": 1040,
+    "sum": 25711718.0
+  },
+  "output_token_count": {
+    "unit": "tokens",
+    "avg": 444.8125,
+    "p1": 41.0,
+    "p5": 52.95,
+    "p10": 63.0,
+    "p25": 91.0,
+    "p50": 191.0,
+    "p75": 719.25,
+    "p90": 1178.5,
+    "p95": 1441.3999999999996,
+    "p99": 1814.199999999998,
+    "min": 37.0,
+    "max": 2260.0,
+    "std": 480.39091297278605,
+    "count": 1040,
+    "sum": 462605.0
+  },
+  "inter_chunk_latency": {
+    "unit": "ms",
+    "avg": 42.68200449884326,
+    "p1": 0.301917,
+    "p5": 0.9618439999999999,
+    "p10": 1.433044,
+    "p25": 27.889001,
+    "p50": 29.672895999999998,
+    "p75": 31.105608,
+    "p90": 60.274801999999994,
+    "p95": 94.72175999999999,
+    "p99": 337.1256566,
+    "min": 0.0,
+    "max": 2462.165148,
+    "std": 73.51356032334849,
+    "count": 247681,
+    "sum": 10571521.556278
+  },
+  "total_output_tokens": {
+    "unit": "tokens",
+    "avg": 462605.0
+  },
+  "benchmark_duration": {
+    "unit": "sec",
+    "avg": 915.2596134400001
+  },
+  "total_isl": {
+    "unit": "tokens",
+    "avg": 25711718.0
+  },
+  "total_osl": {
+    "unit": "tokens",
+    "avg": 462605.0
+  },
+  "telemetry_data": {
+    "summary": {
+      "endpoints_configured": [],
+      "endpoints_successful": [],
+      "start_time": "2026-07-20T14:34:53.711358",
+      "end_time": "2026-07-20T14:50:11.899283"
+    },
+    "endpoints": {},
+    "error_summary": []
+  },
+  "input_config": {
+    "models": {
+      "items": [
+        {
+          "name": "glm-5.2"
+        }
+      ]
+    },
+    "endpoint": {
+      "urls": [
+        "http://127.0.0.1:8391"
+      ],
+      "type": "chat",
+      "api_key": "<redacted>",
+      "streaming": true,
+      "use_server_token_count": true,
+      "extra": {
+        "cache_salt": "166ddc184ab466ba8f8b9f654d3f6dde"
+      }
+    },
+    "datasets": [
+      {
+        "name": "main",
+        "type": "file",
+        "path": "data/capacity/corpus.jsonl",
+        "format": "dag_jsonl",
+        "entries": 1000000,
+        "random_seed": 0
+      }
+    ],
+    "phases": [
+      {
+        "name": "warmup",
+        "type": "concurrency",
+        "exclude_from_results": true,
+        "duration": 120.0,
+        "concurrency": 16
+      },
+      {
+        "name": "profiling",
+        "type": "concurrency",
+        "requests": 1000000,
+        "duration": 900.0,
+        "concurrency": 16,
+        "grace_period": 300.0
+      }
+    ],
+    "artifacts": {
+      "dir": "/home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0016",
+      "records": [
+        "jsonl"
+      ],
+      "raw": true,
+      "trace": true,
+      "slice_duration": 60.0,
+      "auto_plot": false,
+      "plot_required": false
+    },
+    "tokenizer": {
+      "name": "builtin"
+    },
+    "gpu_telemetry": {
+      "enabled": true
+    },
+    "server_metrics": {
+      "enabled": true,
+      "urls": [],
+      "formats": [
+        "json",
+        "csv",
+        "jsonl"
+      ]
+    },
+    "network_latency": {
+      "enabled": false
+    },
+    "otel": {},
+    "mlflow": {},
+    "wandb": {},
+    "runtime": {
+      "ui": "simple"
+    }
+  },
+  "run_info": {
+    "benchmark_id": "56f23bb8ed81",
+    "random_seed": 0,
+    "trial": 0,
+    "cli_command": "aiperf profile --model 'glm-5.2' --url 'http://127.0.0.1:8391' --api-key '<redacted>' --endpoint-type 'chat' --streaming --tokenizer 'builtin' --custom-dataset-type 'dag_jsonl' --input-file 'data/capacity/corpus.jsonl' --extra-inputs '{\"cache_salt\": \"166ddc184ab466ba8f8b9f654d3f6dde\"}' --concurrency 16 --warmup-duration 120 --benchmark-duration 900 --benchmark-grace-period 300 --request-count 1000000 --random-seed 0 --use-server-token-count --artifact-dir 'results/capacity/sweep-hammer-jul20/N0016' --export-level 'raw' --export-http-trace --slice-duration 60 --server-metrics-formats 'json' 'csv' 'jsonl' --ui 'simple'"
+  },
+  "was_cancelled": false,
+  "error_summary": [],
+  "start_time": "2026-07-20T14:34:53.711358",
+  "end_time": "2026-07-20T14:50:08.992428",
+  "warmup_metrics": {
+    "time_to_first_output_token": {
+      "unit": "ms",
+      "avg": 3312.3207381680672,
+      "p1": 363.95109272,
+      "p5": 385.06929255,
+      "p10": 405.0470608,
+      "p25": 670.14210125,
+      "p50": 1274.3365585,
+      "p75": 3123.39454575,
+      "p90": 7794.345230200005,
+      "p95": 14010.37950634998,
+      "p99": 27606.561260259958,
+      "min": 356.62920199999996,
+      "max": 55553.370066999996,
+      "std": 5858.7041688763,
+      "count": 238,
+      "sum": 788332.3356839999
+    },
+    "http_req_sending": {
+      "unit": "ms",
+      "avg": 0.5264382655601659,
+      "p1": 0.18954079999999998,
+      "p5": 0.196619,
+      "p10": 0.20096599999999998,
+      "p25": 0.215077,
+      "p50": 0.245805,
+      "p75": 0.29881399999999997,
+      "p90": 0.44862399999999997,
+      "p95": 2.470495,
+      "p99": 4.9418751999999975,
+      "min": 0.185824,
+      "max": 7.7652719999999995,
+      "std": 1.002811237421406,
+      "count": 241,
+      "sum": 126.87162199999999
+    },
+    "http_req_duration": {
+      "unit": "ms",
+      "avg": 6449.152367053942,
+      "p1": 1154.9478405999998,
+      "p5": 1614.2782129999998,
+      "p10": 1955.136744,
+      "p25": 2690.6977779999997,
+      "p50": 4073.557766,
+      "p75": 6957.057836999999,
+      "p90": 12383.135033999999,
+      "p95": 19791.78773,
+      "p99": 44645.168710799895,
+      "min": 1033.893081,
+      "max": 59098.172647,
+      "std": 7675.135390957702,
+      "count": 241,
+      "sum": 1554245.72046
+    },
+    "usage_total_tokens": {
+      "unit": "tokens",
+      "avg": 16750.651452282156,
+      "p1": 8478.0,
+      "p5": 8836.0,
+      "p10": 10138.0,
+      "p25": 11976.0,
+      "p50": 17148.0,
+      "p75": 20956.0,
+      "p90": 23505.0,
+      "p95": 25656.0,
+      "p99": 27447.4,
+      "min": 8390.0,
+      "max": 27612.0,
+      "std": 5244.638198867555,
+      "count": 241,
+      "sum": 4036907.0
+    },
+    "time_to_first_token": {
+      "unit": "ms",
+      "avg": 857.1175836390041,
+      "p1": 336.7578438,
+      "p5": 351.823126,
+      "p10": 368.719832,
+      "p25": 413.83741799999996,
+      "p50": 638.762841,
+      "p75": 884.8281999999999,
+      "p90": 1354.5215759999999,
+      "p95": 3455.353526,
+      "p99": 4041.0483504,
+      "min": 325.615825,
+      "max": 4311.725353,
+      "std": 800.4737838515733,
+      "count": 241,
+      "sum": 206565.337657
+    },
+    "http_req_chunks_received": {
+      "unit": "count",
+      "avg": 104.18257261410788,
+      "p1": 19.2,
+      "p5": 24.0,
+      "p10": 28.0,
+      "p25": 36.0,
+      "p50": 46.0,
+      "p75": 90.0,
+      "p90": 231.0,
+      "p95": 423.0,
+      "p99": 794.199999999999,
+      "min": 15.0,
+      "max": 1438.0,
+      "std": 161.4887682438018,
+      "count": 241,
+      "sum": 25108.0
+    },
+    "http_req_connecting": {
+      "unit": "ms",
+      "avg": 0.2459006846473029,
+      "p1": 0.0,
+      "p5": 0.0,
+      "p10": 0.0,
+      "p25": 0.0,
+      "p50": 0.0,
+      "p75": 0.0,
+      "p90": 0.0,
+      "p95": 1.925215,
+      "p99": 4.238834599999996,
+      "min": 0.0,
+      "max": 7.167301999999999,
+      "std": 0.910956729743431,
+      "count": 241,
+      "sum": 59.262065
+    },
+    "http_req_chunks_sent": {
+      "unit": "count",
+      "avg": 1.0,
+      "p1": 1.0,
+      "p5": 1.0,
+      "p10": 1.0,
+      "p25": 1.0,
+      "p50": 1.0,
+      "p75": 1.0,
+      "p90": 1.0,
+      "p95": 1.0,
+      "p99": 1.0,
+      "min": 1.0,
+      "max": 1.0,
+      "std": 0.0,
+      "count": 241,
+      "sum": 241.0
+    },
+    "request_latency": {
+      "unit": "ms",
+      "avg": 6422.717482829876,
+      "p1": 1155.0942315999998,
+      "p5": 1560.282295,
+      "p10": 1902.84609,
+      "p25": 2650.730955,
+      "p50": 4073.688706,
+      "p75": 6957.186683,
+      "p90": 12322.890518999999,
+      "p95": 19762.904714,
+      "p99": 44633.07320979989,
+      "min": 1034.033584,
+      "max": 59098.328463,
+      "std": 7676.918071447369,
+      "count": 241,
+      "sum": 1547874.913362
+    },
+    "output_sequence_length": {
+      "unit": "tokens",
+      "avg": 195.3278008298755,
+      "p1": 36.8,
+      "p5": 46.0,
+      "p10": 57.0,
+      "p25": 67.0,
+      "p50": 88.0,
+      "p75": 164.0,
+      "p90": 457.0,
+      "p95": 750.0,
+      "p99": 1468.999999999998,
+      "min": 28.0,
+      "max": 2878.0,
+      "std": 308.63626622413256,
+      "count": 241,
+      "sum": 47074.0
+    },
+    "http_req_blocked": {
+      "unit": "ms",
+      "avg": 0.0,
+      "p1": 0.0,
+      "p5": 0.0,
+      "p10": 0.0,
+      "p25": 0.0,
+      "p50": 0.0,
+      "p75": 0.0,
+      "p90": 0.0,
+      "p95": 0.0,
+      "p99": 0.0,
+      "min": 0.0,
+      "max": 0.0,
+      "std": 0.0,
+      "count": 241,
+      "sum": 0.0
+    },
+    "http_req_dns_lookup": {
+      "unit": "ms",
+      "avg": 0.0,
+      "p1": 0.0,
+      "p5": 0.0,
+      "p10": 0.0,
+      "p25": 0.0,
+      "p50": 0.0,
+      "p75": 0.0,
+      "p90": 0.0,
+      "p95": 0.0,
+      "p99": 0.0,
+      "min": 0.0,
+      "max": 0.0,
+      "std": 0.0,
+      "count": 241,
+      "sum": 0.0
+    },
+    "http_req_receiving": {
+      "unit": "ms",
+      "avg": 5626.230182568464,
+      "p1": 701.5307297999999,
+      "p5": 1123.760787,
+      "p10": 1373.987821,
+      "p25": 2034.526365,
+      "p50": 3259.794828,
+      "p75": 5397.5569669999995,
+      "p90": 11631.309148,
+      "p95": 19187.784516,
+      "p99": 43537.881680599894,
+      "min": 445.953744,
+      "max": 58249.024164999995,
+      "std": 7643.19976410911,
+      "count": 241,
+      "sum": 1355921.473999
+    },
+    "http_req_waiting": {
+      "unit": "ms",
+      "avg": 822.395746219917,
+      "p1": 332.6865742,
+      "p5": 346.007613,
+      "p10": 357.003831,
+      "p25": 396.79574199999996,
+      "p50": 627.5535679999999,
+      "p75": 823.2372389999999,
+      "p90": 1191.0382479999998,
+      "p95": 3452.320644,
+      "p99": 4034.8570638,
+      "min": 325.26465099999996,
+      "max": 4306.694697,
+      "std": 796.7750434797204,
+      "count": 241,
+      "sum": 198197.374839
+    },
+    "http_req_data_received": {
+      "unit": "KB",
+      "avg": 26.084170772821576,
+      "p1": 5.1009765625,
+      "p5": 6.5625,
+      "p10": 7.7939453125,
+      "p25": 9.357421875,
+      "p50": 12.12890625,
+      "p75": 22.4501953125,
+      "p90": 58.4638671875,
+      "p95": 105.990234375,
+      "p99": 189.90039062499977,
+      "min": 4.3134765625,
+      "max": 354.3505859375,
+      "std": 39.15713173079086,
+      "count": 241,
+      "sum": 6286.28515625
+    },
+    "usage_prompt_tokens": {
+      "unit": "tokens",
+      "avg": 16555.323651452283,
+      "p1": 8405.8,
+      "p5": 8779.0,
+      "p10": 10072.0,
+      "p25": 11919.0,
+      "p50": 16625.0,
+      "p75": 20737.0,
+      "p90": 23349.0,
+      "p95": 25437.0,
+      "p99": 27405.2,
+      "min": 8312.0,
+      "max": 27526.0,
+      "std": 5197.023662558373,
+      "count": 241,
+      "sum": 3989833.0
+    },
+    "usage_completion_tokens": {
+      "unit": "tokens",
+      "avg": 195.3278008298755,
+      "p1": 36.8,
+      "p5": 46.0,
+      "p10": 57.0,
+      "p25": 67.0,
+      "p50": 88.0,
+      "p75": 164.0,
+      "p90": 457.0,
+      "p95": 750.0,
+      "p99": 1468.999999999998,
+      "min": 28.0,
+      "max": 2878.0,
+      "std": 308.63626622413256,
+      "count": 241,
+      "sum": 47074.0
+    },
+    "input_sequence_length": {
+      "unit": "tokens",
+      "avg": 16555.323651452283,
+      "p1": 8405.8,
+      "p5": 8779.0,
+      "p10": 10072.0,
+      "p25": 11919.0,
+      "p50": 16625.0,
+      "p75": 20737.0,
+      "p90": 23349.0,
+      "p95": 25437.0,
+      "p99": 27405.2,
+      "min": 8312.0,
+      "max": 27526.0,
+      "std": 5197.023662558373,
+      "count": 241,
+      "sum": 3989833.0
+    },
+    "time_to_second_token": {
+      "unit": "ms",
+      "avg": 102.24075110788381,
+      "p1": 0.0,
+      "p5": 0.298794,
+      "p10": 1.034012,
+      "p25": 25.60322,
+      "p50": 29.317065999999997,
+      "p75": 36.668600999999995,
+      "p90": 318.22306,
+      "p95": 593.610564,
+      "p99": 914.1068544,
+      "min": 0.0,
+      "max": 1536.806096,
+      "std": 229.04650007719096,
+      "count": 241,
+      "sum": 24640.021017
+    },
+    "output_token_count": {
+      "unit": "tokens",
+      "avg": 195.3278008298755,
+      "p1": 36.8,
+      "p5": 46.0,
+      "p10": 57.0,
+      "p25": 67.0,
+      "p50": 88.0,
+      "p75": 164.0,
+      "p90": 457.0,
+      "p95": 750.0,
+      "p99": 1468.999999999998,
+      "min": 28.0,
+      "max": 2878.0,
+      "std": 308.63626622413256,
+      "count": 241,
+      "sum": 47074.0
+    },
+    "http_req_data_sent": {
+      "unit": "KB",
+      "avg": 68.2171656185166,
+      "p1": 38.2875,
+      "p5": 39.5390625,
+      "p10": 42.7568359375,
+      "p25": 50.34765625,
+      "p50": 67.736328125,
+      "p75": 83.9091796875,
+      "p90": 94.8212890625,
+      "p95": 102.5244140625,
+      "p99": 111.03515625,
+      "min": 37.669921875,
+      "max": 112.720703125,
+      "std": 19.711629692469305,
+      "count": 241,
+      "sum": 16440.3369140625
+    },
+    "request_count": {
+      "unit": "requests",
+      "avg": 241.0
+    },
+    "http_req_connection_reused": {
+      "unit": "ratio",
+      "avg": 0.9045643153526971,
+      "p1": 0.0,
+      "p5": 0.0,
+      "p10": 1.0,
+      "p25": 1.0,
+      "p50": 1.0,
+      "p75": 1.0,
+      "p90": 1.0,
+      "p95": 1.0,
+      "p99": 1.0,
+      "min": 0.0,
+      "max": 1.0,
+      "std": 0.293815783686315,
+      "count": 241,
+      "sum": 218.0
+    },
+    "e2e_output_token_throughput": {
+      "unit": "tokens/sec/user",
+      "avg": 28.073209567126753,
+      "p1": 12.303453588885763,
+      "p5": 13.421027339307955,
+      "p10": 14.95022189884373,
+      "p25": 18.744244585557578,
+      "p50": 27.895661836342473,
+      "p75": 34.9827699355213,
+      "p90": 42.429732865755355,
+      "p95": 46.287404954916354,
+      "p99": 56.17375225119115,
+      "min": 11.8194698217423,
+      "max": 82.31327195849677,
+      "std": 10.917999025360475,
+      "count": 241,
+      "sum": 6765.643505677547
+    },
+    "inter_token_latency": {
+      "unit": "ms",
+      "avg": 32.304113948249444,
+      "p1": 13.313789233952381,
+      "p5": 14.796461031249999,
+      "p10": 18.884587533707865,
+      "p25": 24.04125092244224,
+      "p50": 30.77191221153846,
+      "p75": 40.55283363829787,
+      "p90": 48.90667095890411,
+      "p95": 54.306152174999994,
+      "p99": 63.34027829762324,
+      "min": 11.369600952380951,
+      "max": 69.28420243902438,
+      "std": 11.869366124787447,
+      "count": 241,
+      "sum": 7785.291461528116
+    },
+    "http_req_connection_overhead": {
+      "unit": "ms",
+      "avg": 0.2459006846473029,
+      "p1": 0.0,
+      "p5": 0.0,
+      "p10": 0.0,
+      "p25": 0.0,
+      "p50": 0.0,
+      "p75": 0.0,
+      "p90": 0.0,
+      "p95": 1.925215,
+      "p99": 4.238834599999996,
+      "min": 0.0,
+      "max": 7.167301999999999,
+      "std": 0.910956729743431,
+      "count": 241,
+      "sum": 59.262065
+    },
+    "osl_mismatch_diff_pct": {
+      "unit": "%",
+      "avg": -84.58128590053441,
+      "p1": -96.68068068068068,
+      "p5": -95.86397058823529,
+      "p10": -95.19492293744334,
+      "p25": -94.0934065934066,
+      "p50": -92.16965742251223,
+      "p75": -86.74496644295301,
+      "p90": -66.87898089171973,
+      "p95": -34.62603878116344,
+      "p99": 0.0,
+      "min": -97.4145891043398,
+      "max": 0.0,
+      "std": 19.96768070248593,
+      "count": 241,
+      "sum": -20384.089902028794
+    },
+    "http_req_total": {
+      "unit": "ms",
+      "avg": 6449.398267738589,
+      "p1": 1154.9478405999998,
+      "p5": 1614.2782129999998,
+      "p10": 1955.136744,
+      "p25": 2690.6977779999997,
+      "p50": 4073.557766,
+      "p75": 6957.057836999999,
+      "p90": 12383.135033999999,
+      "p95": 19791.78773,
+      "p99": 44645.168710799895,
+      "min": 1033.893081,
+      "max": 59098.172647,
+      "std": 7675.137857205938,
+      "count": 241,
+      "sum": 1554304.9825249999
+    },
+    "prefill_throughput_per_user": {
+      "unit": "tokens/sec/user",
+      "avg": 28757.530241561442,
+      "p1": 2543.625016035158,
+      "p5": 3191.421949925116,
+      "p10": 9601.310913153096,
+      "p25": 16017.355159311057,
+      "p50": 27380.808748900938,
+      "p75": 39038.876632615465,
+      "p90": 52834.96166280653,
+      "p95": 58278.102440923496,
+      "p99": 64599.47731380815,
+      "min": 2114.425940608069,
+      "max": 68699.48998485897,
+      "std": 16055.83049564443,
+      "count": 241,
+      "sum": 6930564.788216308
+    },
+    "output_token_throughput_per_user": {
+      "unit": "tokens/sec/user",
+      "avg": 35.63661659786708,
+      "p1": 15.791388827568547,
+      "p5": 18.41411994680693,
+      "p10": 20.447108347249642,
+      "p25": 24.659189267001196,
+      "p50": 32.49716797336477,
+      "p75": 41.59517336373337,
+      "p90": 52.95323491789584,
+      "p95": 67.58372815553722,
+      "p99": 75.13944276145915,
+      "min": 14.433304632179025,
+      "max": 87.95383445630834,
+      "std": 14.156716017368579,
+      "count": 241,
+      "sum": 8588.424600085968
+    },
+    "osl_mismatch_count": {
+      "unit": "requests",
+      "avg": 235.0
+    },
+    "usage_prompt_cache_read_tokens": {
+      "unit": "tokens",
+      "avg": 15873.094017094018,
+      "p1": 7744.0,
+      "p5": 8320.0,
+      "p10": 8787.2,
+      "p25": 11088.0,
+      "p50": 15424.0,
+      "p75": 20416.0,
+      "p90": 22976.0,
+      "p95": 24640.0,
+      "p99": 27221.76,
+      "min": 7744.0,
+      "max": 27392.0,
+      "std": 5317.08858382619,
+      "count": 234,
+      "sum": 3714304.0
+    },
+    "inter_chunk_latency": {
+      "unit": "ms",
+      "avg": 53.96321112427582,
+      "p1": 0.09511065,
+      "p5": 0.80385675,
+      "p10": 1.5959759999999998,
+      "p25": 26.22379625,
+      "p50": 29.84988,
+      "p75": 32.31039425,
+      "p90": 67.355098,
+      "p95": 296.10059325,
+      "p99": 575.7694134,
+      "min": 0.0,
+      "max": 1536.806096,
+      "std": 103.90145705545507,
+      "count": 24856,
+      "sum": 1341309.575705
+    },
+    "total_osl": {
+      "unit": "tokens",
+      "avg": 47074.0
+    },
+    "total_usage_total_tokens": {
+      "unit": "tokens",
+      "avg": 4036907.0
+    },
+    "total_usage_prompt_cache_read_tokens": {
+      "unit": "tokens",
+      "avg": 3714304.0
+    },
+    "total_output_tokens": {
+      "unit": "tokens",
+      "avg": 47074.0
+    },
+    "total_usage_prompt_tokens": {
+      "unit": "tokens",
+      "avg": 3989833.0
+    },
+    "total_isl": {
+      "unit": "tokens",
+      "avg": 3989833.0
+    },
+    "total_usage_completion_tokens": {
+      "unit": "tokens",
+      "avg": 47074.0
+    },
+    "overall_usage_prompt_cache_read_pct": {
+      "unit": "%",
+      "avg": 93.09422223937693
+    },
+    "benchmark_duration": {
+      "unit": "sec",
+      "avg": 132.77050496
+    },
+    "output_token_throughput": {
+      "unit": "tokens/sec",
+      "avg": 354.51728647131426
+    },
+    "total_token_throughput": {
+      "unit": "tokens/sec",
+      "avg": 30402.2032412171
+    },
+    "request_throughput": {
+      "unit": "requests/sec",
+      "avg": 1.8149863202529366
+    },
+    "input_token_throughput": {
+      "unit": "tokens/sec",
+      "avg": 30047.685954745786
+    },
+    "effective_concurrency": {
+      "unit": "requests",
+      "avg": 11.705373278957007,
+      "p50": 12.0,
+      "p90": 16.0,
+      "p95": 16.0,
+      "p99": 16.0,
+      "min": 0.0,
+      "max": 16.0,
+      "std": 3.7394016309983344
+    },
+    "effective_decode_throughput": {
+      "unit": "tokens/sec",
+      "avg": 352.7022846386635,
+      "p50": 107.35102818780142,
+      "p90": 817.5541656890833,
+      "p95": 908.2879381943676,
+      "p99": 1472.3290044875414,
+      "min": 0.0,
+      "max": 78748.34602473544,
+      "std": 810.0301685156334
+    },
+    "effective_prefill_throughput": {
+      "unit": "tokens/sec",
+      "avg": 30047.686002945786,
+      "p50": 24354.863644473295,
+      "p90": 71913.8557978175,
+      "p95": 93336.66757332934,
+      "p99": 120764.60696853818,
+      "min": 0.0,
+      "max": 160735.59725270994,
+      "std": 33409.30463987072
+    },
+    "effective_decode_concurrency": {
+      "unit": "requests",
+      "avg": 10.149716593521635,
+      "p50": 11.0,
+      "p90": 14.0,
+      "p95": 15.0,
+      "p99": 16.0,
+      "min": 0.0,
+      "max": 16.0,
+      "std": 3.7335543117713055
+    },
+    "effective_prefill_concurrency": {
+      "unit": "requests",
+      "avg": 1.555656685435372,
+      "p50": 1.0,
+      "p90": 4.0,
+      "p95": 5.0,
+      "p99": 16.0,
+      "min": 0.0,
+      "max": 16.0,
+      "std": 2.794357095099477
+    },
+    "effective_total_throughput": {
+      "unit": "tokens/sec",
+      "avg": 30400.388303662654,
+      "p50": 24626.127136969517,
+      "p90": 72224.17428326508,
+      "p95": 93674.97634741483,
+      "p99": 121108.61706407806,
+      "min": 0.0,
+      "max": 161023.43729087771,
+      "std": 33382.539007240004
+    },
+    "effective_decode_throughput_per_user": {
+      "unit": "tokens/sec/user",
+      "avg": 36.45854118283859,
+      "p50": 18.42288738011495,
+      "p90": 68.95586151364934,
+      "p95": 88.68757528776258,
+      "p99": 136.75632614476245,
+      "min": 0.0,
+      "max": 29984.680421644767,
+      "std": 106.01240973860584
+    },
+    "effective_prefill_throughput_per_user": {
+      "unit": "tokens/sec/user",
+      "avg": 16590.414384638345,
+      "p50": 11007.704946190575,
+      "p90": 48417.875086118525,
+      "p95": 55230.72267643953,
+      "p99": 63210.21094572981,
+      "min": 0.0,
+      "max": 68699.5051488535,
+      "std": 19235.866952668322
+    },
+    "tokens_in_flight": {
+      "unit": "tokens",
+      "avg": 198004.4614081333,
+      "p50": 209962.27705657147,
+      "p90": 259201.86061257482,
+      "p95": 268496.7809212032,
+      "p99": 277122.0464846955,
+      "min": 0.0,
+      "max": 280727.84753053763,
+      "std": 58938.618319417976
+    },
+    "active_decode_throughput": {
+      "unit": "tokens/sec",
+      "avg": 360.4415740767312,
+      "p50": 119.57492628128502,
+      "p90": 819.5537394369597,
+      "p95": 909.094515594917,
+      "p99": 1477.7213948672293,
+      "min": 0.0,
+      "max": 78748.34602473544,
+      "std": 817.1640480751794
+    },
+    "active_prefill_throughput": {
+      "unit": "tokens/sec",
+      "avg": 54638.057855494684,
+      "p50": 50337.065273143446,
+      "p90": 87977.73774487291,
+      "p95": 106897.41349482762,
+      "p99": 148976.17115349078,
+      "min": 2629.109968682532,
+      "max": 160735.59725270994,
+      "std": 26192.942177387486
+    },
+    "active_decode_throughput_per_user": {
+      "unit": "tokens/sec/user",
+      "avg": 37.258545088142334,
+      "p50": 21.59865785298506,
+      "p90": 69.20210707410051,
+      "p95": 89.56082913485018,
+      "p99": 139.52112630985422,
+      "min": 0.0,
+      "max": 29984.680421644767,
+      "std": 107.03004956490568
+    },
+    "active_prefill_throughput_per_user": {
+      "unit": "tokens/sec/user",
+      "avg": 30167.648214429337,
+      "p50": 28944.185059330954,
+      "p90": 54740.743672204386,
+      "p95": 59775.0596437083,
+      "p99": 67875.00181486599,
+      "min": 2533.6672632417926,
+      "max": 68699.5051488535,
+      "std": 16224.688258701028
+    },
+    "active_total_throughput": {
+      "unit": "tokens/sec",
+      "avg": 30403.333942400368,
+      "p50": 24626.127136969517,
+      "p90": 72224.17428326508,
+      "p95": 93674.97634741483,
+      "p99": 121108.61706407806,
+      "min": 84.45666473309552,
+      "max": 161023.43729087771,
+      "std": 33382.81492440671
+    },
+    "credit_to_start_latency": {
+      "unit": "ms",
+      "avg": 1.6703893775933611,
+      "p1": 0.938752,
+      "p5": 1.05984,
+      "p10": 1.090816,
+      "p25": 1.206528,
+      "p50": 1.318144,
+      "p75": 1.48864,
+      "p90": 1.930752,
+      "p95": 4.247808,
+      "p99": 7.972863999999998,
+      "min": 0.893952,
+      "max": 11.598336,
+      "std": 1.3275723332826808,
+      "count": 241,
+      "sum": 402.56384
+    },
+    "effective_latency": {
+      "unit": "ms",
+      "avg": 6450.96046366805,
+      "p1": 1157.7427456,
+      "p5": 1615.69408,
+      "p10": 1956.862976,
+      "p25": 2692.097024,
+      "p50": 4074.9824,
+      "p75": 6958.10048,
+      "p90": 12384.707328,
+      "p95": 19793.08416,
+      "p99": 44646.3783423999,
+      "min": 1035.595776,
+      "max": 59099.826176,
+      "std": 7675.142337640313,
+      "count": 241,
+      "sum": 1554681.471744
+    }
+  },
+  "time_to_first_output_token": {
+    "unit": "ms",
+    "avg": 4987.187347749462,
+    "p1": 380.32252808999993,
+    "p5": 403.5403151,
+    "p10": 426.5102391,
+    "p25": 638.1150037499999,
+    "p50": 1329.972233,
+    "p75": 5279.2070942499995,
+    "p90": 17187.5171505,
+    "p95": 23047.71537219999,
+    "p99": 31966.107753890024,
+    "min": 356.05555899999996,
+    "max": 38854.914205,
+    "std": 7503.357305740956,
+    "count": 930,
+    "sum": 4638084.233407
+  },
+  "http_req_sending": {
+    "unit": "ms",
+    "avg": 0.34954696826923076,
+    "p1": 0.18586379,
+    "p5": 0.196381,
+    "p10": 0.20171909999999998,
+    "p25": 0.216939,
+    "p50": 0.24878599999999998,
+    "p75": 0.3022105,
+    "p90": 0.39130560000000003,
+    "p95": 0.6372417999999987,
+    "p99": 2.2343898199999748,
+    "min": 0.14196599999999998,
+    "max": 9.155971,
+    "std": 0.47233247756759456,
+    "count": 1040,
+    "sum": 363.528847
+  },
+  "http_req_duration": {
+    "unit": "ms",
+    "avg": 10815.387401517306,
+    "p1": 1158.35047863,
+    "p5": 1582.1962592,
+    "p10": 1871.6814244999998,
+    "p25": 2776.5147105,
+    "p50": 5273.4430935,
+    "p75": 17245.55761275,
+    "p90": 27950.497951499998,
+    "p95": 33243.40889325,
+    "p99": 40890.98136956999,
+    "min": 934.6873459999999,
+    "max": 52220.236654,
+    "std": 10897.950335877096,
+    "count": 1040,
+    "sum": 11248002.897578
+  },
+  "usage_total_tokens": {
+    "unit": "tokens",
+    "avg": 25167.61826923077,
+    "p1": 8334.24,
+    "p5": 9195.7,
+    "p10": 10307.1,
+    "p25": 15575.25,
+    "p50": 23067.5,
+    "p75": 32788.25,
+    "p90": 43592.3,
+    "p95": 48338.899999999994,
+    "p99": 58307.57,
+    "min": 8187.0,
+    "max": 65628.0,
+    "std": 12363.772111941094,
+    "count": 1040,
+    "sum": 26174323.0
+  },
+  "http_req_chunks_received": {
+    "unit": "count",
+    "avg": 238.75480769230768,
+    "p1": 21.39,
+    "p5": 27.0,
+    "p10": 33.0,
+    "p25": 49.0,
+    "p50": 104.5,
+    "p75": 393.0,
+    "p90": 639.0,
+    "p95": 767.1999999999998,
+    "p99": 961.9299999999987,
+    "min": 19.0,
+    "max": 1195.0,
+    "std": 257.05740992260246,
+    "count": 1040,
+    "sum": 248305.0
+  },
+  "http_req_connecting": {
+    "unit": "ms",
+    "avg": 0.07612035673076922,
+    "p1": 0.0,
+    "p5": 0.0,
+    "p10": 0.0,
+    "p25": 0.0,
+    "p50": 0.0,
+    "p75": 0.0,
+    "p90": 0.0,
+    "p95": 0.0,
+    "p99": 1.5905187199999968,
+    "min": 0.0,
+    "max": 8.317432,
+    "std": 0.42438744503995973,
+    "count": 1040,
+    "sum": 79.165171
+  },
+  "http_req_chunks_sent": {
+    "unit": "count",
+    "avg": 1.0,
+    "p1": 1.0,
+    "p5": 1.0,
+    "p10": 1.0,
+    "p25": 1.0,
+    "p50": 1.0,
+    "p75": 1.0,
+    "p90": 1.0,
+    "p95": 1.0,
+    "p99": 1.0,
+    "min": 1.0,
+    "max": 1.0,
+    "std": 0.0,
+    "count": 1040,
+    "sum": 1040.0
+  },
+  "http_req_blocked": {
+    "unit": "ms",
+    "avg": 0.0,
+    "p1": 0.0,
+    "p5": 0.0,
+    "p10": 0.0,
+    "p25": 0.0,
+    "p50": 0.0,
+    "p75": 0.0,
+    "p90": 0.0,
+    "p95": 0.0,
+    "p99": 0.0,
+    "min": 0.0,
+    "max": 0.0,
+    "std": 0.0,
+    "count": 1040,
+    "sum": 0.0
+  },
+  "http_req_dns_lookup": {
+    "unit": "ms",
+    "avg": 0.0,
+    "p1": 0.0,
+    "p5": 0.0,
+    "p10": 0.0,
+    "p25": 0.0,
+    "p50": 0.0,
+    "p75": 0.0,
+    "p90": 0.0,
+    "p95": 0.0,
+    "p99": 0.0,
+    "min": 0.0,
+    "max": 0.0,
+    "std": 0.0,
+    "count": 1040,
+    "sum": 0.0
+  },
+  "http_req_receiving": {
+    "unit": "ms",
+    "avg": 10207.194423883653,
+    "p1": 691.8602401699999,
+    "p5": 1053.8283281499998,
+    "p10": 1306.6632351,
+    "p25": 2201.88973175,
+    "p50": 4530.178124,
+    "p75": 16586.29723225,
+    "p90": 27173.622890000002,
+    "p95": 32563.80672329999,
+    "p99": 40360.41726267997,
+    "min": 556.006323,
+    "max": 51774.446939999994,
+    "std": 10908.788578871703,
+    "count": 1040,
+    "sum": 10615482.200839
+  },
+  "http_req_waiting": {
+    "unit": "ms",
+    "avg": 607.8434306653846,
+    "p1": 331.86272102999993,
+    "p5": 352.31707819999997,
+    "p10": 365.47957219999995,
+    "p25": 395.62904449999996,
+    "p50": 525.3089165,
+    "p75": 683.72413575,
+    "p90": 850.6335966,
+    "p95": 946.6596416,
+    "p99": 3580.4136180799997,
+    "min": 325.619899,
+    "max": 3592.007803,
+    "std": 413.1271143233787,
+    "count": 1040,
+    "sum": 632157.1678919999
+  },
+  "http_req_data_received": {
+    "unit": "KB",
+    "avg": 58.571347280649036,
+    "p1": 5.815810546875,
+    "p5": 7.5029296875,
+    "p10": 8.84482421875,
+    "p25": 12.827392578125,
+    "p50": 26.22802734375,
+    "p75": 95.880615234375,
+    "p90": 153.00087890625,
+    "p95": 185.29370117187494,
+    "p99": 233.35729492187414,
+    "min": 5.310546875,
+    "max": 290.322265625,
+    "std": 61.808473261880096,
+    "count": 1040,
+    "sum": 60914.201171875
+  },
+  "usage_prompt_tokens": {
+    "unit": "tokens",
+    "avg": 24722.80576923077,
+    "p1": 8265.39,
+    "p5": 9116.45,
+    "p10": 10190.5,
+    "p25": 15107.75,
+    "p50": 22658.0,
+    "p75": 32484.25,
+    "p90": 42932.3,
+    "p95": 47748.049999999996,
+    "p99": 57856.36999999999,
+    "min": 8100.0,
+    "max": 64531.0,
+    "std": 12184.919747090813,
+    "count": 1040,
+    "sum": 25711718.0
+  },
+  "usage_completion_tokens": {
+    "unit": "tokens",
+    "avg": 444.8125,
+    "p1": 41.0,
+    "p5": 52.95,
+    "p10": 63.0,
+    "p25": 91.0,
+    "p50": 191.0,
+    "p75": 719.25,
+    "p90": 1178.5,
+    "p95": 1441.3999999999996,
+    "p99": 1814.199999999998,
+    "min": 37.0,
+    "max": 2260.0,
+    "std": 480.39091297278605,
+    "count": 1040,
+    "sum": 462605.0
+  },
+  "http_req_data_sent": {
+    "unit": "KB",
+    "avg": 102.13143028846154,
+    "p1": 37.593271484375,
+    "p5": 40.617333984375,
+    "p10": 44.775,
+    "p25": 63.240478515625,
+    "p50": 94.4453125,
+    "p75": 130.843505859375,
+    "p90": 174.1220703125,
+    "p95": 196.86909179687498,
+    "p99": 239.30867187499987,
+    "min": 36.833984375,
+    "max": 258.052734375,
+    "std": 49.09589484385997,
+    "count": 1040,
+    "sum": 106216.6875
+  },
+  "http_req_connection_reused": {
+    "unit": "ratio",
+    "avg": 0.9509615384615384,
+    "p1": 0.0,
+    "p5": 1.0,
+    "p10": 1.0,
+    "p25": 1.0,
+    "p50": 1.0,
+    "p75": 1.0,
+    "p90": 1.0,
+    "p95": 1.0,
+    "p99": 1.0,
+    "min": 0.0,
+    "max": 1.0,
+    "std": 0.21594835222432773,
+    "count": 1040,
+    "sum": 989.0
+  },
+  "e2e_output_token_throughput": {
+    "unit": "tokens/sec/user",
+    "avg": 38.66858205694003,
+    "p1": 14.265670708255813,
+    "p5": 20.513176826218007,
+    "p10": 27.22410395050239,
+    "p25": 34.68689948893622,
+    "p50": 40.4632161941492,
+    "p75": 44.406453609027274,
+    "p90": 47.42086631460469,
+    "p95": 48.88881164666021,
+    "p99": 54.15890769274403,
+    "min": 11.906680767535468,
+    "max": 59.100714351564186,
+    "std": 8.377195675004616,
+    "count": 1040,
+    "sum": 40215.325339217634
+  },
+  "http_req_connection_overhead": {
+    "unit": "ms",
+    "avg": 0.07612035673076922,
+    "p1": 0.0,
+    "p5": 0.0,
+    "p10": 0.0,
+    "p25": 0.0,
+    "p50": 0.0,
+    "p75": 0.0,
+    "p90": 0.0,
+    "p95": 0.0,
+    "p99": 1.5905187199999968,
+    "min": 0.0,
+    "max": 8.317432,
+    "std": 0.42438744503995973,
+    "count": 1040,
+    "sum": 79.165171
+  },
+  "osl_mismatch_diff_pct": {
+    "unit": "%",
+    "avg": -65.6649534967913,
+    "p1": -96.56954955384836,
+    "p5": -95.48013942099699,
+    "p10": -94.65989339165947,
+    "p25": -92.00194766648323,
+    "p50": -84.13059124750745,
+    "p75": -42.054703978204834,
+    "p90": 0.0,
+    "p95": 0.0,
+    "p99": 0.0,
+    "min": -97.82850779510022,
+    "max": 0.0,
+    "std": 34.405816468417115,
+    "count": 1040,
+    "sum": -68291.55163666295
+  },
+  "http_req_total": {
+    "unit": "ms",
+    "avg": 10815.463521874039,
+    "p1": 1158.35047863,
+    "p5": 1582.1962592,
+    "p10": 1871.6814244999998,
+    "p25": 2776.5147105,
+    "p50": 5273.4430935,
+    "p75": 17245.55761275,
+    "p90": 27950.6551214,
+    "p95": 33243.40889325,
+    "p99": 40890.98136956999,
+    "min": 934.6873459999999,
+    "max": 52220.236654,
+    "std": 10897.931651080315,
+    "count": 1040,
+    "sum": 11248082.062749
+  },
+  "prefill_throughput_per_user": {
+    "unit": "tokens/sec/user",
+    "avg": 47005.625091708695,
+    "p1": 2776.0337281079533,
+    "p5": 12061.96888020083,
+    "p10": 15786.467627695452,
+    "p25": 26469.88396329444,
+    "p50": 42607.352726775556,
+    "p75": 63832.66218674383,
+    "p90": 84377.64797846085,
+    "p95": 98446.65827287588,
+    "p99": 117078.30373636782,
+    "min": 2261.548883006647,
+    "max": 141468.65381190402,
+    "std": 26442.50438075831,
+    "count": 1040,
+    "sum": 48885850.09537704
+  },
+  "osl_mismatch_count": {
+    "unit": "requests",
+    "avg": 895.0
+  },
+  "usage_prompt_cache_read_tokens": {
+    "unit": "tokens",
+    "avg": 23983.751937984496,
+    "p1": 7744.0,
+    "p5": 8512.0,
+    "p10": 9792.0,
+    "p25": 14208.0,
+    "p50": 21696.0,
+    "p75": 30928.0,
+    "p90": 42163.2,
+    "p95": 47126.39999999999,
+    "p99": 56792.31999999999,
+    "min": 7744.0,
+    "max": 64384.0,
+    "std": 12131.374464094235,
+    "count": 1032,
+    "sum": 24751232.0
+  },
+  "total_usage_total_tokens": {
+    "unit": "tokens",
+    "avg": 26174323.0
+  },
+  "total_usage_prompt_cache_read_tokens": {
+    "unit": "tokens",
+    "avg": 24751232.0
+  },
+  "total_usage_prompt_tokens": {
+    "unit": "tokens",
+    "avg": 25711718.0
+  },
+  "total_usage_completion_tokens": {
+    "unit": "tokens",
+    "avg": 462605.0
+  },
+  "overall_usage_prompt_cache_read_pct": {
+    "unit": "%",
+    "avg": 96.26440364661747
+  },
+  "total_token_throughput": {
+    "unit": "tokens/sec",
+    "avg": 28597.033018295126
+  },
+  "input_token_throughput": {
+    "unit": "tokens/sec",
+    "avg": 28091.60904001579
+  },
+  "effective_concurrency": {
+    "unit": "requests",
+    "avg": 12.289278364405316,
+    "p50": 12.0,
+    "p90": 15.0,
+    "p95": 16.0,
+    "p99": 16.0,
+    "min": 0.0,
+    "max": 16.0,
+    "std": 1.8224280612135288
+  },
+  "effective_decode_throughput": {
+    "unit": "tokens/sec",
+    "avg": 504.28766325286847,
+    "p50": 446.39935993434557,
+    "p90": 821.3824949016414,
+    "p95": 898.3375598394971,
+    "p99": 2081.033259475957,
+    "min": 0.0,
+    "max": 136616.39698529555,
+    "std": 979.264515848831
+  },
+  "effective_prefill_throughput": {
+    "unit": "tokens/sec",
+    "avg": 28091.609043514654,
+    "p50": 0.0,
+    "p90": 89733.77036939427,
+    "p95": 110602.02697906621,
+    "p99": 148988.76790548197,
+    "min": 0.0,
+    "max": 211712.97380551466,
+    "std": 40048.6595601544
+  },
+  "effective_decode_concurrency": {
+    "unit": "requests",
+    "avg": 11.575364724013559,
+    "p50": 12.0,
+    "p90": 14.0,
+    "p95": 15.0,
+    "p99": 16.0,
+    "min": 0.0,
+    "max": 16.0,
+    "std": 2.0191769328892524
+  },
+  "effective_prefill_concurrency": {
+    "unit": "requests",
+    "avg": 0.7139136403917576,
+    "p50": 0.0,
+    "p90": 2.0,
+    "p95": 3.0,
+    "p99": 5.0,
+    "min": 0.0,
+    "max": 16.0,
+    "std": 1.3822099195208082
+  },
+  "effective_total_throughput": {
+    "unit": "tokens/sec",
+    "avg": 28595.89675887357,
+    "p50": 606.7579312905584,
+    "p90": 90235.29587586965,
+    "p95": 111132.48128457196,
+    "p99": 149438.8542609219,
+    "min": 0.0,
+    "max": 212194.50129677076,
+    "std": 40016.369871469615
+  },
+  "effective_decode_throughput_per_user": {
+    "unit": "tokens/sec/user",
+    "avg": 42.600143389284334,
+    "p50": 38.191944592181784,
+    "p90": 65.00810449860268,
+    "p95": 68.18693986204015,
+    "p99": 173.8643721571205,
+    "min": 0.0,
+    "max": 13960.980687776044,
+    "std": 85.16906233888278
+  },
+  "effective_prefill_throughput_per_user": {
+    "unit": "tokens/sec/user",
+    "avg": 20089.116929706553,
+    "p50": 0.0,
+    "p90": 65049.86997857938,
+    "p95": 80961.87438319871,
+    "p99": 109360.34377311573,
+    "min": 0.0,
+    "max": 141468.65133081685,
+    "std": 29067.60671947812
+  },
+  "tokens_in_flight": {
+    "unit": "tokens",
+    "avg": 356248.92639257736,
+    "p50": 362879.28496338654,
+    "p90": 453099.60343566246,
+    "p95": 466969.64088061627,
+    "p99": 492375.694775443,
+    "min": 0.0,
+    "max": 505096.38866028364,
+    "std": 77305.98573160678
+  },
+  "active_decode_throughput": {
+    "unit": "tokens/sec",
+    "avg": 505.96648965200035,
+    "p50": 452.52790957399634,
+    "p90": 821.6787159639439,
+    "p95": 899.0543123444412,
+    "p99": 2087.5273766103946,
+    "min": 0.0,
+    "max": 136616.39698529555,
+    "std": 980.4601149406376
+  },
+  "active_prefill_throughput": {
+    "unit": "tokens/sec",
+    "avg": 67997.99065468695,
+    "p50": 62853.70861922176,
+    "p90": 114519.99072188952,
+    "p95": 134481.28214545592,
+    "p99": 163691.59652235563,
+    "min": 2294.053449267356,
+    "max": 211712.97380551466,
+    "std": 34187.75985248943
+  },
+  "active_decode_throughput_per_user": {
+    "unit": "tokens/sec/user",
+    "avg": 42.74196372644652,
+    "p50": 38.48725696119699,
+    "p90": 65.01984841710521,
+    "p95": 68.22508143640773,
+    "p99": 174.18495510848527,
+    "min": 0.0,
+    "max": 13960.980687776044,
+    "std": 85.27517843414842
+  },
+  "active_prefill_throughput_per_user": {
+    "unit": "tokens/sec/user",
+    "avg": 48627.3172580146,
+    "p50": 44712.08094179233,
+    "p90": 85023.03147703139,
+    "p95": 99464.69907407404,
+    "p99": 118005.1681195887,
+    "min": 2294.053449267356,
+    "max": 141468.65133081685,
+    "std": 25641.263329510908
+  },
+  "active_total_throughput": {
+    "unit": "tokens/sec",
+    "avg": 28596.567155004155,
+    "p50": 606.7579312905584,
+    "p90": 90235.29587586965,
+    "p95": 111132.48128457196,
+    "p99": 149438.8542609219,
+    "min": 57.173891829611435,
+    "max": 212194.50129677076,
+    "std": 40016.599398631064
+  },
+  "credit_to_start_latency": {
+    "unit": "ms",
+    "avg": 1.539191876923077,
+    "p1": 0.903104,
+    "p5": 1.0443264,
+    "p10": 1.0991360000000001,
+    "p25": 1.202688,
+    "p50": 1.3497599999999998,
+    "p75": 1.5452160000000001,
+    "p90": 1.8894336,
+    "p95": 2.7408639999999993,
+    "p99": 5.6142847999999566,
+    "min": 0.827648,
+    "max": 14.525696,
+    "std": 0.8454953385283775,
+    "count": 1040,
+    "sum": 1600.759552
+  },
+  "effective_latency": {
+    "unit": "ms",
+    "avg": 10817.06213243077,
+    "p1": 1160.33114112,
+    "p5": 1583.6966400000001,
+    "p10": 1872.934016,
+    "p25": 2778.061184,
+    "p50": 5275.087104,
+    "p75": 17247.145024,
+    "p90": 27953.632896000003,
+    "p95": 33244.6491264,
+    "p99": 40892.59686143999,
+    "min": 935.748352,
+    "max": 52221.874688,
+    "std": 10897.928973515816,
+    "count": 1040,
+    "sum": 11249744.617728
+  }
+}

experiments/kv-cache-fp8/capacity/hammer/N0016/profile_export_console.txt

@@ -0,0 +1,87 @@
+
+
+                                              NVIDIA AIPerf | LLM Metrics: Effective                                               
+┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━┓
+┃                                    Metric ┃        avg ┃    min ┃        max ┃        p99 ┃        p90 ┃        p50 ┃       std ┃
+┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━┩
+│          Effective Concurrency (requests) │      12.29 │   0.00 │      16.00 │      16.00 │      15.00 │      12.00 │      1.82 │
+│  Effective Decode Throughput (tokens/sec) │     504.29 │   0.00 │ 136,616.40 │   2,081.03 │     821.38 │     446.40 │    979.26 │
+│ Effective Prefill Throughput (tokens/sec) │  28,091.61 │   0.00 │ 211,712.97 │ 148,988.77 │  89,733.77 │       0.00 │ 40,048.66 │
+│   Effective Decode Concurrency (requests) │      11.58 │   0.00 │      16.00 │      16.00 │      14.00 │      12.00 │      2.02 │
+│  Effective Prefill Concurrency (requests) │       0.71 │   0.00 │      16.00 │       5.00 │       2.00 │       0.00 │      1.38 │
+│   Effective Total Throughput (tokens/sec) │  28,595.90 │   0.00 │ 212,194.50 │ 149,438.85 │  90,235.30 │     606.76 │ 40,016.37 │
+│      Effective Decode Throughput Per User │      42.60 │   0.00 │  13,960.98 │     173.86 │      65.01 │      38.19 │     85.17 │
+│                         (tokens/sec/user) │            │        │            │            │            │            │           │
+│     Effective Prefill Throughput Per User │  20,089.12 │   0.00 │ 141,468.65 │ 109,360.34 │  65,049.87 │       0.00 │ 29,067.61 │
+│                         (tokens/sec/user) │            │        │            │            │            │            │           │
+│                 Tokens In Flight (tokens) │ 356,248.93 │   0.00 │ 505,096.39 │ 492,375.69 │ 453,099.60 │ 362,879.28 │ 77,305.99 │
+│         Effective Latency (CO-aware) (ms) │  10,817.06 │ 935.75 │  52,221.87 │  40,892.60 │  27,953.63 │   5,275.09 │ 10,897.93 │
+└───────────────────────────────────────────┴────────────┴────────┴────────────┴────────────┴────────────┴────────────┴───────────┘
+                                              NVIDIA AIPerf | LLM Metrics: Active                                               
+┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┓
+┃                                 Metric ┃       avg ┃      min ┃        max ┃        p99 ┃        p90 ┃       p50 ┃       std ┃
+┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━┩
+│  Active Decode Throughput (tokens/sec) │    505.97 │     0.00 │ 136,616.40 │   2,087.53 │     821.68 │    452.53 │    980.46 │
+│ Active Prefill Throughput (tokens/sec) │ 67,997.99 │ 2,294.05 │ 211,712.97 │ 163,691.60 │ 114,519.99 │ 62,853.71 │ 34,187.76 │
+│      Active Decode Throughput Per User │     42.74 │     0.00 │  13,960.98 │     174.18 │      65.02 │     38.49 │     85.28 │
+│                      (tokens/sec/user) │           │          │            │            │            │           │           │
+│     Active Prefill Throughput Per User │ 48,627.32 │ 2,294.05 │ 141,468.65 │ 118,005.17 │  85,023.03 │ 44,712.08 │ 25,641.26 │
+│                      (tokens/sec/user) │           │          │            │            │            │           │           │
+│   Active Total Throughput (tokens/sec) │ 28,596.57 │    57.17 │ 212,194.50 │ 149,438.85 │  90,235.30 │    606.76 │ 40,016.60 │
+└────────────────────────────────────────┴───────────┴──────────┴────────────┴────────────┴────────────┴───────────┴───────────┘
+                                                NVIDIA AIPerf | LLM Metrics: Usage                                                
+┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┓
+┃                                  Metric ┃           avg ┃      min ┃       max ┃       p99 ┃       p90 ┃       p50 ┃       std ┃
+┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━┩
+│            Usage Prompt Tokens (tokens) │     24,722.81 │ 8,100.00 │ 64,531.00 │ 57,856.37 │ 42,932.30 │ 22,658.00 │ 12,184.92 │
+│ Usage Prompt Cache Read Tokens (tokens) │     23,983.75 │ 7,744.00 │ 64,384.00 │ 56,792.32 │ 42,163.20 │ 21,696.00 │ 12,131.37 │
+│        Usage Completion Tokens (tokens) │        444.81 │    37.00 │  2,260.00 │  1,814.20 │  1,178.50 │    191.00 │    480.39 │
+│             Usage Total Tokens (tokens) │     25,167.62 │ 8,187.00 │ 65,628.00 │ 58,307.57 │ 43,592.30 │ 23,067.50 │ 12,363.77 │
+│      Total Usage Prompt Tokens (tokens) │ 25,711,718.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│    Total Usage Prompt Cache Read Tokens │ 24,751,232.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│                                (tokens) │               │          │           │           │           │           │           │
+│       Overall Usage Prompt Cache Read % │         96.26 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│                                     (%) │               │          │           │           │           │           │           │
+│  Total Usage Completion Tokens (tokens) │    462,605.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│       Total Usage Total Tokens (tokens) │ 26,174,323.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+└─────────────────────────────────────────┴───────────────┴──────────┴───────────┴───────────┴───────────┴───────────┴───────────┘
+                                                    NVIDIA AIPerf | LLM Metrics                                                     
+┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┓
+┃                                        Metric ┃       avg ┃      min ┃       max ┃       p99 ┃       p90 ┃       p50 ┃       std ┃
+┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━┩
+│                      Time to First Token (ms) │    628.30 │   325.98 │  3,592.81 │  3,582.36 │    900.76 │    539.64 │    422.70 │
+│                     Time to Second Token (ms) │     62.88 │     0.00 │  1,867.07 │    604.22 │     75.17 │     28.46 │    134.29 │
+│               Time to First Output Token (ms) │  4,987.19 │   356.06 │ 38,854.91 │ 31,966.11 │ 17,187.52 │  1,329.97 │  7,503.36 │
+│                          Request Latency (ms) │ 10,793.22 │   934.78 │ 52,220.37 │ 40,891.13 │ 27,950.81 │  5,243.05 │ 10,904.72 │
+│                      Inter Token Latency (ms) │     23.30 │    11.93 │     58.21 │     46.99 │     30.81 │     22.12 │      6.31 │
+│              Output Token Throughput Per User │     45.51 │    17.18 │     83.82 │     71.13 │     60.38 │     45.20 │     10.33 │
+│                             (tokens/sec/user) │           │          │           │           │           │           │           │
+│ E2E Output Token Throughput (tokens/sec/user) │     38.67 │    11.91 │     59.10 │     54.16 │     47.42 │     40.46 │      8.38 │
+│               Output Sequence Length (tokens) │    444.81 │    37.00 │  2,260.00 │  1,814.20 │  1,178.50 │    191.00 │    480.39 │
+│                Input Sequence Length (tokens) │ 24,722.81 │ 8,100.00 │ 64,531.00 │ 57,856.37 │ 42,932.30 │ 22,658.00 │ 12,184.92 │
+│          Output Token Throughput (tokens/sec) │    505.42 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│           Input Token Throughput (tokens/sec) │ 28,091.61 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│             Request Throughput (requests/sec) │      1.14 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│                      Request Count (requests) │  1,040.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+└───────────────────────────────────────────────┴───────────┴──────────┴───────────┴───────────┴───────────┴───────────┴───────────┘
+
+
+No GPU telemetry data collected during the benchmarking run.
+
+╭──────────────────────────────── Output Sequence Length Mismatch Warning ────────────────────────────────╮
+│  895 of 1,040 requests (86.1%) have output length differing from requested by more than the threshold.  │
+│  Threshold (tokens): min(requested x 5%, 50)                                                            │
+│  Average mismatch: -65.7%                                                                               │
+│                                                                                                         │
+│  Why: Server hit EOS token before reaching requested output length.                                     │
+│                                                                                                         │
+│  Fix Options:                                                                                           │
+│    - --extra-inputs ignore_eos:true - Generate until max_tokens (vLLM, TensorRT-LLM)                    │
+│    - --extra-inputs min_tokens:<N> - Set minimum output length (vLLM, TensorRT-LLM, SGLang)             │
+│    - --use-server-token-count - Use server-reported token counts if tokenizer mismatch suspected        │
+│                                                                                                         │
+│  Diagnostics:                                                                                           │
+│    - Review profile_export.jsonl -> osl_mismatch_diff_pct for per-request values                        │
+│    - Adjust: AIPERF_METRICS_OSL_MISMATCH_PCT_THRESHOLD=5                                                │
+│    - Adjust: AIPERF_METRICS_OSL_MISMATCH_MAX_TOKEN_THRESHOLD=50                                         │
+╰─────────────────────────────────────────────────────────────────────────────────────────────────────────╯

experiments/kv-cache-fp8/capacity/hammer/N0016/summary.json

@@ -0,0 +1,25 @@
+{
+  "sent": 1040,
+  "records": 1040,
+  "valid": 1040,
+  "valid_frac": 1.0,
+  "p25_speed_tok_s": 39.9,
+  "p50_speed_tok_s": 45.2,
+  "p95_ttft_ms": 1046.0,
+  "p50_ttft_ms": 539.0,
+  "cache_read_ratio": 0.9626,
+  "max_turn_index": 49,
+  "conversations": 32,
+  "recycled_record_frac": 0.0,
+  "inflight_mean": 14.6,
+  "inflight_p95": 19.0,
+  "foreign_p50": 2.0,
+  "foreign_p95": 4.0,
+  "active_sessions_mean": 15.4,
+  "active_sessions_p95": 16.0,
+  "tiers": {
+    "20tok": true,
+    "60tok": false,
+    "180tok": false
+  }
+}

experiments/kv-cache-fp8/capacity/hammer/N0016/verify_cache_salt.log

@@ -0,0 +1 @@
+records_checked=1040 mismatches=0

experiments/kv-cache-fp8/capacity/hammer/N0016/verify_pin.log

@@ -0,0 +1 @@
+mode=pinned records_checked=1040 mismatches=0

experiments/kv-cache-fp8/capacity/hammer/N0036/aiperf_stdout.log

@@ -0,0 +1,236 @@
+  115,295.03 │  39,885.87 │  48,575.04 │
+│      Effective Decode Throughput Per User │      25.34 │   0.00 │    24,742.90 │       127.65 │        52.19 │       6.45 │      49.35 │
+│                         (tokens/sec/user) │            │        │              │              │              │            │            │
+│     Effective Prefill Throughput Per User │  21,052.55 │   0.00 │   109,836.09 │    87,080.67 │    52,205.55 │  15,019.37 │  21,514.58 │
+│                         (tokens/sec/user) │            │        │              │              │              │            │            │
+│                 Tokens In Flight (tokens) │ 861,160.37 │   0.00 │ 1,269,245.62 │ 1,230,231.58 │ 1,164,760.30 │ 894,248.97 │ 259,402.48 │
+│         Effective Latency (CO-aware) (ms) │  15,809.56 │ 940.19 │    93,149.09 │    67,596.82 │    43,665.83 │   7,883.86 │  17,014.83 │
+└───────────────────────────────────────────┴────────────┴────────┴──────────────┴──────────────┴──────────────┴────────────┴────────────┘
+                                             NVIDIA AIPerf | LLM Metrics: Active                                              
+┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┓
+┃                                 Metric ┃       avg ┃    min ┃        max ┃        p99 ┃        p90 ┃       p50 ┃       std ┃
+┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━┩
+│  Active Decode Throughput (tokens/sec) │    740.87 │   0.00 │ 103,688.33 │   3,778.06 │   1,643.35 │    196.70 │  1,345.24 │
+│ Active Prefill Throughput (tokens/sec) │ 60,858.36 │ 421.33 │ 341,378.40 │ 207,284.22 │ 121,240.64 │ 54,718.68 │ 47,009.73 │
+│      Active Decode Throughput Per User │     25.42 │   0.00 │  24,742.90 │     127.78 │      52.20 │      6.47 │     49.41 │
+│                      (tokens/sec/user) │           │        │            │            │            │           │           │
+│     Active Prefill Throughput Per User │ 26,316.38 │ 421.33 │ 109,836.09 │  91,734.79 │  57,126.36 │ 20,993.12 │ 20,978.24 │
+│                      (tokens/sec/user) │           │        │            │            │            │           │           │
+│   Active Total Throughput (tokens/sec) │ 49,425.04 │  60.16 │ 342,154.15 │ 201,893.82 │ 115,295.03 │ 39,885.87 │ 48,575.02 │
+└────────────────────────────────────────┴───────────┴────────┴────────────┴────────────┴────────────┴───────────┴───────────┘
+                                                NVIDIA AIPerf | LLM Metrics: Usage                                                
+┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┓
+┃                                  Metric ┃           avg ┃      min ┃       max ┃       p99 ┃       p90 ┃       p50 ┃       std ┃
+┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━┩
+│            Usage Prompt Tokens (tokens) │     24,458.55 │ 8,085.00 │ 53,769.00 │ 49,306.60 │ 40,353.00 │ 23,127.00 │ 10,898.51 │
+│ Usage Prompt Cache Read Tokens (tokens) │     23,677.33 │ 7,744.00 │ 53,312.00 │ 48,816.64 │ 39,398.40 │ 22,464.00 │ 10,926.98 │
+│        Usage Completion Tokens (tokens) │        372.01 │    30.00 │  2,286.00 │  1,708.48 │  1,119.80 │    142.00 │    444.56 │
+│             Usage Total Tokens (tokens) │     24,830.56 │ 8,151.00 │ 55,332.00 │ 49,647.76 │ 40,813.60 │ 23,476.00 │ 11,031.19 │
+│      Total Usage Prompt Tokens (tokens) │ 44,832,528.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│    Total Usage Prompt Cache Read Tokens │ 43,211,136.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│                                (tokens) │               │          │           │           │           │           │           │
+│       Overall Usage Prompt Cache Read % │         96.38 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│                                     (%) │               │          │           │           │           │           │           │
+│  Total Usage Completion Tokens (tokens) │    681,890.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│       Total Usage Total Tokens (tokens) │ 45,514,418.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+└─────────────────────────────────────────┴───────────────┴──────────┴───────────┴───────────┴───────────┴───────────┴───────────┘
+                                                    NVIDIA AIPerf | LLM Metrics                                                     
+┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┓
+┃                                        Metric ┃       avg ┃      min ┃       max ┃       p99 ┃       p90 ┃       p50 ┃       std ┃
+┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━┩
+│                      Time to First Token (ms) │  1,077.26 │   327.87 │ 23,117.39 │  7,327.97 │  1,981.57 │    688.48 │  1,422.87 │
+│                     Time to Second Token (ms) │    100.35 │     0.00 │  1,192.87 │    670.46 │    327.15 │     36.29 │    165.68 │
+│               Time to First Output Token (ms) │  7,077.20 │   359.05 │ 77,151.91 │ 51,945.21 │ 23,607.91 │  2,167.23 │ 11,407.64 │
+│                          Request Latency (ms) │ 15,769.31 │   902.28 │ 93,147.30 │ 67,595.36 │ 43,565.18 │  7,860.28 │ 17,025.99 │
+│                      Inter Token Latency (ms) │     42.87 │     9.41 │    134.79 │    116.65 │     62.79 │     38.08 │     18.59 │
+│              Output Token Throughput Per User │     26.53 │     7.42 │    106.30 │     57.90 │     35.67 │     26.26 │      8.98 │
+│                             (tokens/sec/user) │           │          │           │           │           │           │           │
+│ E2E Output Token Throughput (tokens/sec/user) │     22.46 │     4.66 │     63.42 │     38.51 │     29.40 │     23.69 │      6.96 │
+│               Output Sequence Length (tokens) │    372.01 │    30.00 │  2,286.00 │  1,708.48 │  1,119.80 │    142.00 │    444.56 │
+│                Input Sequence Length (tokens) │ 24,458.55 │ 8,085.00 │ 53,769.00 │ 49,306.60 │ 40,353.00 │ 23,127.00 │ 10,898.51 │
+│          Output Token Throughput (tokens/sec) │    740.49 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│           Input Token Throughput (tokens/sec) │ 48,685.42 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│             Request Throughput (requests/sec) │      1.99 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│                      Request Count (requests) │  1,833.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+└───────────────────────────────────────────────┴───────────┴──────────┴───────────┴───────────┴───────────┴───────────┴───────────┘
+
+
+No GPU telemetry data collected during the benchmarking run.
+
+╭───────────────────────────────── Output Sequence Length Mismatch Warning ─────────────────────────────────╮
+│  1,629 of 1,833 requests (88.9%) have output length differing from requested by more than the threshold.  │
+│  Threshold (tokens): min(requested x 5%, 50)                                                              │
+│  Average mismatch: -71.4%                                                                                 │
+│                                                                                                           │
+│  Why: Server hit EOS token before reaching requested output length.                                       │
+│                                                                                                           │
+│  Fix Options:                                                                                             │
+│    - --extra-inputs ignore_eos:true - Generate until max_tokens (vLLM, TensorRT-LLM)                      │
+│    - --extra-inputs min_tokens:<N> - Set minimum output length (vLLM, TensorRT-LLM, SGLang)               │
+│    - --use-server-token-count - Use server-reported token counts if tokenizer mismatch suspected          │
+│                                                                                                           │
+│  Diagnostics:                                                                                             │
+│    - Review profile_export.jsonl -> osl_mismatch_diff_pct for per-request values                          │
+│    - Adjust: AIPERF_METRICS_OSL_MISMATCH_PCT_THRESHOLD=5                                                  │
+│    - Adjust: AIPERF_METRICS_OSL_MISMATCH_MAX_TOKEN_THRESHOLD=50                                           │
+╰───────────────────────────────────────────────────────────────────────────────────────────────────────────╯
+
+CLI Command: aiperf profile --model 'glm-5.2' --url 'http://127.0.0.1:8391' --api-key '<redacted>' --endpoint-type 'chat' --streaming --tokenizer 'builtin' --custom-dataset-type 
+'dag_jsonl' --input-file 'data/capacity/corpus.jsonl' --extra-inputs '{"cache_salt": "a66726cf3754197730e7c940182d5edb"}' --concurrency 36 --warmup-duration 120 --benchmark-duration 900 
+--benchmark-grace-period 300 --request-count 1000000 --random-seed 0 --use-server-token-count --artifact-dir 'results/capacity/sweep-hammer-jul20/N0036' --export-level 'raw' 
+--export-http-trace --slice-duration 60 --server-metrics-formats 'json' 'csv' 'jsonl' --ui 'simple'
+Benchmark Duration: 920.84 sec
+CSV Export: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0036/profile_export_aiperf.csv
+JSON Export: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0036/profile_export_aiperf.json
+Raw Records: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0036/profile_export_raw.jsonl
+Server Metrics CSV Export: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0036/server_metrics_export.csv
+Server Metrics JSON Export: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0036/server_metrics_export.json
+Timeslice CSV Export: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0036/profile_export_aiperf_timeslices.csv
+Timeslice JSON Export: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0036/profile_export_aiperf_timeslices.json
+Log File: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0036/logs/aiperf.log
+
+
+Warmup:   0% |          | [00:00<?]
+Warmup:   0% |          | [00:00<?]
+                                   
+
+Warmup:   2% |▏         | [00:01<01:57]
+Warmup:   2% |▏         | [00:01<01:57]
+Warmup:   3% |▎         | [00:03<01:55]
+Warmup:   3% |▎         | [00:03<01:55]
+Warmup:   5% |▌         | [00:05<01:53]
+Warmup:   5% |▌         | [00:05<01:53]
+Warmup:   7% |▋         | [00:07<01:51]
+Warmup:   7% |▋         | [00:07<01:51]
+Warmup:   8% |▊         | [00:09<01:49]
+Warmup:   8% |▊         | [00:09<01:49]
+Warmup:  10% |█         | [00:11<01:47]
+Warmup:  10% |█         | [00:11<01:47]
+Warmup:  12% |█▏        | [00:13<01:45]
+Warmup:  12% |█▏        | [00:13<01:45]
+Warmup:  13% |█▎        | [00:15<01:43]
+Warmup:  13% |█▎        | [00:15<01:43]
+Warmup:  15% |█▌        | [00:17<01:41]
+Warmup:  15% |█▌        | [00:17<01:41]
+Warmup:  17% |█▋        | [00:19<01:39]
+Warmup:  17% |█▋        | [00:19<01:39]
+Warmup:  18% |█▊        | [00:21<01:37]
+Warmup:  18% |█▊        | [00:21<01:37]
+Warmup:  20% |██        | [00:23<01:35]
+Warmup:  20% |██        | [00:23<01:35]
+Warmup:  22% |██▏       | [00:25<01:33]
+Warmup:  22% |██▏       | [00:25<01:33]
+Warmup:  23% |██▎       | [00:27<01:31]
+Warmup:  23% |██▎       | [00:27<01:31]
+Warmup:  25% |██▌       | [00:29<01:29]
+Warmup:  25% |██▌       | [00:29<01:29]
+Warmup:  27% |██▋       | [00:31<01:27]
+Warmup:  27% |██▋       | [00:31<01:27]
+Warmup:  28% |██▊       | [00:33<01:25]
+Warmup:  28% |██▊       | [00:33<01:25]
+Warmup:  30% |███       | [00:35<01:23]
+Warmup:  30% |███       | [00:35<01:23]
+Warmup:  32% |███▏      | [00:37<01:21]
+Warmup:  32% |███▏      | [00:37<01:21]
+Warmup:  33% |███▎      | [00:39<01:19]
+Warmup:  33% |███▎      | [00:39<01:19]
+Warmup:  35% |███▌      | [00:41<01:17]
+Warmup:  35% |███▌      | [00:41<01:17]
+Warmup:  37% |███▋      | [00:43<01:15]
+Warmup:  37% |███▋      | [00:43<01:15]
+Warmup:  38% |███▊      | [00:45<01:13]
+Warmup:  38% |███▊      | [00:45<01:13]
+Warmup:  40% |████      | [00:47<01:11]
+Warmup:  40% |████      | [00:47<01:11]
+Warmup:  42% |████▏     | [00:49<01:09]
+Warmup:  42% |████▏     | [00:49<01:09]
+Warmup:  43% |████▎     | [00:51<01:07]
+Warmup:  43% |████▎     | [00:51<01:07]
+Warmup:  45% |████▌     | [00:53<01:05]
+Warmup:  45% |████▌     | [00:53<01:05]
+Warmup:  47% |████▋     | [00:55<01:03]
+Warmup:  47% |████▋     | [00:55<01:03]
+Warmup:  48% |████▊     | [00:57<01:01]
+Warmup:  48% |████▊     | [00:57<01:01]
+Warmup:  50% |█████     | [00:59<00:59]
+Warmup:  50% |█████     | [00:59<00:59]
+Warmup:  52% |█████▏    | [01:01<00:57]
+Warmup:  52% |█████▏    | [01:01<00:57]
+Warmup:  53% |█████▎    | [01:03<00:55]
+Warmup:  53% |█████▎    | [01:03<00:55]
+Warmup:  55% |█████▌    | [01:05<00:53]
+Warmup:  55% |█████▌    | [01:05<00:53]
+Warmup:  57% |█████▋    | [01:07<00:51]
+Warmup:  57% |█████▋    | [01:07<00:51]
+Warmup:  58% |█████▊    | [01:09<00:49]
+Warmup:  58% |█████▊    | [01:09<00:49]
+Warmup:  60% |██████    | [01:11<00:47]
+Warmup:  60% |██████    | [01:11<00:47]
+Warmup:  62% |██████▏   | [01:13<00:45]
+Warmup:  62% |██████▏   | [01:13<00:45]
+Warmup:  63% |██████▎   | [01:15<00:43]
+Warmup:  63% |██████▎   | [01:15<00:43]
+Warmup:  65% |██████▌   | [01:17<00:41]
+Warmup:  65% |██████▌   | [01:17<00:41]
+Warmup:  67% |██████▋   | [01:19<00:39]
+Warmup:  67% |██████▋   | [01:19<00:39]
+Warmup:  68% |██████▊   | [01:21<00:37]
+Warmup:  68% |██████▊   | [01:21<00:37]
+Warmup:  70% |███████   | [01:23<00:35]
+Warmup:  70% |███████   | [01:23<00:35]
+Warmup:  72% |███████▏  | [01:25<00:33]
+Warmup:  72% |███████▏  | [01:25<00:33]
+Warmup:  73% |███████▎  | [01:27<00:31]
+Warmup:  73% |███████▎  | [01:27<00:31]
+Warmup:  75% |███████▌  | [01:29<00:29]
+Warmup:  75% |███████▌  | [01:29<00:29]
+Warmup:  77% |███████▋  | [01:31<00:27]
+Warmup:  77% |███████▋  | [01:31<00:27]
+Warmup:  78% |███████▊  | [01:33<00:25]
+Warmup:  78% |███████▊  | [01:33<00:25]
+Warmup:  80% |████████  | [01:35<00:23]
+Warmup:  80% |████████  | [01:35<00:23]
+Warmup:  82% |████████▏ | [01:37<00:21]
+Warmup:  82% |████████▏ | [01:37<00:21]
+Warmup:  83% |████████▎ | [01:39<00:19]
+Warmup:  83% |████████▎ | [01:39<00:19]
+Warmup:  85% |████████▌ | [01:41<00:17]
+Warmup:  85% |████████▌ | [01:41<00:17]
+Warmup:  87% |████████▋ | [01:43<00:15]
+Warmup:  87% |████████▋ | [01:43<00:15]
+Warmup:  88% |████████▊ | [01:45<00:13]
+Warmup:  88% |████████▊ | [01:45<00:13]
+Warmup:  90% |█████████ | [01:47<00:11]
+Warmup:  90% |█████████ | [01:47<00:11]
+Warmup:  92% |█████████▏| [01:49<00:09]
+Warmup:  92% |█████████▏| [01:49<00:09]
+Warmup:  93% |█████████▎| [01:51<00:07]
+Warmup:  93% |█████████▎| [01:51<00:07]
+Warmup:  95% |█████████▌| [01:53<00:05]
+Warmup:  95% |█████████▌| [01:53<00:05]
+Warmup:  97% |█████████▋| [01:55<00:03]
+Warmup:  97% |█████████▋| [01:55<00:03]
+Warmup:  98% |█████████▊| [01:57<00:01]
+Warmup:  98% |█████████▊| [01:57<00:01]
+Warmup Grace Period: 0/382 |          |   0% [00:00<?]
+Warmup Grace Period: 351/382 |█████████▏|  92% [00:00<00:00]
+Warmup Grace Period: 357/382 |█████████▎|  93% [00:02<00:00]
+Warmup Grace Period: 357/382 |█████████▎|  93% [00:02<00:00]
+Warmup Grace Period: 364/382 |█████████▌|  95% [00:04<00:00]
+Warmup Grace Period: 368/382 |█████████▋|  96% [00:06<00:00]
+Warmup Grace Period: 373/382 |█████████▊|  98% [00:08<00:00]
+Warmup Grace Period: 377/382 |█████████▊|  99% [00:14<00:00]
+Warmup Grace Period: 377/382 |█████████▊|  99% [00:14<00:00]
+Warmup Grace Period: 381/382 |█████████▉| 100% [00:20<00:00]
+Warmup Grace Period: 381/382 |█████████▉| 100% [00:20<00:00]
+Warmup Grace Period: 381/382 |█████████▉| 100% [00:30<00:00]
+Processing Records: 0/382 |          |   0% [00:00<?]
+Profiling: 0/1,000,000 |          |   0% [00:00<?]
+Profiling Grace Period: 0/1,833 |          |   0% [00:00<?]
+Profiling Grace Period: 1,802/1,833 |█████████▊|  98% [00:00<00:00]
+Profiling Grace Period: 1,821/1,833 |█████████▉|  99% [00:05<00:00]
+Profiling Grace Period: 1,821/1,833 |█████████▉|  99% [00:05<00:00]
+Profiling Grace Period: 1,821/1,833 |█████████▉|  99% [00:17<00:00]
+Processing Records: 0/1,833 |          |   0% [00:00<?]
+Processing Records: 1,833/1,833 |██████████| 100% [00:00<00:00]
+Processing Records: 1,833/1,833 |██████████| 100% [00:00<00:00]

experiments/kv-cache-fp8/capacity/hammer/N0036/manifest.json

@@ -0,0 +1,111 @@
+{
+  "concurrency": 36,
+  "mode": "hammer",
+  "cache_salt": "a66726cf3754197730e7c940182d5edb",
+  "command": [
+    "aiperf",
+    "profile",
+    "--model",
+    "glm-5.2",
+    "--url",
+    "http://127.0.0.1:8391",
+    "--api-key",
+    "<key>",
+    "--endpoint-type",
+    "chat",
+    "--streaming",
+    "--tokenizer",
+    "builtin",
+    "--custom-dataset-type",
+    "dag_jsonl",
+    "--input-file",
+    "data/capacity/corpus.jsonl",
+    "--extra-inputs",
+    "{\"cache_salt\": \"a66726cf3754197730e7c940182d5edb\"}",
+    "--concurrency",
+    "36",
+    "--warmup-duration",
+    "120",
+    "--benchmark-duration",
+    "900",
+    "--benchmark-grace-period",
+    "300",
+    "--request-count",
+    "1000000",
+    "--random-seed",
+    "0",
+    "--use-server-token-count",
+    "--artifact-dir",
+    "results/capacity/sweep-hammer-jul20/N0036",
+    "--export-level",
+    "raw",
+    "--export-http-trace",
+    "--slice-duration",
+    "60",
+    "--server-metrics-formats",
+    "json",
+    "csv",
+    "jsonl",
+    "--ui",
+    "simple"
+  ],
+  "dataset_path": "data/capacity/corpus.jsonl",
+  "signature": {
+    "dataset_sha256": "0327d658a267715aebdab45b67b6e0d324616a0faeec8a75d09b8defcd0b2ac2",
+    "url": "http://127.0.0.1:8391",
+    "model": "glm-5.2",
+    "duration": 900,
+    "warmup": 120,
+    "grace": 300,
+    "seed": 0,
+    "request_ceiling": 1000000,
+    "python": "3.12.3",
+    "aiperf": {
+      "version": "0.11.0",
+      "rev": "73d0524f98f30199ca659a8ab586454a823c887d"
+    },
+    "server": {
+      "version": "0.5.15.post1",
+      "model_path": "/home/jl_fs/models/GLM-5.2-FP8",
+      "served_model_name": "glm-5.2",
+      "tp_size": 8,
+      "dp_size": 8,
+      "enable_cache_report": true,
+      "speculative_algorithm": "EAGLE",
+      "max_running_requests": 48,
+      "context_length": 262144,
+      "kv_cache_dtype": "fp8_e4m3",
+      "mem_fraction_static": 0.872,
+      "scheduler_pids": [
+        82149,
+        82150,
+        82151,
+        82152,
+        82153,
+        82154,
+        82155,
+        82156
+      ],
+      "max_total_num_tokens": 234176,
+      "max_model_len": 262144,
+      "model_id": "glm-5.2"
+    },
+    "evaluator_sha256": "3e48dfede43cbf47244f560547eda95d29c671d4fa23b203a5d220a14edf1c80",
+    "verifier_sha256": "6f84bee84368a72ea1445d0961185a0d5d2fd6ba8f8ec4c3fefac619b46a32aa",
+    "runner_sha256": "aa3e1a0f66bcfac6663b5d2e59a37cae9d7264dda6b4c13130a3edf403e77104"
+  },
+  "started_utc": "2026-07-21T00:59:32Z",
+  "loadavg_start": [
+    1.18310546875,
+    1.52099609375,
+    1.60693359375
+  ],
+  "status": "ok",
+  "elapsed_s": 1089.1,
+  "loadavg_end": [
+    1.79052734375,
+    1.81201171875,
+    1.73583984375
+  ],
+  "returncode": 0
+}

experiments/kv-cache-fp8/capacity/hammer/N0036/profile_export_aiperf.csv

@@ -0,0 +1,68 @@
+Metric,avg,min,max,sum,p1,p5,p10,p25,p50,p75,p90,p95,p99,std
+Active Decode Throughput (tokens/sec),740.87,0.00,103688.33,,,,,,196.70,,1643.35,1723.21,3778.06,1345.24
+Active Decode Throughput Per User (tokens/sec/user),25.42,0.00,24742.90,,,,,,6.47,,52.20,55.28,127.78,49.41
+Active Prefill Throughput (tokens/sec),60858.36,421.33,341378.40,,,,,,54718.68,,121240.64,147287.06,207284.22,47009.73
+Active Prefill Throughput Per User (tokens/sec/user),26316.38,421.33,109836.09,,,,,,20993.12,,57126.36,68612.05,91734.79,20978.24
+Active Total Throughput (tokens/sec),49425.04,60.16,342154.15,,,,,,39885.87,,115295.03,139091.13,201893.82,48575.02
+Credit-to-Start Latency (ms),1.69,0.48,19.68,3090.26,0.94,1.07,1.13,1.24,1.39,1.57,1.84,2.19,12.13,1.69
+E2E Output Token Throughput (tokens/sec/user),22.46,4.66,63.42,41163.43,7.17,9.36,11.17,18.73,23.69,26.83,29.40,31.35,38.51,6.96
+Effective Concurrency,31.47,0.00,36.00,,,,,,32.00,,35.00,36.00,36.00,4.33
+Effective Decode Concurrency,29.32,0.00,35.00,,,,,,30.00,,33.00,33.00,34.00,4.33
+Effective Decode Throughput (tokens/sec),738.50,0.00,103688.33,,,,,,196.14,,1643.03,1723.05,3765.82,1343.74
+Effective Decode Throughput Per User (tokens/sec/user),25.34,0.00,24742.90,,,,,,6.45,,52.19,55.24,127.65,49.35
+Effective Latency (CO-aware) (ms),15809.56,940.19,93149.09,28978930.35,1477.58,2136.61,2764.94,4264.71,7883.86,20931.09,43665.83,51696.34,67596.82,17014.83
+Effective Prefill Concurrency,2.14,0.00,36.00,,,,,,2.00,,4.00,5.00,7.00,2.65
+Effective Prefill Throughput (tokens/sec),48685.42,0.00,341378.40,,,,,,39133.32,,114454.82,138332.35,201085.20,48585.29
+Effective Prefill Throughput Per User (tokens/sec/user),21052.55,0.00,109836.09,,,,,,15019.37,,52205.55,64587.88,87080.67,21514.58
+Effective Total Throughput (tokens/sec),49423.92,0.00,342154.15,,,,,,39885.87,,115295.03,139091.13,201893.82,48575.04
+HTTP Blocked (ms),0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00
+HTTP Chunks Received,199.69,16.00,1216.00,366028.00,20.00,27.00,31.00,43.00,77.00,271.00,604.80,707.20,914.40,238.01
+HTTP Chunks Sent,1.00,1.00,1.00,1833.00,1.00,1.00,1.00,1.00,1.00,1.00,1.00,1.00,1.00,0.00
+HTTP Connecting (ms),0.09,0.00,10.60,162.55,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,2.59,0.69
+HTTP Connection Overhead (ms),0.09,0.00,10.60,162.55,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,2.59,0.69
+HTTP Connection Reused (ratio),0.97,0.00,1.00,1780.00,0.00,1.00,1.00,1.00,1.00,1.00,1.00,1.00,1.00,0.17
+HTTP Data Received (KB),49.01,4.31,289.06,89829.87,5.31,7.31,8.45,11.22,19.66,65.86,145.74,169.90,217.27,57.06
+HTTP Data Sent (KB),101.90,36.74,224.62,186785.61,37.59,41.42,45.24,64.78,95.62,136.09,167.25,182.04,204.25,44.80
+HTTP DNS Lookup (ms),0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00
+HTTP Duration (excl. conn) (ms),15807.74,938.73,93147.12,28975581.28,1469.84,2135.06,2763.53,4263.53,7882.24,20929.90,43664.23,51694.79,67595.22,17014.84
+HTTP Receiving (ms),14768.08,534.17,92360.82,27069886.28,920.12,1515.83,2049.81,3472.78,6603.17,19859.32,42807.80,50808.91,67045.99,16950.62
+HTTP Sending (ms),0.36,0.09,11.06,666.56,0.19,0.20,0.21,0.23,0.26,0.31,0.38,0.46,2.73,0.70
+HTTP Total Time (ms),15807.83,938.73,93147.12,28975743.83,1469.84,2135.06,2763.53,4263.53,7882.24,20929.90,43664.23,51694.79,67595.22,17014.82
+HTTP Waiting (TTFB) (ms),1039.30,324.61,23117.10,1905028.44,355.54,376.91,395.42,468.29,663.85,894.31,1972.96,3547.34,7108.91,1417.91
+Input Sequence Length (tokens),24458.55,8085.00,53769.00,44832528.00,8269.12,9343.40,10456.00,15545.00,23127.00,32590.00,40353.00,43615.80,49306.60,10898.51
+Inter Chunk Latency (ms),74.15,0.00,1768.95,26930534.92,1.64,2.52,3.80,34.60,36.48,42.38,274.39,320.90,610.12,125.49
+Inter Token Latency (ms),42.87,9.41,134.79,78578.08,17.27,23.63,28.04,33.67,38.08,44.98,62.79,88.90,116.65,18.59
+OSL Mismatch Diff % (%),-71.38,-98.11,0.00,-130845.78,-96.91,-95.69,-94.76,-93.02,-88.05,-61.41,0.00,0.00,0.00,32.11
+Output Sequence Length (tokens),372.01,30.00,2286.00,681890.00,37.32,52.00,61.00,81.00,142.00,497.00,1119.80,1314.80,1708.48,444.56
+Output Token Count (tokens),372.01,30.00,2286.00,681890.00,37.32,52.00,61.00,81.00,142.00,497.00,1119.80,1314.80,1708.48,444.56
+Output Token Throughput Per User (tokens/sec/user),26.53,7.42,106.30,48624.60,8.57,11.25,15.93,22.23,26.26,29.70,35.67,42.32,57.90,8.98
+Prefill Throughput Per User (tokens/sec/user),36862.55,421.33,119600.62,67569045.45,2058.51,4206.04,9117.51,17891.45,34137.23,52341.73,69050.92,81390.52,100448.24,23424.18
+Request Latency (ms),15769.31,902.28,93147.30,28905152.67,1441.76,2097.20,2733.78,4239.28,7860.28,20930.01,43565.18,51692.06,67595.36,17025.99
+Time to First Output Token (ms),7077.20,359.05,77151.91,11896771.88,399.76,429.88,485.82,800.67,2167.23,6773.50,23607.91,34731.04,51945.21,11407.64
+Time to First Token (ms),1077.26,327.87,23117.39,1974617.75,357.39,384.93,406.98,481.87,688.48,957.58,1981.57,3552.35,7327.97,1422.87
+Time to Second Token (ms),100.35,0.00,1192.87,183941.77,0.11,1.30,2.44,32.34,36.29,58.00,327.15,528.16,670.46,165.68
+Tokens In Flight (tokens),861160.37,0.00,1269245.62,,,,,,894248.97,,1164760.30,1188237.67,1230231.58,259402.48
+Usage Completion Tokens (tokens),372.01,30.00,2286.00,681890.00,37.32,52.00,61.00,81.00,142.00,497.00,1119.80,1314.80,1708.48,444.56
+Usage Prompt Cache Read Tokens (tokens),23677.33,7744.00,53312.00,43211136.00,7744.00,8384.00,9920.00,14528.00,22464.00,31872.00,39398.40,43059.20,48816.64,10926.98
+Usage Prompt Tokens (tokens),24458.55,8085.00,53769.00,44832528.00,8269.12,9343.40,10456.00,15545.00,23127.00,32590.00,40353.00,43615.80,49306.60,10898.51
+Usage Total Tokens (tokens),24830.56,8151.00,55332.00,45514418.00,8343.80,9410.40,10572.20,15792.00,23476.00,33087.00,40813.60,44097.20,49647.76,11031.19
+
+Metric,Value
+Benchmark Duration (sec),920.84
+Input Token Throughput (tokens/sec),48685.42
+OSL Mismatch Count,1629.00
+Output Token Throughput (tokens/sec),740.49
+Overall Usage Prompt Cache Read % (%),96.38
+Request Count,1833.00
+Request Throughput (requests/sec),1.99
+Total Input Sequence Length (tokens),44832528.00
+Total Output Sequence Length (tokens),681890.00
+Total Output Tokens (tokens),681890.00
+Total Token Throughput (tokens/sec),49425.91
+Total Usage Completion Tokens (tokens),681890.00
+Total Usage Prompt Cache Read Tokens (tokens),43211136.00
+Total Usage Prompt Tokens (tokens),44832528.00
+Total Usage Total Tokens (tokens),45514418.00
+
+
+Endpoint,GPU_Index,GPU_Name,GPU_UUID,Metric,avg,min,max,sum,p1,p5,p10,p25,p50,p75,p90,p95,p99,std

experiments/kv-cache-fp8/capacity/hammer/N0036/profile_export_aiperf.json

@@ -0,0 +1,1726 @@
+{
+  "schema_version": "1.4",
+  "aiperf_version": "0.11.0",
+  "benchmark_id": "098c77ae5139",
+  "request_throughput": {
+    "unit": "requests/sec",
+    "avg": 1.9905271936236324
+  },
+  "request_latency": {
+    "unit": "ms",
+    "avg": 15769.314056669939,
+    "p1": 1441.755021,
+    "p5": 2097.1977734,
+    "p10": 2733.7756556,
+    "p25": 4239.280417999999,
+    "p50": 7860.280895,
+    "p75": 20930.014814,
+    "p90": 43565.182013599995,
+    "p95": 51692.06049019997,
+    "p99": 67595.35532172001,
+    "min": 902.2796609999999,
+    "max": 93147.29717199999,
+    "std": 17025.994024591906,
+    "count": 1833,
+    "sum": 28905152.665875997
+  },
+  "request_count": {
+    "unit": "requests",
+    "avg": 1833.0
+  },
+  "time_to_first_token": {
+    "unit": "ms",
+    "avg": 1077.2600919787233,
+    "p1": 357.39456712,
+    "p5": 384.92517599999996,
+    "p10": 406.9814218,
+    "p25": 481.869221,
+    "p50": 688.47692,
+    "p75": 957.580464,
+    "p90": 1981.5728491999996,
+    "p95": 3552.3474893999996,
+    "p99": 7327.969477400008,
+    "min": 327.86751699999996,
+    "max": 23117.391744,
+    "std": 1422.8695957370228,
+    "count": 1833,
+    "sum": 1974617.7485969998
+  },
+  "time_to_second_token": {
+    "unit": "ms",
+    "avg": 100.35011845117295,
+    "p1": 0.11365539999999999,
+    "p5": 1.3047138,
+    "p10": 2.4356268,
+    "p25": 32.344072,
+    "p50": 36.294849,
+    "p75": 58.001981,
+    "p90": 327.149249,
+    "p95": 528.162649,
+    "p99": 670.4603927600001,
+    "min": 0.0,
+    "max": 1192.865088,
+    "std": 165.67630653807336,
+    "count": 1833,
+    "sum": 183941.76712099998
+  },
+  "inter_token_latency": {
+    "unit": "ms",
+    "avg": 42.86856385541416,
+    "p1": 17.269983464939422,
+    "p5": 23.627729548122723,
+    "p10": 28.036997852080773,
+    "p25": 33.66676564402174,
+    "p50": 38.077614775510206,
+    "p75": 44.98051880733945,
+    "p90": 62.79358821196765,
+    "p95": 88.90447680505147,
+    "p99": 116.65273318918248,
+    "min": 9.40706044117647,
+    "max": 134.79140067741935,
+    "std": 18.59016234798549,
+    "count": 1833,
+    "sum": 78578.07754697415
+  },
+  "output_token_throughput": {
+    "unit": "tokens/sec",
+    "avg": 740.4913191816796
+  },
+  "output_token_throughput_per_user": {
+    "unit": "tokens/sec/user",
+    "avg": 26.527331141120772,
+    "p1": 8.57280374033913,
+    "p5": 11.248042106901575,
+    "p10": 15.92520344600637,
+    "p25": 22.231846730874754,
+    "p50": 26.262149189112407,
+    "p75": 29.70288297288728,
+    "p90": 35.667157196662224,
+    "p95": 42.323153682650705,
+    "p99": 57.90395884042379,
+    "min": 7.418870899584937,
+    "max": 106.30313329579687,
+    "std": 8.977475465600003,
+    "count": 1833,
+    "sum": 48624.59798167438
+  },
+  "output_sequence_length": {
+    "unit": "tokens",
+    "avg": 372.0076377523186,
+    "p1": 37.32,
+    "p5": 52.0,
+    "p10": 61.0,
+    "p25": 81.0,
+    "p50": 142.0,
+    "p75": 497.0,
+    "p90": 1119.8,
+    "p95": 1314.799999999999,
+    "p99": 1708.4800000000007,
+    "min": 30.0,
+    "max": 2286.0,
+    "std": 444.5551924008996,
+    "count": 1833,
+    "sum": 681890.0
+  },
+  "input_sequence_length": {
+    "unit": "tokens",
+    "avg": 24458.55319148936,
+    "p1": 8269.12,
+    "p5": 9343.400000000001,
+    "p10": 10456.0,
+    "p25": 15545.0,
+    "p50": 23127.0,
+    "p75": 32590.0,
+    "p90": 40352.99999999999,
+    "p95": 43615.799999999996,
+    "p99": 49306.60000000001,
+    "min": 8085.0,
+    "max": 53769.0,
+    "std": 10898.511761819209,
+    "count": 1833,
+    "sum": 44832528.0
+  },
+  "output_token_count": {
+    "unit": "tokens",
+    "avg": 372.0076377523186,
+    "p1": 37.32,
+    "p5": 52.0,
+    "p10": 61.0,
+    "p25": 81.0,
+    "p50": 142.0,
+    "p75": 497.0,
+    "p90": 1119.8,
+    "p95": 1314.799999999999,
+    "p99": 1708.4800000000007,
+    "min": 30.0,
+    "max": 2286.0,
+    "std": 444.5551924008996,
+    "count": 1833,
+    "sum": 681890.0
+  },
+  "inter_chunk_latency": {
+    "unit": "ms",
+    "avg": 74.152644350493,
+    "p1": 1.64218188,
+    "p5": 2.5163493999999997,
+    "p10": 3.803649599999999,
+    "p25": 34.597128,
+    "p50": 36.481742,
+    "p75": 42.384133999999996,
+    "p90": 274.38537600000006,
+    "p95": 320.9018818,
+    "p99": 610.1216108799999,
+    "min": 0.0,
+    "max": 1768.948034,
+    "std": 125.49221517987243,
+    "count": 363177,
+    "sum": 26930534.917278998
+  },
+  "total_output_tokens": {
+    "unit": "tokens",
+    "avg": 681890.0
+  },
+  "benchmark_duration": {
+    "unit": "sec",
+    "avg": 920.8407121920001
+  },
+  "total_isl": {
+    "unit": "tokens",
+    "avg": 44832528.0
+  },
+  "total_osl": {
+    "unit": "tokens",
+    "avg": 681890.0
+  },
+  "telemetry_data": {
+    "summary": {
+      "endpoints_configured": [],
+      "endpoints_successful": [],
+      "start_time": "2026-07-21T01:02:11.910776",
+      "end_time": "2026-07-21T01:17:36.107075"
+    },
+    "endpoints": {},
+    "error_summary": []
+  },
+  "input_config": {
+    "models": {
+      "items": [
+        {
+          "name": "glm-5.2"
+        }
+      ]
+    },
+    "endpoint": {
+      "urls": [
+        "http://127.0.0.1:8391"
+      ],
+      "type": "chat",
+      "api_key": "<redacted>",
+      "streaming": true,
+      "use_server_token_count": true,
+      "extra": {
+        "cache_salt": "a66726cf3754197730e7c940182d5edb"
+      }
+    },
+    "datasets": [
+      {
+        "name": "main",
+        "type": "file",
+        "path": "data/capacity/corpus.jsonl",
+        "format": "dag_jsonl",
+        "entries": 1000000,
+        "random_seed": 0
+      }
+    ],
+    "phases": [
+      {
+        "name": "warmup",
+        "type": "concurrency",
+        "exclude_from_results": true,
+        "duration": 120.0,
+        "concurrency": 36
+      },
+      {
+        "name": "profiling",
+        "type": "concurrency",
+        "requests": 1000000,
+        "duration": 900.0,
+        "concurrency": 36,
+        "grace_period": 300.0
+      }
+    ],
+    "artifacts": {
+      "dir": "/home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0036",
+      "records": [
+        "jsonl"
+      ],
+      "raw": true,
+      "trace": true,
+      "slice_duration": 60.0,
+      "auto_plot": false,
+      "plot_required": false
+    },
+    "tokenizer": {
+      "name": "builtin"
+    },
+    "gpu_telemetry": {
+      "enabled": true
+    },
+    "server_metrics": {
+      "enabled": true,
+      "urls": [],
+      "formats": [
+        "json",
+        "csv",
+        "jsonl"
+      ]
+    },
+    "network_latency": {
+      "enabled": false
+    },
+    "otel": {},
+    "mlflow": {},
+    "wandb": {},
+    "runtime": {
+      "ui": "simple"
+    }
+  },
+  "run_info": {
+    "benchmark_id": "098c77ae5139",
+    "random_seed": 0,
+    "trial": 0,
+    "cli_command": "aiperf profile --model 'glm-5.2' --url 'http://127.0.0.1:8391' --api-key '<redacted>' --endpoint-type 'chat' --streaming --tokenizer 'builtin' --custom-dataset-type 'dag_jsonl' --input-file 'data/capacity/corpus.jsonl' --extra-inputs '{\"cache_salt\": \"a66726cf3754197730e7c940182d5edb\"}' --concurrency 36 --warmup-duration 120 --benchmark-duration 900 --benchmark-grace-period 300 --request-count 1000000 --random-seed 0 --use-server-token-count --artifact-dir 'results/capacity/sweep-hammer-jul20/N0036' --export-level 'raw' --export-http-trace --slice-duration 60 --server-metrics-formats 'json' 'csv' 'jsonl' --ui 'simple'"
+  },
+  "was_cancelled": false,
+  "error_summary": [],
+  "start_time": "2026-07-21T01:02:11.910776",
+  "end_time": "2026-07-21T01:17:32.772348",
+  "warmup_metrics": {
+    "request_latency": {
+      "unit": "ms",
+      "avg": 10961.967085060209,
+      "p1": 1601.236388,
+      "p5": 2831.5084384499996,
+      "p10": 3421.0641208,
+      "p25": 4644.42431575,
+      "p50": 6308.6769779999995,
+      "p75": 10558.454769,
+      "p90": 23725.055832200003,
+      "p95": 41934.440651299956,
+      "p99": 68754.94225290998,
+      "min": 1024.250575,
+      "max": 79463.47651,
+      "std": 12905.665728316419,
+      "count": 382,
+      "sum": 4187471.426493
+    },
+    "time_to_second_token": {
+      "unit": "ms",
+      "avg": 180.67785293717276,
+      "p1": 0.32062355000000003,
+      "p5": 2.1225950000000005,
+      "p10": 11.0339548,
+      "p25": 34.321180999999996,
+      "p50": 40.8235635,
+      "p75": 297.93165375,
+      "p90": 526.8693344999999,
+      "p95": 688.0145784999997,
+      "p99": 1093.07750852,
+      "min": 0.0,
+      "max": 1815.9494009999999,
+      "std": 270.89730107172977,
+      "count": 382,
+      "sum": 69018.939822
+    },
+    "output_sequence_length": {
+      "unit": "tokens",
+      "avg": 219.84031413612564,
+      "p1": 38.480000000000004,
+      "p5": 46.0,
+      "p10": 55.0,
+      "p25": 66.25,
+      "p50": 90.5,
+      "p75": 164.25,
+      "p90": 518.7000000000005,
+      "p95": 1124.0499999999997,
+      "p99": 1616.98,
+      "min": 26.0,
+      "max": 2878.0,
+      "std": 347.9035539878663,
+      "count": 382,
+      "sum": 83979.0
+    },
+    "http_req_chunks_sent": {
+      "unit": "count",
+      "avg": 1.0,
+      "p1": 1.0,
+      "p5": 1.0,
+      "p10": 1.0,
+      "p25": 1.0,
+      "p50": 1.0,
+      "p75": 1.0,
+      "p90": 1.0,
+      "p95": 1.0,
+      "p99": 1.0,
+      "min": 1.0,
+      "max": 1.0,
+      "std": 0.0,
+      "count": 382,
+      "sum": 382.0
+    },
+    "time_to_first_token": {
+      "unit": "ms",
+      "avg": 1079.35016213089,
+      "p1": 345.87024942,
+      "p5": 376.3528694,
+      "p10": 405.5414735,
+      "p25": 563.746896,
+      "p50": 693.438488,
+      "p75": 985.56215075,
+      "p90": 1646.9660999000005,
+      "p95": 4271.4353832,
+      "p99": 4779.892527,
+      "min": 329.46910299999996,
+      "max": 8265.636117,
+      "std": 1115.1191649166426,
+      "count": 382,
+      "sum": 412311.761934
+    },
+    "http_req_data_received": {
+      "unit": "KB",
+      "avg": 29.416255930955497,
+      "p1": 5.427451171875,
+      "p5": 6.319384765625,
+      "p10": 7.5721679687500005,
+      "p25": 9.33837890625,
+      "p50": 12.80126953125,
+      "p75": 22.387451171875,
+      "p90": 66.45800781250001,
+      "p95": 148.540380859375,
+      "p99": 206.029375,
+      "min": 3.8017578125,
+      "max": 358.02734375,
+      "std": 44.894654557351224,
+      "count": 382,
+      "sum": 11237.009765625
+    },
+    "input_sequence_length": {
+      "unit": "tokens",
+      "avg": 13426.429319371728,
+      "p1": 8203.72,
+      "p5": 8437.45,
+      "p10": 8699.3,
+      "p25": 10319.5,
+      "p50": 12627.0,
+      "p75": 16277.0,
+      "p90": 19019.800000000003,
+      "p95": 20537.85,
+      "p99": 22455.94,
+      "min": 8106.0,
+      "max": 24197.0,
+      "std": 3839.6592793019936,
+      "count": 382,
+      "sum": 5128896.0
+    },
+    "http_req_sending": {
+      "unit": "ms",
+      "avg": 0.6373236492146597,
+      "p1": 0.19737692999999998,
+      "p5": 0.20938104999999999,
+      "p10": 0.215393,
+      "p25": 0.23289274999999998,
+      "p50": 0.268194,
+      "p75": 0.32110825,
+      "p90": 0.8689646000000026,
+      "p95": 2.2207378499999995,
+      "p99": 8.338110069999999,
+      "min": 0.132028,
+      "max": 14.169241,
+      "std": 1.4865581905392793,
+      "count": 382,
+      "sum": 243.45763399999998
+    },
+    "http_req_connection_reused": {
+      "unit": "ratio",
+      "avg": 0.8979057591623036,
+      "p1": 0.0,
+      "p5": 0.0,
+      "p10": 0.10000000000000142,
+      "p25": 1.0,
+      "p50": 1.0,
+      "p75": 1.0,
+      "p90": 1.0,
+      "p95": 1.0,
+      "p99": 1.0,
+      "min": 0.0,
+      "max": 1.0,
+      "std": 0.3027722028612779,
+      "count": 382,
+      "sum": 343.0
+    },
+    "http_req_dns_lookup": {
+      "unit": "ms",
+      "avg": 0.0,
+      "p1": 0.0,
+      "p5": 0.0,
+      "p10": 0.0,
+      "p25": 0.0,
+      "p50": 0.0,
+      "p75": 0.0,
+      "p90": 0.0,
+      "p95": 0.0,
+      "p99": 0.0,
+      "min": 0.0,
+      "max": 0.0,
+      "std": 0.0,
+      "count": 382,
+      "sum": 0.0
+    },
+    "http_req_connecting": {
+      "unit": "ms",
+      "avg": 0.35926221465968583,
+      "p1": 0.0,
+      "p5": 0.0,
+      "p10": 0.0,
+      "p25": 0.0,
+      "p50": 0.0,
+      "p75": 0.0,
+      "p90": 0.6244785000000236,
+      "p95": 2.107126899999998,
+      "p99": 7.768330689999999,
+      "min": 0.0,
+      "max": 13.522357999999999,
+      "std": 1.4405627481337169,
+      "count": 382,
+      "sum": 137.238166
+    },
+    "http_req_blocked": {
+      "unit": "ms",
+      "avg": 0.0,
+      "p1": 0.0,
+      "p5": 0.0,
+      "p10": 0.0,
+      "p25": 0.0,
+      "p50": 0.0,
+      "p75": 0.0,
+      "p90": 0.0,
+      "p95": 0.0,
+      "p99": 0.0,
+      "min": 0.0,
+      "max": 0.0,
+      "std": 0.0,
+      "count": 382,
+      "sum": 0.0
+    },
+    "http_req_waiting": {
+      "unit": "ms",
+      "avg": 1041.5014919869109,
+      "p1": 342.55653654,
+      "p5": 362.59700349999997,
+      "p10": 395.88457839999995,
+      "p25": 554.77368475,
+      "p50": 656.0515045,
+      "p75": 905.0306637499999,
+      "p90": 1620.3741292,
+      "p95": 4269.89524685,
+      "p99": 4776.26398385,
+      "min": 329.12118,
+      "max": 8260.712108,
+      "std": 1118.193395727573,
+      "count": 382,
+      "sum": 397853.569939
+    },
+    "http_req_receiving": {
+      "unit": "ms",
+      "avg": 9980.680220863875,
+      "p1": 845.28412939,
+      "p5": 2063.82307985,
+      "p10": 2743.9750733,
+      "p25": 3878.05936425,
+      "p50": 5331.744436,
+      "p75": 9476.4805215,
+      "p90": 23042.461581000018,
+      "p95": 41225.41567709996,
+      "p99": 67762.2219679,
+      "min": 507.297582,
+      "max": 78190.609407,
+      "std": 12829.9639468509,
+      "count": 382,
+      "sum": 3812619.8443699996
+    },
+    "usage_total_tokens": {
+      "unit": "tokens",
+      "avg": 13646.269633507853,
+      "p1": 8283.77,
+      "p5": 8505.35,
+      "p10": 8765.3,
+      "p25": 10394.5,
+      "p50": 12800.0,
+      "p75": 16509.5,
+      "p90": 19206.3,
+      "p95": 21062.6,
+      "p99": 23029.899999999998,
+      "min": 8187.0,
+      "max": 24818.0,
+      "std": 3965.9205764766407,
+      "count": 382,
+      "sum": 5212875.0
+    },
+    "request_count": {
+      "unit": "requests",
+      "avg": 382.0
+    },
+    "http_req_duration": {
+      "unit": "ms",
+      "avg": 11022.8190365,
+      "p1": 1632.63471796,
+      "p5": 2907.28809595,
+      "p10": 3442.1320522,
+      "p25": 4807.008181,
+      "p50": 6354.2891039999995,
+      "p75": 10558.30346875,
+      "p90": 23725.239699400005,
+      "p95": 41934.321728599956,
+      "p99": 68754.80068390998,
+      "min": 1024.145066,
+      "max": 79463.346425,
+      "std": 12892.458125454752,
+      "count": 382,
+      "sum": 4210716.871943
+    },
+    "time_to_first_output_token": {
+      "unit": "ms",
+      "avg": 5542.050525594595,
+      "p1": 385.3025806,
+      "p5": 531.5209125,
+      "p10": 667.1954756,
+      "p25": 1092.0464339999999,
+      "p50": 2213.330363,
+      "p75": 5115.88511675,
+      "p90": 12233.0822461,
+      "p95": 23728.3557382,
+      "p99": 46154.78581368,
+      "min": 369.829075,
+      "max": 59673.10264,
+      "std": 8959.932021875109,
+      "count": 370,
+      "sum": 2050558.69447
+    },
+    "http_req_data_sent": {
+      "unit": "KB",
+      "avg": 56.75916230366492,
+      "p1": 37.296904296875,
+      "p5": 38.290625,
+      "p10": 39.274609375,
+      "p25": 44.789794921875,
+      "p50": 53.34521484375,
+      "p75": 66.79541015625,
+      "p90": 77.3892578125,
+      "p95": 84.718701171875,
+      "p99": 92.87625,
+      "min": 36.841796875,
+      "max": 99.873046875,
+      "std": 14.598115769998916,
+      "count": 382,
+      "sum": 21682.0
+    },
+    "usage_prompt_cache_read_tokens": {
+      "unit": "tokens",
+      "avg": 12554.609625668449,
+      "p1": 7744.0,
+      "p5": 7744.0,
+      "p10": 8211.2,
+      "p25": 9680.0,
+      "p50": 11744.0,
+      "p75": 15296.0,
+      "p90": 17900.8,
+      "p95": 19542.399999999998,
+      "p99": 21905.28,
+      "min": 7744.0,
+      "max": 22848.0,
+      "std": 3749.101287319555,
+      "count": 374,
+      "sum": 4695424.0
+    },
+    "output_token_count": {
+      "unit": "tokens",
+      "avg": 219.84031413612564,
+      "p1": 38.480000000000004,
+      "p5": 46.0,
+      "p10": 55.0,
+      "p25": 66.25,
+      "p50": 90.5,
+      "p75": 164.25,
+      "p90": 518.7000000000005,
+      "p95": 1124.0499999999997,
+      "p99": 1616.98,
+      "min": 26.0,
+      "max": 2878.0,
+      "std": 347.9035539878663,
+      "count": 382,
+      "sum": 83979.0
+    },
+    "usage_prompt_tokens": {
+      "unit": "tokens",
+      "avg": 13426.429319371728,
+      "p1": 8203.72,
+      "p5": 8437.45,
+      "p10": 8699.3,
+      "p25": 10319.5,
+      "p50": 12627.0,
+      "p75": 16277.0,
+      "p90": 19019.800000000003,
+      "p95": 20537.85,
+      "p99": 22455.94,
+      "min": 8106.0,
+      "max": 24197.0,
+      "std": 3839.6592793019936,
+      "count": 382,
+      "sum": 5128896.0
+    },
+    "http_req_chunks_received": {
+      "unit": "count",
+      "avg": 118.64921465968587,
+      "p1": 19.43,
+      "p5": 24.0,
+      "p10": 28.0,
+      "p25": 35.0,
+      "p50": 49.0,
+      "p75": 89.0,
+      "p90": 274.0,
+      "p95": 620.9,
+      "p99": 861.95,
+      "min": 14.0,
+      "max": 1390.0,
+      "std": 186.0436319001202,
+      "count": 382,
+      "sum": 45324.0
+    },
+    "usage_completion_tokens": {
+      "unit": "tokens",
+      "avg": 219.84031413612564,
+      "p1": 38.480000000000004,
+      "p5": 46.0,
+      "p10": 55.0,
+      "p25": 66.25,
+      "p50": 90.5,
+      "p75": 164.25,
+      "p90": 518.7000000000005,
+      "p95": 1124.0499999999997,
+      "p99": 1616.98,
+      "min": 26.0,
+      "max": 2878.0,
+      "std": 347.9035539878663,
+      "count": 382,
+      "sum": 83979.0
+    },
+    "e2e_output_token_throughput": {
+      "unit": "tokens/sec/user",
+      "avg": 17.743823723942544,
+      "p1": 7.582229485061348,
+      "p5": 9.1688574206221,
+      "p10": 9.864761218027578,
+      "p25": 11.650816992103701,
+      "p50": 15.76266628302917,
+      "p75": 22.508623511253994,
+      "p90": 25.842234591800004,
+      "p95": 31.679216977410896,
+      "p99": 44.40313217790437,
+      "min": 5.6165066597389615,
+      "max": 68.34245573917983,
+      "std": 8.008653172349742,
+      "count": 382,
+      "sum": 6778.140662546051
+    },
+    "osl_mismatch_diff_pct": {
+      "unit": "%",
+      "avg": -82.66032680290645,
+      "p1": -96.55063667205478,
+      "p5": -95.87710051940115,
+      "p10": -95.21266799890286,
+      "p25": -94.05872627947096,
+      "p50": -91.990984616191,
+      "p75": -85.44191163982484,
+      "p90": -58.64102124615276,
+      "p95": -1.9972948992620716,
+      "p99": 0.0,
+      "min": -97.78534923339012,
+      "max": 0.0,
+      "std": 23.731519257415446,
+      "count": 382,
+      "sum": -31576.244838710267
+    },
+    "inter_token_latency": {
+      "unit": "ms",
+      "avg": 55.338085839271166,
+      "p1": 16.74054031407807,
+      "p5": 23.41532577403226,
+      "p10": 31.854518150106806,
+      "p25": 39.294519456739096,
+      "p50": 52.64029700838034,
+      "p75": 69.42866331632277,
+      "p90": 83.19095793368984,
+      "p95": 94.95387318070333,
+      "p99": 112.31235720219405,
+      "min": 11.949715655172414,
+      "max": 124.1912240609756,
+      "std": 21.33483739666194,
+      "count": 382,
+      "sum": 21139.148790601583
+    },
+    "prefill_throughput_per_user": {
+      "unit": "tokens/sec/user",
+      "avg": 19675.17179254096,
+      "p1": 1903.8762038680454,
+      "p5": 2417.443352144764,
+      "p10": 7154.129261562466,
+      "p25": 12027.5818985683,
+      "p50": 17809.445067164794,
+      "p75": 25174.70103012896,
+      "p90": 35084.48831190863,
+      "p95": 44857.59336039875,
+      "p99": 52282.288214323744,
+      "min": 980.6867717450476,
+      "max": 55761.860615878475,
+      "std": 11492.510639465438,
+      "count": 382,
+      "sum": 7515915.624750647
+    },
+    "http_req_connection_overhead": {
+      "unit": "ms",
+      "avg": 0.35926221465968583,
+      "p1": 0.0,
+      "p5": 0.0,
+      "p10": 0.0,
+      "p25": 0.0,
+      "p50": 0.0,
+      "p75": 0.0,
+      "p90": 0.6244785000000236,
+      "p95": 2.107126899999998,
+      "p99": 7.768330689999999,
+      "min": 0.0,
+      "max": 13.522357999999999,
+      "std": 1.4405627481337169,
+      "count": 382,
+      "sum": 137.238166
+    },
+    "http_req_total": {
+      "unit": "ms",
+      "avg": 11023.178298714658,
+      "p1": 1632.63471796,
+      "p5": 2907.28809595,
+      "p10": 3442.1320522,
+      "p25": 4807.008181,
+      "p50": 6354.2891039999995,
+      "p75": 10558.30346875,
+      "p90": 23725.239699400005,
+      "p95": 41934.321728599956,
+      "p99": 68754.80068390998,
+      "min": 1025.357027,
+      "max": 79463.346425,
+      "std": 12892.48215339059,
+      "count": 382,
+      "sum": 4210854.110109
+    },
+    "osl_mismatch_count": {
+      "unit": "requests",
+      "avg": 361.0
+    },
+    "output_token_throughput_per_user": {
+      "unit": "tokens/sec/user",
+      "avg": 21.484933078913674,
+      "p1": 8.904753360251904,
+      "p5": 10.531430481225932,
+      "p10": 12.02054942993839,
+      "p25": 14.403274308865985,
+      "p50": 18.996897185585397,
+      "p75": 25.448864432922633,
+      "p90": 31.39272097603246,
+      "p95": 42.70711835325918,
+      "p99": 59.75461445683822,
+      "min": 8.052098749820022,
+      "max": 83.68399959099878,
+      "std": 10.552668409827588,
+      "count": 382,
+      "sum": 8207.244436145023
+    },
+    "inter_chunk_latency": {
+      "unit": "ms",
+      "avg": 84.15049851899157,
+      "p1": 0.80857589,
+      "p5": 2.7875655,
+      "p10": 11.133312600000002,
+      "p25": 31.10146025,
+      "p50": 36.161820999999996,
+      "p75": 45.19338,
+      "p90": 299.6567431,
+      "p95": 333.5365473499999,
+      "p99": 886.02949565,
+      "min": 0.0,
+      "max": 1815.9494009999999,
+      "std": 157.9710515475207,
+      "count": 44862,
+      "sum": 3775159.6645589997
+    },
+    "total_osl": {
+      "unit": "tokens",
+      "avg": 83979.0
+    },
+    "total_usage_completion_tokens": {
+      "unit": "tokens",
+      "avg": 83979.0
+    },
+    "total_usage_total_tokens": {
+      "unit": "tokens",
+      "avg": 5212875.0
+    },
+    "total_isl": {
+      "unit": "tokens",
+      "avg": 5128896.0
+    },
+    "total_usage_prompt_cache_read_tokens": {
+      "unit": "tokens",
+      "avg": 4695424.0
+    },
+    "total_usage_prompt_tokens": {
+      "unit": "tokens",
+      "avg": 5128896.0
+    },
+    "total_output_tokens": {
+      "unit": "tokens",
+      "avg": 83979.0
+    },
+    "benchmark_duration": {
+      "unit": "sec",
+      "avg": 152.605379328
+    },
+    "overall_usage_prompt_cache_read_pct": {
+      "unit": "%",
+      "avg": 91.54843459489138
+    },
+    "total_token_throughput": {
+      "unit": "tokens/sec",
+      "avg": 34153.34332133807
+    },
+    "input_token_throughput": {
+      "unit": "tokens/sec",
+      "avg": 33603.13568758843
+    },
+    "output_token_throughput": {
+      "unit": "tokens/sec",
+      "avg": 550.2076337496391
+    },
+    "request_throughput": {
+      "unit": "requests/sec",
+      "avg": 2.502760405486635
+    },
+    "effective_concurrency": {
+      "unit": "requests",
+      "avg": 27.587808291441963,
+      "p50": 32.0,
+      "p90": 36.0,
+      "p95": 36.0,
+      "p99": 36.0,
+      "min": 0.0,
+      "max": 36.0,
+      "std": 11.518262942062973
+    },
+    "effective_decode_throughput": {
+      "unit": "tokens/sec",
+      "avg": 547.7047688603631,
+      "p50": 177.41454854753064,
+      "p90": 1617.487758355731,
+      "p95": 1736.70965556401,
+      "p99": 2898.98371677276,
+      "min": 0.0,
+      "max": 62141.7225861912,
+      "std": 1030.6446807975476
+    },
+    "effective_prefill_throughput": {
+      "unit": "tokens/sec",
+      "avg": 33603.13569221181,
+      "p50": 26022.141239064982,
+      "p90": 87556.88509141632,
+      "p95": 107058.63730322958,
+      "p99": 155269.31704341943,
+      "min": 0.0,
+      "max": 227759.47004351753,
+      "std": 38390.895381233
+    },
+    "effective_decode_concurrency": {
+      "unit": "requests",
+      "avg": 24.886453435016275,
+      "p50": 30.0,
+      "p90": 34.0,
+      "p95": 34.0,
+      "p99": 35.0,
+      "min": 0.0,
+      "max": 36.0,
+      "std": 11.356987378351379
+    },
+    "effective_prefill_concurrency": {
+      "unit": "requests",
+      "avg": 2.7013548564256875,
+      "p50": 1.0,
+      "p90": 5.0,
+      "p95": 10.0,
+      "p99": 36.0,
+      "min": 0.0,
+      "max": 36.0,
+      "std": 5.855703852525868
+    },
+    "effective_total_throughput": {
+      "unit": "tokens/sec",
+      "avg": 34150.84056563132,
+      "p50": 26589.819142040215,
+      "p90": 88207.72437398243,
+      "p95": 107532.68523225182,
+      "p99": 155719.6814629649,
+      "min": 0.0,
+      "max": 228005.9573292448,
+      "std": 38389.46247001123
+    },
+    "effective_decode_throughput_per_user": {
+      "unit": "tokens/sec/user",
+      "avg": 30.73818299219542,
+      "p50": 6.431683552777989,
+      "p90": 73.32497910894692,
+      "p95": 108.98332486860785,
+      "p99": 125.2574251332404,
+      "min": 0.0,
+      "max": 29289.623401108092,
+      "std": 72.18014951299381
+    },
+    "effective_prefill_throughput_per_user": {
+      "unit": "tokens/sec/user",
+      "avg": 12444.145174346122,
+      "p50": 11334.014325786773,
+      "p90": 30206.577814163014,
+      "p95": 36809.625637746925,
+      "p99": 52209.82210897042,
+      "min": 0.0,
+      "max": 55761.87960990282,
+      "std": 13305.477311018485
+    },
+    "tokens_in_flight": {
+      "unit": "tokens",
+      "avg": 402895.09022163914,
+      "p50": 482486.1792351537,
+      "p90": 558487.8782349091,
+      "p95": 570672.1660263103,
+      "p99": 590996.883793484,
+      "min": 0.0,
+      "max": 596576.0639588513,
+      "std": 172249.95136862595
+    },
+    "active_decode_throughput": {
+      "unit": "tokens/sec",
+      "avg": 558.7640925047532,
+      "p50": 179.53231569804706,
+      "p90": 1619.7960724009229,
+      "p95": 1743.1484553565626,
+      "p99": 2911.1913271542776,
+      "min": 0.0,
+      "max": 62141.7225861912,
+      "std": 1038.025797303407
+    },
+    "active_prefill_throughput": {
+      "unit": "tokens/sec",
+      "avg": 55193.22175913491,
+      "p50": 47015.51937976078,
+      "p90": 100657.05570262866,
+      "p95": 119352.91589011741,
+      "p99": 171773.24434189,
+      "min": 980.686774236624,
+      "max": 227759.47004351753,
+      "std": 35059.85119796225
+    },
+    "active_decode_throughput_per_user": {
+      "unit": "tokens/sec/user",
+      "avg": 31.358852252860288,
+      "p50": 6.572624485257236,
+      "p90": 74.9651586399057,
+      "p95": 109.04999277699417,
+      "p99": 129.11434662875445,
+      "min": 0.0,
+      "max": 29289.623401108092,
+      "std": 72.77163567384969
+    },
+    "active_prefill_throughput_per_user": {
+      "unit": "tokens/sec/user",
+      "avg": 20439.534884529876,
+      "p50": 19418.51004364779,
+      "p90": 36014.52824028737,
+      "p95": 45293.11181630933,
+      "p99": 52591.23290502495,
+      "min": 980.686774236624,
+      "max": 55761.87960990282,
+      "std": 11285.367070767516
+    },
+    "active_total_throughput": {
+      "unit": "tokens/sec",
+      "avg": 34156.67929238992,
+      "p50": 26589.819142040215,
+      "p90": 88207.72437398243,
+      "p95": 107532.68523225182,
+      "p99": 155719.6814629649,
+      "min": 69.49715288034847,
+      "max": 228005.9573292448,
+      "std": 38390.14668591671
+    },
+    "credit_to_start_latency": {
+      "unit": "ms",
+      "avg": 2.035722722513089,
+      "p1": 0.9849472,
+      "p5": 1.1019776,
+      "p10": 1.1666176000000001,
+      "p25": 1.273728,
+      "p50": 1.393152,
+      "p75": 1.570816,
+      "p90": 2.609536000000001,
+      "p95": 7.544691199999998,
+      "p99": 13.276254719999999,
+      "min": 0.804096,
+      "max": 15.110144,
+      "std": 2.192599651139441,
+      "count": 382,
+      "sum": 777.64608
+    },
+    "effective_latency": {
+      "unit": "ms",
+      "avg": 11024.987910701571,
+      "p1": 1634.01657856,
+      "p5": 2908.9813631999996,
+      "p10": 3443.391744,
+      "p25": 4808.467264000001,
+      "p50": 6355.869056,
+      "p75": 10559.805632,
+      "p90": 23726.815052800008,
+      "p95": 41936.00168959996,
+      "p99": 68756.30119423999,
+      "min": 1025.623552,
+      "max": 79464.803072,
+      "std": 12892.684567089169,
+      "count": 382,
+      "sum": 4211545.381888
+    }
+  },
+  "http_req_chunks_sent": {
+    "unit": "count",
+    "avg": 1.0,
+    "p1": 1.0,
+    "p5": 1.0,
+    "p10": 1.0,
+    "p25": 1.0,
+    "p50": 1.0,
+    "p75": 1.0,
+    "p90": 1.0,
+    "p95": 1.0,
+    "p99": 1.0,
+    "min": 1.0,
+    "max": 1.0,
+    "std": 0.0,
+    "count": 1833,
+    "sum": 1833.0
+  },
+  "http_req_data_received": {
+    "unit": "KB",
+    "avg": 49.00701867669122,
+    "p1": 5.3125,
+    "p5": 7.312890625,
+    "p10": 8.4501953125,
+    "p25": 11.2216796875,
+    "p50": 19.658203125,
+    "p75": 65.859375,
+    "p90": 145.736328125,
+    "p95": 169.9013671875,
+    "p99": 217.26667968750004,
+    "min": 4.310546875,
+    "max": 289.056640625,
+    "std": 57.059825619938785,
+    "count": 1833,
+    "sum": 89829.865234375
+  },
+  "http_req_sending": {
+    "unit": "ms",
+    "avg": 0.3636418052373159,
+    "p1": 0.1916262,
+    "p5": 0.20320259999999998,
+    "p10": 0.209981,
+    "p25": 0.22847199999999998,
+    "p50": 0.25812599999999997,
+    "p75": 0.310316,
+    "p90": 0.37533459999999985,
+    "p95": 0.4622059999999999,
+    "p99": 2.73468816000001,
+    "min": 0.087752,
+    "max": 11.064639,
+    "std": 0.6985535223267454,
+    "count": 1833,
+    "sum": 666.555429
+  },
+  "http_req_connection_reused": {
+    "unit": "ratio",
+    "avg": 0.9710856519367158,
+    "p1": 0.0,
+    "p5": 1.0,
+    "p10": 1.0,
+    "p25": 1.0,
+    "p50": 1.0,
+    "p75": 1.0,
+    "p90": 1.0,
+    "p95": 1.0,
+    "p99": 1.0,
+    "min": 0.0,
+    "max": 1.0,
+    "std": 0.16756583344870607,
+    "count": 1833,
+    "sum": 1780.0
+  },
+  "http_req_dns_lookup": {
+    "unit": "ms",
+    "avg": 0.0,
+    "p1": 0.0,
+    "p5": 0.0,
+    "p10": 0.0,
+    "p25": 0.0,
+    "p50": 0.0,
+    "p75": 0.0,
+    "p90": 0.0,
+    "p95": 0.0,
+    "p99": 0.0,
+    "min": 0.0,
+    "max": 0.0,
+    "std": 0.0,
+    "count": 1833,
+    "sum": 0.0
+  },
+  "http_req_connecting": {
+    "unit": "ms",
+    "avg": 0.08867839279869066,
+    "p1": 0.0,
+    "p5": 0.0,
+    "p10": 0.0,
+    "p25": 0.0,
+    "p50": 0.0,
+    "p75": 0.0,
+    "p90": 0.0,
+    "p95": 0.0,
+    "p99": 2.5887481200000058,
+    "min": 0.0,
+    "max": 10.602114,
+    "std": 0.6882316148067046,
+    "count": 1833,
+    "sum": 162.547494
+  },
+  "http_req_blocked": {
+    "unit": "ms",
+    "avg": 0.0,
+    "p1": 0.0,
+    "p5": 0.0,
+    "p10": 0.0,
+    "p25": 0.0,
+    "p50": 0.0,
+    "p75": 0.0,
+    "p90": 0.0,
+    "p95": 0.0,
+    "p99": 0.0,
+    "min": 0.0,
+    "max": 0.0,
+    "std": 0.0,
+    "count": 1833,
+    "sum": 0.0
+  },
+  "http_req_waiting": {
+    "unit": "ms",
+    "avg": 1039.2953858134206,
+    "p1": 355.54133427999994,
+    "p5": 376.9109166,
+    "p10": 395.4207618,
+    "p25": 468.287131,
+    "p50": 663.854683,
+    "p75": 894.307588,
+    "p90": 1972.9591419999997,
+    "p95": 3547.3441784,
+    "p99": 7108.9073627600255,
+    "min": 324.612912,
+    "max": 23117.103763,
+    "std": 1417.9073894564901,
+    "count": 1833,
+    "sum": 1905028.442196
+  },
+  "http_req_receiving": {
+    "unit": "ms",
+    "avg": 14768.077621624112,
+    "p1": 920.1236415999999,
+    "p5": 1515.8277492,
+    "p10": 2049.8109308000007,
+    "p25": 3472.784331,
+    "p50": 6603.17094,
+    "p75": 19859.318268,
+    "p90": 42807.804033199995,
+    "p95": 50808.90785339996,
+    "p99": 67045.99440220004,
+    "min": 534.167635,
+    "max": 92360.81533699999,
+    "std": 16950.620324969597,
+    "count": 1833,
+    "sum": 27069886.280437
+  },
+  "usage_total_tokens": {
+    "unit": "tokens",
+    "avg": 24830.56082924168,
+    "p1": 8343.8,
+    "p5": 9410.400000000001,
+    "p10": 10572.2,
+    "p25": 15792.0,
+    "p50": 23476.0,
+    "p75": 33087.0,
+    "p90": 40813.6,
+    "p95": 44097.2,
+    "p99": 49647.76000000001,
+    "min": 8151.0,
+    "max": 55332.0,
+    "std": 11031.185402400899,
+    "count": 1833,
+    "sum": 45514418.0
+  },
+  "http_req_duration": {
+    "unit": "ms",
+    "avg": 15807.73664924277,
+    "p1": 1469.8422951199998,
+    "p5": 2135.0551336,
+    "p10": 2763.5318492,
+    "p25": 4263.534342999999,
+    "p50": 7882.243818999999,
+    "p75": 20929.902606,
+    "p90": 43664.22651999999,
+    "p95": 51694.79024079998,
+    "p99": 67595.22217316001,
+    "min": 938.7277029999999,
+    "max": 93147.119146,
+    "std": 17014.843278212917,
+    "count": 1833,
+    "sum": 28975581.278061997
+  },
+  "time_to_first_output_token": {
+    "unit": "ms",
+    "avg": 7077.199212370613,
+    "p1": 399.7647762,
+    "p5": 429.87731299999996,
+    "p10": 485.824848,
+    "p25": 800.673821,
+    "p50": 2167.231703,
+    "p75": 6773.496749,
+    "p90": 23607.910196,
+    "p95": 34731.035478,
+    "p99": 51945.21263580002,
+    "min": 359.04769799999997,
+    "max": 77151.907209,
+    "std": 11407.636786376237,
+    "count": 1681,
+    "sum": 11896771.875994999
+  },
+  "http_req_data_sent": {
+    "unit": "KB",
+    "avg": 101.90158722040371,
+    "p1": 37.59234375,
+    "p5": 41.4248046875,
+    "p10": 45.239453125000004,
+    "p25": 64.783203125,
+    "p50": 95.6240234375,
+    "p75": 136.0927734375,
+    "p90": 167.24765625,
+    "p95": 182.039453125,
+    "p99": 204.25050781250002,
+    "min": 36.7392578125,
+    "max": 224.6240234375,
+    "std": 44.80318219000962,
+    "count": 1833,
+    "sum": 186785.609375
+  },
+  "usage_prompt_cache_read_tokens": {
+    "unit": "tokens",
+    "avg": 23677.33479452055,
+    "p1": 7744.0,
+    "p5": 8384.0,
+    "p10": 9920.0,
+    "p25": 14528.0,
+    "p50": 22464.0,
+    "p75": 31872.0,
+    "p90": 39398.40000000001,
+    "p95": 43059.2,
+    "p99": 48816.64,
+    "min": 7744.0,
+    "max": 53312.0,
+    "std": 10926.981607926657,
+    "count": 1825,
+    "sum": 43211136.0
+  },
+  "usage_prompt_tokens": {
+    "unit": "tokens",
+    "avg": 24458.55319148936,
+    "p1": 8269.12,
+    "p5": 9343.400000000001,
+    "p10": 10456.0,
+    "p25": 15545.0,
+    "p50": 23127.0,
+    "p75": 32590.0,
+    "p90": 40352.99999999999,
+    "p95": 43615.799999999996,
+    "p99": 49306.60000000001,
+    "min": 8085.0,
+    "max": 53769.0,
+    "std": 10898.511761819209,
+    "count": 1833,
+    "sum": 44832528.0
+  },
+  "http_req_chunks_received": {
+    "unit": "count",
+    "avg": 199.68794326241135,
+    "p1": 20.0,
+    "p5": 27.0,
+    "p10": 31.0,
+    "p25": 43.0,
+    "p50": 77.0,
+    "p75": 271.0,
+    "p90": 604.8,
+    "p95": 707.1999999999996,
+    "p99": 914.4000000000003,
+    "min": 16.0,
+    "max": 1216.0,
+    "std": 238.0134155487954,
+    "count": 1833,
+    "sum": 366028.0
+  },
+  "usage_completion_tokens": {
+    "unit": "tokens",
+    "avg": 372.0076377523186,
+    "p1": 37.32,
+    "p5": 52.0,
+    "p10": 61.0,
+    "p25": 81.0,
+    "p50": 142.0,
+    "p75": 497.0,
+    "p90": 1119.8,
+    "p95": 1314.799999999999,
+    "p99": 1708.4800000000007,
+    "min": 30.0,
+    "max": 2286.0,
+    "std": 444.5551924008996,
+    "count": 1833,
+    "sum": 681890.0
+  },
+  "e2e_output_token_throughput": {
+    "unit": "tokens/sec/user",
+    "avg": 22.456862643521227,
+    "p1": 7.165413299348929,
+    "p5": 9.363918846179239,
+    "p10": 11.171552970368136,
+    "p25": 18.734693721491215,
+    "p50": 23.69377593876951,
+    "p75": 26.83260211026705,
+    "p90": 29.39927335415507,
+    "p95": 31.35160079822535,
+    "p99": 38.51314302591598,
+    "min": 4.664493979279221,
+    "max": 63.41870948475335,
+    "std": 6.960197504633241,
+    "count": 1833,
+    "sum": 41163.42922557441
+  },
+  "osl_mismatch_diff_pct": {
+    "unit": "%",
+    "avg": -71.38340431114183,
+    "p1": -96.90628556465339,
+    "p5": -95.6853013002442,
+    "p10": -94.76078859491395,
+    "p25": -93.01675977653632,
+    "p50": -88.04537521815008,
+    "p75": -61.40939597315436,
+    "p90": 0.0,
+    "p95": 0.0,
+    "p99": 0.0,
+    "min": -98.11202013845187,
+    "max": 0.0,
+    "std": 32.113422363029635,
+    "count": 1833,
+    "sum": -130845.78010232298
+  },
+  "prefill_throughput_per_user": {
+    "unit": "tokens/sec/user",
+    "avg": 36862.54525232061,
+    "p1": 2058.5099834030957,
+    "p5": 4206.042645068822,
+    "p10": 9117.511711102494,
+    "p25": 17891.44647122276,
+    "p50": 34137.229431362204,
+    "p75": 52341.732701243855,
+    "p90": 69050.92398635374,
+    "p95": 81390.52488325739,
+    "p99": 100448.23638957745,
+    "min": 421.32780842492605,
+    "max": 119600.61957722774,
+    "std": 23424.182268467495,
+    "count": 1833,
+    "sum": 67569045.44750369
+  },
+  "http_req_connection_overhead": {
+    "unit": "ms",
+    "avg": 0.08867839279869066,
+    "p1": 0.0,
+    "p5": 0.0,
+    "p10": 0.0,
+    "p25": 0.0,
+    "p50": 0.0,
+    "p75": 0.0,
+    "p90": 0.0,
+    "p95": 0.0,
+    "p99": 2.5887481200000058,
+    "min": 0.0,
+    "max": 10.602114,
+    "std": 0.6882316148067046,
+    "count": 1833,
+    "sum": 162.547494
+  },
+  "http_req_total": {
+    "unit": "ms",
+    "avg": 15807.82532763557,
+    "p1": 1469.8422951199998,
+    "p5": 2135.0551336,
+    "p10": 2763.5318492,
+    "p25": 4263.534342999999,
+    "p50": 7882.243818999999,
+    "p75": 20929.902606,
+    "p90": 43664.22651999999,
+    "p95": 51694.79024079998,
+    "p99": 67595.22217316001,
+    "min": 938.7277029999999,
+    "max": 93147.119146,
+    "std": 17014.815124881872,
+    "count": 1833,
+    "sum": 28975743.825556
+  },
+  "osl_mismatch_count": {
+    "unit": "requests",
+    "avg": 1629.0
+  },
+  "total_usage_completion_tokens": {
+    "unit": "tokens",
+    "avg": 681890.0
+  },
+  "total_usage_total_tokens": {
+    "unit": "tokens",
+    "avg": 45514418.0
+  },
+  "total_usage_prompt_cache_read_tokens": {
+    "unit": "tokens",
+    "avg": 43211136.0
+  },
+  "total_usage_prompt_tokens": {
+    "unit": "tokens",
+    "avg": 44832528.0
+  },
+  "overall_usage_prompt_cache_read_pct": {
+    "unit": "%",
+    "avg": 96.3834473041538
+  },
+  "total_token_throughput": {
+    "unit": "tokens/sec",
+    "avg": 49425.90656353134
+  },
+  "input_token_throughput": {
+    "unit": "tokens/sec",
+    "avg": 48685.41524434966
+  },
+  "effective_concurrency": {
+    "unit": "requests",
+    "avg": 31.46601072192701,
+    "p50": 32.0,
+    "p90": 35.0,
+    "p95": 36.0,
+    "p99": 36.0,
+    "min": 0.0,
+    "max": 36.0,
+    "std": 4.330697977516465
+  },
+  "effective_decode_throughput": {
+    "unit": "tokens/sec",
+    "avg": 738.5007978045049,
+    "p50": 196.14242068567836,
+    "p90": 1643.032399479645,
+    "p95": 1723.0487109839144,
+    "p99": 3765.824859539257,
+    "min": 0.0,
+    "max": 103688.32793895199,
+    "std": 1343.7373209273323
+  },
+  "effective_prefill_throughput": {
+    "unit": "tokens/sec",
+    "avg": 48685.415236102046,
+    "p50": 39133.31562665334,
+    "p90": 114454.81584751369,
+    "p95": 138332.34747057554,
+    "p99": 201085.20251311825,
+    "min": 0.0,
+    "max": 341378.3982821152,
+    "std": 48585.294193134076
+  },
+  "effective_decode_concurrency": {
+    "unit": "requests",
+    "avg": 29.32169521806419,
+    "p50": 30.0,
+    "p90": 33.0,
+    "p95": 33.0,
+    "p99": 34.0,
+    "min": 0.0,
+    "max": 35.0,
+    "std": 4.328438573969661
+  },
+  "effective_prefill_concurrency": {
+    "unit": "requests",
+    "avg": 2.1443155038628223,
+    "p50": 2.0,
+    "p90": 4.0,
+    "p95": 5.0,
+    "p99": 7.0,
+    "min": 0.0,
+    "max": 36.0,
+    "std": 2.6484094362541937
+  },
+  "effective_total_throughput": {
+    "unit": "tokens/sec",
+    "avg": 49423.916027964915,
+    "p50": 39885.869115089976,
+    "p90": 115295.03157272971,
+    "p95": 139091.13381342386,
+    "p99": 201893.82313504344,
+    "min": 0.0,
+    "max": 342154.1455002162,
+    "std": 48575.042569395126
+  },
+  "effective_decode_throughput_per_user": {
+    "unit": "tokens/sec/user",
+    "avg": 25.33568450157711,
+    "p50": 6.4489116182282284,
+    "p90": 52.19257800817146,
+    "p95": 55.239882851637,
+    "p99": 127.65188925845355,
+    "min": 0.0,
+    "max": 24742.8959058855,
+    "std": 49.348513104590204
+  },
+  "effective_prefill_throughput_per_user": {
+    "unit": "tokens/sec/user",
+    "avg": 21052.55465715755,
+    "p50": 15019.368915184494,
+    "p90": 52205.55445350879,
+    "p95": 64587.875155543086,
+    "p99": 87080.67285291648,
+    "min": 0.0,
+    "max": 109836.09216898837,
+    "std": 21514.579010306734
+  },
+  "tokens_in_flight": {
+    "unit": "tokens",
+    "avg": 861160.3739496818,
+    "p50": 894248.9684221098,
+    "p90": 1164760.3000667104,
+    "p95": 1188237.6709985093,
+    "p99": 1230231.5759041905,
+    "min": 0.0,
+    "max": 1269245.6211502475,
+    "std": 259402.47739595448
+  },
+  "active_decode_throughput": {
+    "unit": "tokens/sec",
+    "avg": 740.8670871275966,
+    "p50": 196.70238027570443,
+    "p90": 1643.3509923907156,
+    "p95": 1723.2059704431092,
+    "p99": 3778.0614423198986,
+    "min": 0.0,
+    "max": 103688.32793895199,
+    "std": 1345.2369474816087
+  },
+  "active_prefill_throughput": {
+    "unit": "tokens/sec",
+    "avg": 60858.36094264382,
+    "p50": 54718.6802445697,
+    "p90": 121240.63984119502,
+    "p95": 147287.06328146087,
+    "p99": 207284.219883027,
+    "min": 421.32781075792377,
+    "max": 341378.3982821152,
+    "std": 47009.73489610732
+  },
+  "active_decode_throughput_per_user": {
+    "unit": "tokens/sec/user",
+    "avg": 25.416864589543888,
+    "p50": 6.471412539241991,
+    "p90": 52.19898919666795,
+    "p95": 55.27545727868979,
+    "p99": 127.77627857081926,
+    "min": 0.0,
+    "max": 24742.8959058855,
+    "std": 49.406633801860174
+  },
+  "active_prefill_throughput_per_user": {
+    "unit": "tokens/sec/user",
+    "avg": 26316.38169001291,
+    "p50": 20993.118765487987,
+    "p90": 57126.36402388474,
+    "p95": 68612.04821381268,
+    "p99": 91734.79284402734,
+    "min": 421.32781075792377,
+    "max": 109836.09216898837,
+    "std": 20978.242088515763
+  },
+  "active_total_throughput": {
+    "unit": "tokens/sec",
+    "avg": 49425.03561952671,
+    "p50": 39885.869115089976,
+    "p90": 115295.03157272971,
+    "p95": 139091.13381342386,
+    "p99": 201893.82313504344,
+    "min": 60.15561921797327,
+    "max": 342154.1455002162,
+    "std": 48575.02315908101
+  },
+  "credit_to_start_latency": {
+    "unit": "ms",
+    "avg": 1.6859010583742502,
+    "p1": 0.94043136,
+    "p5": 1.067776,
+    "p10": 1.129984,
+    "p25": 1.244928,
+    "p50": 1.39392,
+    "p75": 1.567232,
+    "p90": 1.8423296,
+    "p95": 2.1909503999999993,
+    "p99": 12.127703040000021,
+    "min": 0.478464,
+    "max": 19.676928,
+    "std": 1.6880070586351763,
+    "count": 1833,
+    "sum": 3090.2566400000005
+  },
+  "effective_latency": {
+    "unit": "ms",
+    "avg": 15809.563747718496,
+    "p1": 1477.58349312,
+    "p5": 2136.6119424,
+    "p10": 2764.9425408,
+    "p25": 4264.713472,
+    "p50": 7883.856128,
+    "p75": 20931.086592,
+    "p90": 43665.830604799994,
+    "p95": 51696.33643519998,
+    "p99": 67596.81550336002,
+    "min": 940.19328,
+    "max": 93149.087488,
+    "std": 17014.830127079284,
+    "count": 1833,
+    "sum": 28978930.349568002
+  }
+}

experiments/kv-cache-fp8/capacity/hammer/N0036/profile_export_console.txt

@@ -0,0 +1,87 @@
+
+
+                                                  NVIDIA AIPerf | LLM Metrics: Effective                                                  
+┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━┳━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━┓
+┃                                    Metric ┃        avg ┃    min ┃          max ┃          p99 ┃          p90 ┃        p50 ┃        std ┃
+┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━╇━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━┩
+│          Effective Concurrency (requests) │      31.47 │   0.00 │        36.00 │        36.00 │        35.00 │      32.00 │       4.33 │
+│  Effective Decode Throughput (tokens/sec) │     738.50 │   0.00 │   103,688.33 │     3,765.82 │     1,643.03 │     196.14 │   1,343.74 │
+│ Effective Prefill Throughput (tokens/sec) │  48,685.42 │   0.00 │   341,378.40 │   201,085.20 │   114,454.82 │  39,133.32 │  48,585.29 │
+│   Effective Decode Concurrency (requests) │      29.32 │   0.00 │        35.00 │        34.00 │        33.00 │      30.00 │       4.33 │
+│  Effective Prefill Concurrency (requests) │       2.14 │   0.00 │        36.00 │         7.00 │         4.00 │       2.00 │       2.65 │
+│   Effective Total Throughput (tokens/sec) │  49,423.92 │   0.00 │   342,154.15 │   201,893.82 │   115,295.03 │  39,885.87 │  48,575.04 │
+│      Effective Decode Throughput Per User │      25.34 │   0.00 │    24,742.90 │       127.65 │        52.19 │       6.45 │      49.35 │
+│                         (tokens/sec/user) │            │        │              │              │              │            │            │
+│     Effective Prefill Throughput Per User │  21,052.55 │   0.00 │   109,836.09 │    87,080.67 │    52,205.55 │  15,019.37 │  21,514.58 │
+│                         (tokens/sec/user) │            │        │              │              │              │            │            │
+│                 Tokens In Flight (tokens) │ 861,160.37 │   0.00 │ 1,269,245.62 │ 1,230,231.58 │ 1,164,760.30 │ 894,248.97 │ 259,402.48 │
+│         Effective Latency (CO-aware) (ms) │  15,809.56 │ 940.19 │    93,149.09 │    67,596.82 │    43,665.83 │   7,883.86 │  17,014.83 │
+└───────────────────────────────────────────┴────────────┴────────┴──────────────┴──────────────┴──────────────┴────────────┴────────────┘
+                                             NVIDIA AIPerf | LLM Metrics: Active                                              
+┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┓
+┃                                 Metric ┃       avg ┃    min ┃        max ┃        p99 ┃        p90 ┃       p50 ┃       std ┃
+┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━┩
+│  Active Decode Throughput (tokens/sec) │    740.87 │   0.00 │ 103,688.33 │   3,778.06 │   1,643.35 │    196.70 │  1,345.24 │
+│ Active Prefill Throughput (tokens/sec) │ 60,858.36 │ 421.33 │ 341,378.40 │ 207,284.22 │ 121,240.64 │ 54,718.68 │ 47,009.73 │
+│      Active Decode Throughput Per User │     25.42 │   0.00 │  24,742.90 │     127.78 │      52.20 │      6.47 │     49.41 │
+│                      (tokens/sec/user) │           │        │            │            │            │           │           │
+│     Active Prefill Throughput Per User │ 26,316.38 │ 421.33 │ 109,836.09 │  91,734.79 │  57,126.36 │ 20,993.12 │ 20,978.24 │
+│                      (tokens/sec/user) │           │        │            │            │            │           │           │
+│   Active Total Throughput (tokens/sec) │ 49,425.04 │  60.16 │ 342,154.15 │ 201,893.82 │ 115,295.03 │ 39,885.87 │ 48,575.02 │
+└────────────────────────────────────────┴───────────┴────────┴────────────┴────────────┴────────────┴───────────┴───────────┘
+                                                NVIDIA AIPerf | LLM Metrics: Usage                                                
+┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┓
+┃                                  Metric ┃           avg ┃      min ┃       max ┃       p99 ┃       p90 ┃       p50 ┃       std ┃
+┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━┩
+│            Usage Prompt Tokens (tokens) │     24,458.55 │ 8,085.00 │ 53,769.00 │ 49,306.60 │ 40,353.00 │ 23,127.00 │ 10,898.51 │
+│ Usage Prompt Cache Read Tokens (tokens) │     23,677.33 │ 7,744.00 │ 53,312.00 │ 48,816.64 │ 39,398.40 │ 22,464.00 │ 10,926.98 │
+│        Usage Completion Tokens (tokens) │        372.01 │    30.00 │  2,286.00 │  1,708.48 │  1,119.80 │    142.00 │    444.56 │
+│             Usage Total Tokens (tokens) │     24,830.56 │ 8,151.00 │ 55,332.00 │ 49,647.76 │ 40,813.60 │ 23,476.00 │ 11,031.19 │
+│      Total Usage Prompt Tokens (tokens) │ 44,832,528.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│    Total Usage Prompt Cache Read Tokens │ 43,211,136.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│                                (tokens) │               │          │           │           │           │           │           │
+│       Overall Usage Prompt Cache Read % │         96.38 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│                                     (%) │               │          │           │           │           │           │           │
+│  Total Usage Completion Tokens (tokens) │    681,890.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│       Total Usage Total Tokens (tokens) │ 45,514,418.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+└─────────────────────────────────────────┴───────────────┴──────────┴───────────┴───────────┴───────────┴───────────┴───────────┘
+                                                    NVIDIA AIPerf | LLM Metrics                                                     
+┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┓
+┃                                        Metric ┃       avg ┃      min ┃       max ┃       p99 ┃       p90 ┃       p50 ┃       std ┃
+┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━┩
+│                      Time to First Token (ms) │  1,077.26 │   327.87 │ 23,117.39 │  7,327.97 │  1,981.57 │    688.48 │  1,422.87 │
+│                     Time to Second Token (ms) │    100.35 │     0.00 │  1,192.87 │    670.46 │    327.15 │     36.29 │    165.68 │
+│               Time to First Output Token (ms) │  7,077.20 │   359.05 │ 77,151.91 │ 51,945.21 │ 23,607.91 │  2,167.23 │ 11,407.64 │
+│                          Request Latency (ms) │ 15,769.31 │   902.28 │ 93,147.30 │ 67,595.36 │ 43,565.18 │  7,860.28 │ 17,025.99 │
+│                      Inter Token Latency (ms) │     42.87 │     9.41 │    134.79 │    116.65 │     62.79 │     38.08 │     18.59 │
+│              Output Token Throughput Per User │     26.53 │     7.42 │    106.30 │     57.90 │     35.67 │     26.26 │      8.98 │
+│                             (tokens/sec/user) │           │          │           │           │           │           │           │
+│ E2E Output Token Throughput (tokens/sec/user) │     22.46 │     4.66 │     63.42 │     38.51 │     29.40 │     23.69 │      6.96 │
+│               Output Sequence Length (tokens) │    372.01 │    30.00 │  2,286.00 │  1,708.48 │  1,119.80 │    142.00 │    444.56 │
+│                Input Sequence Length (tokens) │ 24,458.55 │ 8,085.00 │ 53,769.00 │ 49,306.60 │ 40,353.00 │ 23,127.00 │ 10,898.51 │
+│          Output Token Throughput (tokens/sec) │    740.49 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│           Input Token Throughput (tokens/sec) │ 48,685.42 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│             Request Throughput (requests/sec) │      1.99 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│                      Request Count (requests) │  1,833.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+└───────────────────────────────────────────────┴───────────┴──────────┴───────────┴───────────┴───────────┴───────────┴───────────┘
+
+
+No GPU telemetry data collected during the benchmarking run.
+
+╭───────────────────────────────── Output Sequence Length Mismatch Warning ─────────────────────────────────╮
+│  1,629 of 1,833 requests (88.9%) have output length differing from requested by more than the threshold.  │
+│  Threshold (tokens): min(requested x 5%, 50)                                                              │
+│  Average mismatch: -71.4%                                                                                 │
+│                                                                                                           │
+│  Why: Server hit EOS token before reaching requested output length.                                       │
+│                                                                                                           │
+│  Fix Options:                                                                                             │
+│    - --extra-inputs ignore_eos:true - Generate until max_tokens (vLLM, TensorRT-LLM)                      │
+│    - --extra-inputs min_tokens:<N> - Set minimum output length (vLLM, TensorRT-LLM, SGLang)               │
+│    - --use-server-token-count - Use server-reported token counts if tokenizer mismatch suspected          │
+│                                                                                                           │
+│  Diagnostics:                                                                                             │
+│    - Review profile_export.jsonl -> osl_mismatch_diff_pct for per-request values                          │
+│    - Adjust: AIPERF_METRICS_OSL_MISMATCH_PCT_THRESHOLD=5                                                  │
+│    - Adjust: AIPERF_METRICS_OSL_MISMATCH_MAX_TOKEN_THRESHOLD=50                                           │
+╰───────────────────────────────────────────────────────────────────────────────────────────────────────────╯

experiments/kv-cache-fp8/capacity/hammer/N0036/summary.json

@@ -0,0 +1,25 @@
+{
+  "sent": 1833,
+  "records": 1833,
+  "valid": 1833,
+  "valid_frac": 1.0,
+  "p25_speed_tok_s": 22.2,
+  "p50_speed_tok_s": 26.3,
+  "p95_ttft_ms": 3553.0,
+  "p50_ttft_ms": 688.0,
+  "cache_read_ratio": 0.9638,
+  "max_turn_index": 49,
+  "conversations": 61,
+  "recycled_record_frac": 0.0,
+  "inflight_mean": 35.4,
+  "inflight_p95": 41.0,
+  "foreign_p50": 4.0,
+  "foreign_p95": 7.0,
+  "active_sessions_mean": 35.1,
+  "active_sessions_p95": 36.0,
+  "tiers": {
+    "20tok": true,
+    "60tok": false,
+    "180tok": false
+  }
+}

experiments/kv-cache-fp8/capacity/hammer/N0036/verify_cache_salt.log

@@ -0,0 +1 @@
+records_checked=1833 mismatches=0

experiments/kv-cache-fp8/capacity/hammer/N0036/verify_pin.log

@@ -0,0 +1 @@
+mode=pinned records_checked=1833 mismatches=0

experiments/kv-cache-fp8/capacity/hammer/N0038/aiperf_stdout.log

@@ -0,0 +1,233 @@
+[32m  36,533.95 │  44,389.47 │
+│      Effective Decode Throughput Per User │      24.14 │     0.00 │    18,104.82 │       118.69 │        51.69 │       6.23 │      44.88 │
+│                         (tokens/sec/user) │            │          │              │              │              │            │            │
+│     Effective Prefill Throughput Per User │  16,173.20 │     0.00 │    89,196.88 │    53,833.32 │    33,870.13 │  12,645.78 │  13,287.23 │
+│                         (tokens/sec/user) │            │          │              │              │              │            │            │
+│                 Tokens In Flight (tokens) │ 916,811.87 │     0.00 │ 1,318,476.16 │ 1,291,951.64 │ 1,226,890.70 │ 948,405.31 │ 266,043.93 │
+│         Effective Latency (CO-aware) (ms) │  16,848.34 │ 1,094.78 │   127,586.08 │    74,915.13 │    46,215.88 │   8,719.87 │  18,045.59 │
+└───────────────────────────────────────────┴────────────┴──────────┴──────────────┴──────────────┴──────────────┴────────────┴────────────┘
+                                             NVIDIA AIPerf | LLM Metrics: Active                                              
+┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┓
+┃                                 Metric ┃       avg ┃    min ┃        max ┃        p99 ┃        p90 ┃       p50 ┃       std ┃
+┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━┩
+│  Active Decode Throughput (tokens/sec) │    749.48 │   0.00 │  87,579.93 │   3,763.49 │   1,700.82 │    202.34 │  1,342.60 │
+│ Active Prefill Throughput (tokens/sec) │ 50,318.09 │ 885.97 │ 264,160.58 │ 176,310.73 │ 111,744.81 │ 39,616.74 │ 44,156.18 │
+│      Active Decode Throughput Per User │     24.22 │   0.00 │  18,104.82 │     118.78 │      51.70 │      6.23 │     44.93 │
+│                      (tokens/sec/user) │           │        │            │            │            │           │           │
+│     Active Prefill Throughput Per User │ 16,957.59 │ 885.97 │  89,196.88 │  55,222.21 │  34,288.81 │ 13,946.93 │ 13,107.70 │
+│                      (tokens/sec/user) │           │        │            │            │            │           │           │
+│   Active Total Throughput (tokens/sec) │ 48,738.67 │  49.82 │ 264,815.60 │ 176,765.59 │ 110,647.52 │ 36,553.61 │ 44,389.37 │
+└────────────────────────────────────────┴───────────┴────────┴────────────┴────────────┴────────────┴───────────┴───────────┘
+                                                NVIDIA AIPerf | LLM Metrics: Usage                                                
+┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┓
+┃                                  Metric ┃           avg ┃      min ┃       max ┃       p99 ┃       p90 ┃       p50 ┃       std ┃
+┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━┩
+│            Usage Prompt Tokens (tokens) │     23,992.62 │ 8,085.00 │ 56,709.00 │ 49,075.84 │ 39,679.20 │ 22,683.00 │ 10,676.69 │
+│ Usage Prompt Cache Read Tokens (tokens) │     23,179.21 │ 4,032.00 │ 54,784.00 │ 48,570.88 │ 38,784.00 │ 22,080.00 │ 10,681.80 │
+│        Usage Completion Tokens (tokens) │        374.50 │    32.00 │  3,291.00 │  1,770.88 │  1,134.60 │    144.00 │    457.15 │
+│             Usage Total Tokens (tokens) │     24,367.13 │ 8,162.00 │ 57,487.00 │ 49,679.18 │ 40,191.80 │ 23,002.00 │ 10,828.79 │
+│      Total Usage Prompt Tokens (tokens) │ 44,026,463.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│    Total Usage Prompt Cache Read Tokens │ 42,348,416.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│                                (tokens) │               │          │           │           │           │           │           │
+│       Overall Usage Prompt Cache Read % │         96.19 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│                                     (%) │               │          │           │           │           │           │           │
+│  Total Usage Completion Tokens (tokens) │    687,214.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+│       Total Usage Total Tokens (tokens) │ 44,713,677.00 │      N/A │       N/A │       N/A │       N/A │       N/A │       N/A │
+└─────────────────────────────────────────┴───────────────┴──────────┴───────────┴───────────┴───────────┴───────────┴───────────┘
+                                                     NVIDIA AIPerf | LLM Metrics                                                     
+┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┓
+┃                                        Metric ┃       avg ┃      min ┃        max ┃       p99 ┃       p90 ┃       p50 ┃       std ┃
+┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━┩
+│                      Time to First Token (ms) │  1,410.98 │   317.70 │  26,271.74 │ 10,406.72 │  3,540.68 │    718.66 │  1,990.34 │
+│                     Time to Second Token (ms) │    111.55 │     0.00 │   2,625.39 │    882.86 │    325.81 │     36.88 │    212.33 │
+│               Time to First Output Token (ms) │  7,720.60 │   372.69 │  92,473.96 │ 57,958.67 │ 23,781.50 │  2,891.77 │ 11,993.33 │
+│                          Request Latency (ms) │ 16,806.98 │ 1,093.06 │ 127,584.69 │ 74,913.55 │ 46,214.42 │  8,674.77 │ 18,056.85 │
+│                      Inter Token Latency (ms) │     44.83 │    15.12 │     131.56 │    109.72 │     71.19 │     40.04 │     18.50 │
+│              Output Token Throughput Per User │     25.18 │     7.60 │      66.12 │     55.18 │     33.98 │     24.98 │      8.17 │
+│                             (tokens/sec/user) │           │          │            │           │           │           │           │
+│ E2E Output Token Throughput (tokens/sec/user) │     20.97 │     2.10 │      51.82 │     35.33 │     27.81 │     22.40 │      6.75 │
+│               Output Sequence Length (tokens) │    374.50 │    32.00 │   3,291.00 │  1,770.88 │  1,134.60 │    144.00 │    457.15 │
+│                Input Sequence Length (tokens) │ 23,992.62 │ 8,085.00 │  56,709.00 │ 49,075.84 │ 39,679.20 │ 22,683.00 │ 10,676.69 │
+│          Output Token Throughput (tokens/sec) │    749.09 │      N/A │        N/A │       N/A │       N/A │       N/A │       N/A │
+│           Input Token Throughput (tokens/sec) │ 47,990.56 │      N/A │        N/A │       N/A │       N/A │       N/A │       N/A │
+│             Request Throughput (requests/sec) │      2.00 │      N/A │        N/A │       N/A │       N/A │       N/A │       N/A │
+│                      Request Count (requests) │  1,835.00 │      N/A │        N/A │       N/A │       N/A │       N/A │       N/A │
+└───────────────────────────────────────────────┴───────────┴──────────┴────────────┴───────────┴───────────┴───────────┴───────────┘
+
+
+No GPU telemetry data collected during the benchmarking run.
+
+╭───────────────────────────────── Output Sequence Length Mismatch Warning ─────────────────────────────────╮
+│  1,602 of 1,835 requests (87.3%) have output length differing from requested by more than the threshold.  │
+│  Threshold (tokens): min(requested x 5%, 50)                                                              │
+│  Average mismatch: -71.1%                                                                                 │
+│                                                                                                           │
+│  Why: Server hit EOS token before reaching requested output length.                                       │
+│                                                                                                           │
+│  Fix Options:                                                                                             │
+│    - --extra-inputs ignore_eos:true - Generate until max_tokens (vLLM, TensorRT-LLM)                      │
+│    - --extra-inputs min_tokens:<N> - Set minimum output length (vLLM, TensorRT-LLM, SGLang)               │
+│    - --use-server-token-count - Use server-reported token counts if tokenizer mismatch suspected          │
+│                                                                                                           │
+│  Diagnostics:                                                                                             │
+│    - Review profile_export.jsonl -> osl_mismatch_diff_pct for per-request values                          │
+│    - Adjust: AIPERF_METRICS_OSL_MISMATCH_PCT_THRESHOLD=5                                                  │
+│    - Adjust: AIPERF_METRICS_OSL_MISMATCH_MAX_TOKEN_THRESHOLD=50                                           │
+╰───────────────────────────────────────────────────────────────────────────────────────────────────────────╯
+
+CLI Command: aiperf profile --model 'glm-5.2' --url 'http://127.0.0.1:8391' --api-key '<redacted>' --endpoint-type 'chat' --streaming --tokenizer 'builtin' --custom-dataset-type 
+'dag_jsonl' --input-file 'data/capacity/corpus.jsonl' --extra-inputs '{"cache_salt": "6b630438f0d1c386ae7365fda9378e4c"}' --concurrency 38 --warmup-duration 120 --benchmark-duration 900 
+--benchmark-grace-period 300 --request-count 1000000 --random-seed 0 --use-server-token-count --artifact-dir 'results/capacity/sweep-hammer-jul20/N0038' --export-level 'raw' 
+--export-http-trace --slice-duration 60 --server-metrics-formats 'json' 'csv' 'jsonl' --ui 'simple'
+Benchmark Duration: 917.36 sec
+CSV Export: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0038/profile_export_aiperf.csv
+JSON Export: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0038/profile_export_aiperf.json
+Raw Records: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0038/profile_export_raw.jsonl
+Server Metrics CSV Export: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0038/server_metrics_export.csv
+Server Metrics JSON Export: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0038/server_metrics_export.json
+Timeslice CSV Export: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0038/profile_export_aiperf_timeslices.csv
+Timeslice JSON Export: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0038/profile_export_aiperf_timeslices.json
+Log File: /home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0038/logs/aiperf.log
+
+
+Warmup:   0% |          | [00:00<?]
+Warmup:   0% |          | [00:00<?]
+                                   
+
+Warmup:   2% |▏         | [00:01<01:57]
+Warmup:   2% |▏         | [00:01<01:57]
+Warmup:   3% |▎         | [00:03<01:55]
+Warmup:   3% |▎         | [00:03<01:55]
+Warmup:   5% |▌         | [00:05<01:53]
+Warmup:   5% |▌         | [00:05<01:53]
+Warmup:   7% |▋         | [00:07<01:51]
+Warmup:   7% |▋         | [00:07<01:51]
+Warmup:   8% |▊         | [00:09<01:49]
+Warmup:   8% |▊         | [00:09<01:49]
+Warmup:  10% |█         | [00:11<01:47]
+Warmup:  10% |█         | [00:11<01:47]
+Warmup:  12% |█▏        | [00:13<01:45]
+Warmup:  12% |█▏        | [00:13<01:45]
+Warmup:  13% |█▎        | [00:15<01:43]
+Warmup:  13% |█▎        | [00:15<01:43]
+Warmup:  15% |█▌        | [00:17<01:41]
+Warmup:  15% |█▌        | [00:17<01:41]
+Warmup:  17% |█▋        | [00:19<01:39]
+Warmup:  17% |█▋        | [00:19<01:39]
+Warmup:  18% |█▊        | [00:21<01:37]
+Warmup:  18% |█▊        | [00:21<01:37]
+Warmup:  20% |██        | [00:23<01:35]
+Warmup:  20% |██        | [00:23<01:35]
+Warmup:  22% |██▏       | [00:25<01:33]
+Warmup:  22% |██▏       | [00:25<01:33]
+Warmup:  23% |██▎       | [00:27<01:31]
+Warmup:  23% |██▎       | [00:27<01:31]
+Warmup:  25% |██▌       | [00:29<01:29]
+Warmup:  25% |██▌       | [00:29<01:29]
+Warmup:  27% |██▋       | [00:31<01:27]
+Warmup:  27% |██▋       | [00:31<01:27]
+Warmup:  28% |██▊       | [00:33<01:25]
+Warmup:  28% |██▊       | [00:33<01:25]
+Warmup:  30% |███       | [00:35<01:23]
+Warmup:  30% |███       | [00:35<01:23]
+Warmup:  32% |███▏      | [00:37<01:21]
+Warmup:  32% |███▏      | [00:37<01:21]
+Warmup:  33% |███▎      | [00:39<01:19]
+Warmup:  33% |███▎      | [00:39<01:19]
+Warmup:  35% |███▌      | [00:41<01:17]
+Warmup:  35% |███▌      | [00:41<01:17]
+Warmup:  37% |███▋      | [00:43<01:15]
+Warmup:  37% |███▋      | [00:43<01:15]
+Warmup:  38% |███▊      | [00:45<01:13]
+Warmup:  38% |███▊      | [00:45<01:13]
+Warmup:  40% |████      | [00:47<01:11]
+Warmup:  40% |████      | [00:47<01:11]
+Warmup:  42% |████▏     | [00:49<01:09]
+Warmup:  42% |████▏     | [00:49<01:09]
+Warmup:  43% |████▎     | [00:51<01:07]
+Warmup:  43% |████▎     | [00:51<01:07]
+Warmup:  45% |████▌     | [00:53<01:05]
+Warmup:  45% |████▌     | [00:53<01:05]
+Warmup:  47% |████▋     | [00:55<01:03]
+Warmup:  47% |████▋     | [00:55<01:03]
+Warmup:  48% |████▊     | [00:57<01:01]
+Warmup:  48% |████▊     | [00:57<01:01]
+Warmup:  50% |█████     | [00:59<00:59]
+Warmup:  50% |█████     | [00:59<00:59]
+Warmup:  52% |█████▏    | [01:01<00:57]
+Warmup:  52% |█████▏    | [01:01<00:57]
+Warmup:  53% |█████▎    | [01:03<00:55]
+Warmup:  53% |█████▎    | [01:03<00:55]
+Warmup:  55% |█████▌    | [01:05<00:53]
+Warmup:  55% |█████▌    | [01:05<00:53]
+Warmup:  57% |█████▋    | [01:07<00:51]
+Warmup:  57% |█████▋    | [01:07<00:51]
+Warmup:  58% |█████▊    | [01:09<00:49]
+Warmup:  58% |█████▊    | [01:09<00:49]
+Warmup:  60% |██████    | [01:11<00:47]
+Warmup:  60% |██████    | [01:11<00:47]
+Warmup:  62% |██████▏   | [01:13<00:45]
+Warmup:  62% |██████▏   | [01:13<00:45]
+Warmup:  63% |██████▎   | [01:15<00:43]
+Warmup:  63% |██████▎   | [01:15<00:43]
+Warmup:  65% |██████▌   | [01:17<00:41]
+Warmup:  65% |██████▌   | [01:17<00:41]
+Warmup:  67% |██████▋   | [01:19<00:39]
+Warmup:  67% |██████▋   | [01:19<00:39]
+Warmup:  68% |██████▊   | [01:21<00:37]
+Warmup:  68% |██████▊   | [01:21<00:37]
+Warmup:  70% |███████   | [01:23<00:35]
+Warmup:  70% |███████   | [01:23<00:35]
+Warmup:  72% |███████▏  | [01:25<00:33]
+Warmup:  72% |███████▏  | [01:25<00:33]
+Warmup:  73% |███████▎  | [01:27<00:31]
+Warmup:  73% |███████▎  | [01:27<00:31]
+Warmup:  75% |███████▌  | [01:29<00:29]
+Warmup:  75% |███████▌  | [01:29<00:29]
+Warmup:  77% |███████▋  | [01:31<00:27]
+Warmup:  77% |███████▋  | [01:31<00:27]
+Warmup:  78% |███████▊  | [01:33<00:25]
+Warmup:  78% |███████▊  | [01:33<00:25]
+Warmup:  80% |████████  | [01:35<00:23]
+Warmup:  80% |████████  | [01:35<00:23]
+Warmup:  82% |████████▏ | [01:37<00:21]
+Warmup:  82% |████████▏ | [01:37<00:21]
+Warmup:  83% |████████▎ | [01:39<00:19]
+Warmup:  83% |████████▎ | [01:39<00:19]
+Warmup:  85% |████████▌ | [01:41<00:17]
+Warmup:  85% |████████▌ | [01:41<00:17]
+Warmup:  87% |████████▋ | [01:43<00:15]
+Warmup:  87% |████████▋ | [01:43<00:15]
+Warmup:  88% |████████▊ | [01:45<00:13]
+Warmup:  88% |████████▊ | [01:45<00:13]
+Warmup:  90% |█████████ | [01:47<00:11]
+Warmup:  90% |█████████ | [01:47<00:11]
+Warmup:  92% |█████████▏| [01:49<00:09]
+Warmup:  92% |█████████▏| [01:49<00:09]
+Warmup:  93% |█████████▎| [01:51<00:07]
+Warmup:  93% |█████████▎| [01:51<00:07]
+Warmup:  95% |█████████▌| [01:53<00:05]
+Warmup:  95% |█████████▌| [01:53<00:05]
+Warmup:  97% |█████████▋| [01:55<00:03]
+Warmup:  97% |█████████▋| [01:55<00:03]
+Warmup:  98% |█████████▊| [01:57<00:01]
+Warmup:  98% |█████████▊| [01:57<00:01]
+Warmup Grace Period: 0/403 |          |   0% [00:00<?]
+Warmup Grace Period: 369/403 |█████████▏|  92% [00:00<00:00]
+Warmup Grace Period: 379/403 |█████████▍|  94% [00:02<00:00]
+Warmup Grace Period: 379/403 |█████████▍|  94% [00:02<00:00]
+Warmup Grace Period: 385/403 |█████████▌|  96% [00:04<00:00]
+Warmup Grace Period: 391/403 |█████████▋|  97% [00:08<00:00]
+Warmup Grace Period: 398/403 |█████████▉|  99% [00:12<00:00]
+Warmup Grace Period: 398/403 |█████████▉|  99% [00:12<00:00]
+Warmup Grace Period: 403/403 |██████████| 100% [00:17<00:00]
+Warmup Grace Period: 403/403 |██████████| 100% [00:17<00:00]
+Processing Records: 0/403 |          |   0% [00:00<?]
+Profiling: 0/1,000,000 |          |   0% [00:00<?]
+Profiling Grace Period: 0/1,835 |          |   0% [00:00<?]
+Profiling Grace Period: 1,802/1,835 |█████████▊|  98% [00:00<00:00]
+Profiling Grace Period: 1,823/1,835 |█████████▉|  99% [00:10<00:00]
+Profiling Grace Period: 1,823/1,835 |█████████▉|  99% [00:10<00:00]
+Processing Records: 0/1,835 |          |   0% [00:00<?]
+Processing Records: 1,835/1,835 |██████████| 100% [00:01<00:00]
+Processing Records: 1,835/1,835 |██████████| 100% [00:01<00:00]

experiments/kv-cache-fp8/capacity/hammer/N0038/manifest.json

@@ -0,0 +1,111 @@
+{
+  "concurrency": 38,
+  "mode": "hammer",
+  "cache_salt": "6b630438f0d1c386ae7365fda9378e4c",
+  "command": [
+    "aiperf",
+    "profile",
+    "--model",
+    "glm-5.2",
+    "--url",
+    "http://127.0.0.1:8391",
+    "--api-key",
+    "<key>",
+    "--endpoint-type",
+    "chat",
+    "--streaming",
+    "--tokenizer",
+    "builtin",
+    "--custom-dataset-type",
+    "dag_jsonl",
+    "--input-file",
+    "data/capacity/corpus.jsonl",
+    "--extra-inputs",
+    "{\"cache_salt\": \"6b630438f0d1c386ae7365fda9378e4c\"}",
+    "--concurrency",
+    "38",
+    "--warmup-duration",
+    "120",
+    "--benchmark-duration",
+    "900",
+    "--benchmark-grace-period",
+    "300",
+    "--request-count",
+    "1000000",
+    "--random-seed",
+    "0",
+    "--use-server-token-count",
+    "--artifact-dir",
+    "results/capacity/sweep-hammer-jul20/N0038",
+    "--export-level",
+    "raw",
+    "--export-http-trace",
+    "--slice-duration",
+    "60",
+    "--server-metrics-formats",
+    "json",
+    "csv",
+    "jsonl",
+    "--ui",
+    "simple"
+  ],
+  "dataset_path": "data/capacity/corpus.jsonl",
+  "signature": {
+    "dataset_sha256": "0327d658a267715aebdab45b67b6e0d324616a0faeec8a75d09b8defcd0b2ac2",
+    "url": "http://127.0.0.1:8391",
+    "model": "glm-5.2",
+    "duration": 900,
+    "warmup": 120,
+    "grace": 300,
+    "seed": 0,
+    "request_ceiling": 1000000,
+    "python": "3.12.3",
+    "aiperf": {
+      "version": "0.11.0",
+      "rev": "73d0524f98f30199ca659a8ab586454a823c887d"
+    },
+    "server": {
+      "version": "0.5.15.post1",
+      "model_path": "/home/jl_fs/models/GLM-5.2-FP8",
+      "served_model_name": "glm-5.2",
+      "tp_size": 8,
+      "dp_size": 8,
+      "enable_cache_report": true,
+      "speculative_algorithm": "EAGLE",
+      "max_running_requests": 48,
+      "context_length": 262144,
+      "kv_cache_dtype": "fp8_e4m3",
+      "mem_fraction_static": 0.872,
+      "scheduler_pids": [
+        82149,
+        82150,
+        82151,
+        82152,
+        82153,
+        82154,
+        82155,
+        82156
+      ],
+      "max_total_num_tokens": 234176,
+      "max_model_len": 262144,
+      "model_id": "glm-5.2"
+    },
+    "evaluator_sha256": "3e48dfede43cbf47244f560547eda95d29c671d4fa23b203a5d220a14edf1c80",
+    "verifier_sha256": "6f84bee84368a72ea1445d0961185a0d5d2fd6ba8f8ec4c3fefac619b46a32aa",
+    "runner_sha256": "aa3e1a0f66bcfac6663b5d2e59a37cae9d7264dda6b4c13130a3edf403e77104"
+  },
+  "started_utc": "2026-07-21T01:18:01Z",
+  "loadavg_start": [
+    1.6279296875,
+    1.77490234375,
+    1.72412109375
+  ],
+  "status": "ok",
+  "elapsed_s": 1070.7,
+  "loadavg_end": [
+    1.51513671875,
+    1.796875,
+    1.77783203125
+  ],
+  "returncode": 0
+}

experiments/kv-cache-fp8/capacity/hammer/N0038/profile_export_aiperf.csv

@@ -0,0 +1,68 @@
+Metric,avg,min,max,sum,p1,p5,p10,p25,p50,p75,p90,p95,p99,std
+Active Decode Throughput (tokens/sec),749.48,0.00,87579.93,,,,,,202.34,,1700.82,1787.81,3763.49,1342.60
+Active Decode Throughput Per User (tokens/sec/user),24.22,0.00,18104.82,,,,,,6.23,,51.70,53.81,118.78,44.93
+Active Prefill Throughput (tokens/sec),50318.09,885.97,264160.58,,,,,,39616.74,,111744.81,135191.17,176310.73,44156.18
+Active Prefill Throughput Per User (tokens/sec/user),16957.59,885.97,89196.88,,,,,,13946.93,,34288.81,39759.06,55222.21,13107.70
+Active Total Throughput (tokens/sec),48738.67,49.82,264815.60,,,,,,36553.61,,110647.52,134301.92,176765.59,44389.37
+Credit-to-Start Latency (ms),1.74,0.56,22.65,3186.02,0.87,1.03,1.10,1.21,1.35,1.55,1.90,2.65,15.33,2.12
+E2E Output Token Throughput (tokens/sec/user),20.97,2.10,51.82,38471.34,5.69,8.63,10.28,17.30,22.40,25.31,27.81,29.73,35.33,6.75
+Effective Concurrency,33.70,0.00,38.00,,,,,,34.00,,37.00,38.00,38.00,3.83
+Effective Decode Concurrency,30.87,0.00,37.00,,,,,,31.00,,34.00,35.00,36.00,4.03
+Effective Decode Throughput (tokens/sec),747.09,0.00,87579.93,,,,,,202.07,,1700.23,1787.48,3742.12,1341.12
+Effective Decode Throughput Per User (tokens/sec/user),24.14,0.00,18104.82,,,,,,6.23,,51.69,53.80,118.69,44.88
+Effective Latency (CO-aware) (ms),16848.34,1094.78,127586.08,30916709.50,1623.24,2376.38,3029.67,4788.53,8719.87,21918.48,46215.88,56440.58,74915.13,18045.59
+Effective Prefill Concurrency,2.82,0.00,38.00,,,,,,2.00,,5.00,6.00,9.00,2.72
+Effective Prefill Throughput (tokens/sec),47990.56,0.00,264160.58,,,,,,35842.58,,109962.49,133498.16,176015.28,44399.08
+Effective Prefill Throughput Per User (tokens/sec/user),16173.20,0.00,89196.88,,,,,,12645.78,,33870.13,39547.55,53833.32,13287.23
+Effective Total Throughput (tokens/sec),48737.65,0.00,264815.60,,,,,,36533.95,,110647.52,134301.92,176765.59,44389.47
+HTTP Blocked (ms),0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00
+HTTP Chunks Received,200.79,17.00,1752.00,368450.00,20.00,28.00,31.00,44.00,78.00,247.00,610.20,725.00,941.60,244.29
+HTTP Chunks Sent,1.00,1.00,1.00,1835.00,1.00,1.00,1.00,1.00,1.00,1.00,1.00,1.00,1.00,0.00
+HTTP Connecting (ms),0.10,0.00,12.67,190.60,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,3.16,0.70
+HTTP Connection Overhead (ms),0.10,0.00,12.67,190.60,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,3.16,0.70
+HTTP Connection Reused (ratio),0.97,0.00,1.00,1771.00,0.00,1.00,1.00,1.00,1.00,1.00,1.00,1.00,1.00,0.18
+HTTP Data Received (KB),49.13,4.56,414.78,90161.06,5.56,7.32,8.45,11.59,20.01,60.66,146.23,173.49,224.43,58.32
+HTTP Data Sent (KB),99.70,36.74,227.03,182951.36,37.59,41.15,45.29,64.24,93.34,130.08,163.91,180.07,200.56,43.49
+HTTP DNS Lookup (ms),0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00
+HTTP Duration (excl. conn) (ms),16846.47,1092.93,127584.55,30913278.08,1621.69,2375.09,3027.19,4787.38,8718.81,21916.85,46214.29,56438.79,74913.42,18045.66
+HTTP Receiving (ms),15471.30,653.53,127194.09,28389843.87,1061.28,1712.07,2268.17,3734.93,6869.47,20033.68,45285.46,55078.82,72024.43,18045.85
+HTTP Sending (ms),0.38,0.13,13.24,689.90,0.19,0.20,0.21,0.22,0.25,0.30,0.37,0.47,3.46,0.74
+HTTP Total Time (ms),16846.58,1092.93,127584.55,30913468.68,1622.64,2375.09,3027.19,4787.38,8718.81,21916.85,46214.29,56438.79,74913.42,18045.64
+HTTP Waiting (TTFB) (ms),1374.79,317.29,26271.38,2522744.31,347.23,375.83,400.44,487.41,690.10,1084.39,3535.00,5004.84,10406.35,1993.29
+Input Sequence Length (tokens),23992.62,8085.00,56709.00,44026463.00,8242.10,9208.10,10450.40,15396.50,22683.00,31444.50,39679.20,43533.30,49075.84,10676.69
+Inter Chunk Latency (ms),77.23,0.00,3463.73,28251659.36,1.58,2.60,4.88,35.01,36.84,46.84,300.25,325.08,614.33,130.86
+Inter Token Latency (ms),44.83,15.12,131.56,82266.48,18.12,26.06,29.43,35.24,40.04,45.96,71.19,90.90,109.72,18.50
+OSL Mismatch Diff % (%),-71.09,-98.10,0.00,-130444.81,-96.81,-95.52,-94.79,-92.93,-87.88,-63.84,0.00,0.00,0.00,32.78
+Output Sequence Length (tokens),374.50,32.00,3291.00,687214.00,38.00,53.00,61.00,83.00,144.00,455.50,1134.60,1363.10,1770.88,457.15
+Output Token Count (tokens),374.50,32.00,3291.00,687214.00,38.00,53.00,61.00,83.00,144.00,455.50,1134.60,1363.10,1770.88,457.15
+Output Token Throughput Per User (tokens/sec/user),25.18,7.60,66.12,46209.08,9.11,11.00,14.05,21.76,24.98,28.38,33.98,38.37,55.18,8.17
+Prefill Throughput Per User (tokens/sec/user),33684.60,411.20,118483.21,61811232.01,1608.71,2873.01,5293.36,14280.72,29410.29,48753.90,69028.45,80199.73,97707.05,24022.73
+Request Latency (ms),16806.98,1093.06,127584.69,30840808.02,1603.24,2344.81,2987.81,4752.45,8674.77,21898.54,46214.42,56438.95,74913.55,18056.85
+Time to First Output Token (ms),7720.60,372.69,92473.96,12708115.28,398.58,447.91,535.37,915.26,2891.77,7990.94,23781.50,34222.63,57958.67,11993.33
+Time to First Token (ms),1410.98,317.70,26271.74,2589148.67,352.35,387.42,410.42,517.89,718.66,1191.10,3540.68,5005.25,10406.72,1990.34
+Time to Second Token (ms),111.55,0.00,2625.39,204698.57,0.09,1.24,2.36,33.05,36.88,64.36,325.81,441.39,882.86,212.33
+Tokens In Flight (tokens),916811.87,0.00,1318476.16,,,,,,948405.31,,1226890.70,1252712.53,1291951.64,266043.93
+Usage Completion Tokens (tokens),374.50,32.00,3291.00,687214.00,38.00,53.00,61.00,83.00,144.00,455.50,1134.60,1363.10,1770.88,457.15
+Usage Prompt Cache Read Tokens (tokens),23179.21,4032.00,54784.00,42348416.00,7744.00,8384.00,9958.40,14144.00,22080.00,30656.00,38784.00,42624.00,48570.88,10681.80
+Usage Prompt Tokens (tokens),23992.62,8085.00,56709.00,44026463.00,8242.10,9208.10,10450.40,15396.50,22683.00,31444.50,39679.20,43533.30,49075.84,10676.69
+Usage Total Tokens (tokens),24367.13,8162.00,57487.00,44713677.00,8327.10,9272.10,10555.00,15704.50,23002.00,32098.00,40191.80,44021.40,49679.18,10828.79
+
+Metric,Value
+Benchmark Duration (sec),917.36
+Input Token Throughput (tokens/sec),47990.56
+OSL Mismatch Count,1602.00
+Output Token Throughput (tokens/sec),749.09
+Overall Usage Prompt Cache Read % (%),96.19
+Request Count,1835.00
+Request Throughput (requests/sec),2.00
+Total Input Sequence Length (tokens),44026463.00
+Total Output Sequence Length (tokens),687214.00
+Total Output Tokens (tokens),687214.00
+Total Token Throughput (tokens/sec),48739.65
+Total Usage Completion Tokens (tokens),687214.00
+Total Usage Prompt Cache Read Tokens (tokens),42348416.00
+Total Usage Prompt Tokens (tokens),44026463.00
+Total Usage Total Tokens (tokens),44713677.00
+
+
+Endpoint,GPU_Index,GPU_Name,GPU_UUID,Metric,avg,min,max,sum,p1,p5,p10,p25,p50,p75,p90,p95,p99,std

experiments/kv-cache-fp8/capacity/hammer/N0038/profile_export_aiperf.json

@@ -0,0 +1,1726 @@
+{
+  "schema_version": "1.4",
+  "aiperf_version": "0.11.0",
+  "benchmark_id": "d86e6a4259f1",
+  "request_throughput": {
+    "unit": "requests/sec",
+    "avg": 2.000221445120071
+  },
+  "request_latency": {
+    "unit": "ms",
+    "avg": 16806.979849603267,
+    "p1": 1603.23827488,
+    "p5": 2344.8141203,
+    "p10": 2987.8110285999996,
+    "p25": 4752.4537789999995,
+    "p50": 8674.772751,
+    "p75": 21898.543643499997,
+    "p90": 46214.41715340001,
+    "p95": 56438.94654889999,
+    "p99": 74913.54827772004,
+    "min": 1093.055823,
+    "max": 127584.685527,
+    "std": 18056.85029188808,
+    "count": 1835,
+    "sum": 30840808.024021998
+  },
+  "request_count": {
+    "unit": "requests",
+    "avg": 1835.0
+  },
+  "time_to_first_token": {
+    "unit": "ms",
+    "avg": 1410.9802003520435,
+    "p1": 352.34842983999994,
+    "p5": 387.4212857,
+    "p10": 410.4191102,
+    "p25": 517.893278,
+    "p50": 718.662834,
+    "p75": 1191.104313,
+    "p90": 3540.6839526,
+    "p95": 5005.245853699999,
+    "p99": 10406.717167440042,
+    "min": 317.699413,
+    "max": 26271.740347,
+    "std": 1990.3361552801002,
+    "count": 1835,
+    "sum": 2589148.6676459997
+  },
+  "time_to_second_token": {
+    "unit": "ms",
+    "avg": 111.55235221362398,
+    "p1": 0.08712175999999999,
+    "p5": 1.2414689,
+    "p10": 2.3555732000000003,
+    "p25": 33.051297,
+    "p50": 36.881087,
+    "p75": 64.35956999999999,
+    "p90": 325.8056698,
+    "p95": 441.3865604999998,
+    "p99": 882.8592712799999,
+    "min": 0.0,
+    "max": 2625.3937339999998,
+    "std": 212.32615995707707,
+    "count": 1835,
+    "sum": 204698.56631199998
+  },
+  "inter_token_latency": {
+    "unit": "ms",
+    "avg": 44.83186806394337,
+    "p1": 18.124280478,
+    "p5": 26.060763992382224,
+    "p10": 29.431966330650994,
+    "p25": 35.24163887147379,
+    "p50": 40.03854213533834,
+    "p75": 45.964908444148485,
+    "p90": 71.19446020027638,
+    "p95": 90.89730554635572,
+    "p99": 109.7228376921206,
+    "min": 15.124286945945945,
+    "max": 131.56233059210527,
+    "std": 18.498670198237335,
+    "count": 1835,
+    "sum": 82266.47789733608
+  },
+  "output_token_throughput": {
+    "unit": "tokens/sec",
+    "avg": 749.0900164505418
+  },
+  "output_token_throughput_per_user": {
+    "unit": "tokens/sec/user",
+    "avg": 25.18206036044854,
+    "p1": 9.113874130007352,
+    "p5": 11.001429957016283,
+    "p10": 14.046037899318875,
+    "p25": 21.755729288834175,
+    "p50": 24.97593435394821,
+    "p75": 28.37552582694066,
+    "p90": 33.9766641105268,
+    "p95": 38.371941473669224,
+    "p99": 55.176029353670636,
+    "min": 7.600959906224156,
+    "max": 66.11881958957736,
+    "std": 8.16871979472531,
+    "count": 1835,
+    "sum": 46209.080761423065
+  },
+  "output_sequence_length": {
+    "unit": "tokens",
+    "avg": 374.50354223433243,
+    "p1": 38.0,
+    "p5": 53.0,
+    "p10": 61.0,
+    "p25": 83.0,
+    "p50": 144.0,
+    "p75": 455.5,
+    "p90": 1134.6000000000001,
+    "p95": 1363.0999999999997,
+    "p99": 1770.8800000000015,
+    "min": 32.0,
+    "max": 3291.0,
+    "std": 457.14648093179557,
+    "count": 1835,
+    "sum": 687214.0
+  },
+  "input_sequence_length": {
+    "unit": "tokens",
+    "avg": 23992.62288828338,
+    "p1": 8242.1,
+    "p5": 9208.1,
+    "p10": 10450.4,
+    "p25": 15396.5,
+    "p50": 22683.0,
+    "p75": 31444.5,
+    "p90": 39679.200000000004,
+    "p95": 43533.299999999996,
+    "p99": 49075.840000000004,
+    "min": 8085.0,
+    "max": 56709.0,
+    "std": 10676.691588236537,
+    "count": 1835,
+    "sum": 44026463.0
+  },
+  "output_token_count": {
+    "unit": "tokens",
+    "avg": 374.50354223433243,
+    "p1": 38.0,
+    "p5": 53.0,
+    "p10": 61.0,
+    "p25": 83.0,
+    "p50": 144.0,
+    "p75": 455.5,
+    "p90": 1134.6000000000001,
+    "p95": 1363.0999999999997,
+    "p99": 1770.8800000000015,
+    "min": 32.0,
+    "max": 3291.0,
+    "std": 457.14648093179557,
+    "count": 1835,
+    "sum": 687214.0
+  },
+  "inter_chunk_latency": {
+    "unit": "ms",
+    "avg": 77.2285183557333,
+    "p1": 1.5752496200000001,
+    "p5": 2.6046449,
+    "p10": 4.8764796,
+    "p25": 35.007186499999996,
+    "p50": 36.840396,
+    "p75": 46.838338,
+    "p90": 300.2537938,
+    "p95": 325.08250769999995,
+    "p99": 614.32558468,
+    "min": 0.0,
+    "max": 3463.73497,
+    "std": 130.86416354309938,
+    "count": 365819,
+    "sum": 28251659.356376
+  },
+  "total_output_tokens": {
+    "unit": "tokens",
+    "avg": 687214.0
+  },
+  "benchmark_duration": {
+    "unit": "sec",
+    "avg": 917.3585139200001
+  },
+  "total_isl": {
+    "unit": "tokens",
+    "avg": 44026463.0
+  },
+  "total_osl": {
+    "unit": "tokens",
+    "avg": 687214.0
+  },
+  "telemetry_data": {
+    "summary": {
+      "endpoints_configured": [],
+      "endpoints_successful": [],
+      "start_time": "2026-07-21T01:20:26.544587",
+      "end_time": "2026-07-21T01:35:47.278375"
+    },
+    "endpoints": {},
+    "error_summary": []
+  },
+  "input_config": {
+    "models": {
+      "items": [
+        {
+          "name": "glm-5.2"
+        }
+      ]
+    },
+    "endpoint": {
+      "urls": [
+        "http://127.0.0.1:8391"
+      ],
+      "type": "chat",
+      "api_key": "<redacted>",
+      "streaming": true,
+      "use_server_token_count": true,
+      "extra": {
+        "cache_salt": "6b630438f0d1c386ae7365fda9378e4c"
+      }
+    },
+    "datasets": [
+      {
+        "name": "main",
+        "type": "file",
+        "path": "data/capacity/corpus.jsonl",
+        "format": "dag_jsonl",
+        "entries": 1000000,
+        "random_seed": 0
+      }
+    ],
+    "phases": [
+      {
+        "name": "warmup",
+        "type": "concurrency",
+        "exclude_from_results": true,
+        "duration": 120.0,
+        "concurrency": 38
+      },
+      {
+        "name": "profiling",
+        "type": "concurrency",
+        "requests": 1000000,
+        "duration": 900.0,
+        "concurrency": 38,
+        "grace_period": 300.0
+      }
+    ],
+    "artifacts": {
+      "dir": "/home/ubuntu/workspace/projects/glm-5.2-experiments/results/capacity/sweep-hammer-jul20/N0038",
+      "records": [
+        "jsonl"
+      ],
+      "raw": true,
+      "trace": true,
+      "slice_duration": 60.0,
+      "auto_plot": false,
+      "plot_required": false
+    },
+    "tokenizer": {
+      "name": "builtin"
+    },
+    "gpu_telemetry": {
+      "enabled": true
+    },
+    "server_metrics": {
+      "enabled": true,
+      "urls": [],
+      "formats": [
+        "json",
+        "csv",
+        "jsonl"
+      ]
+    },
+    "network_latency": {
+      "enabled": false
+    },
+    "otel": {},
+    "mlflow": {},
+    "wandb": {},
+    "runtime": {
+      "ui": "simple"
+    }
+  },
+  "run_info": {
+    "benchmark_id": "d86e6a4259f1",
+    "random_seed": 0,
+    "trial": 0,
+    "cli_command": "aiperf profile --model 'glm-5.2' --url 'http://127.0.0.1:8391' --api-key '<redacted>' --endpoint-type 'chat' --streaming --tokenizer 'builtin' --custom-dataset-type 'dag_jsonl' --input-file 'data/capacity/corpus.jsonl' --extra-inputs '{\"cache_salt\": \"6b630438f0d1c386ae7365fda9378e4c\"}' --concurrency 38 --warmup-duration 120 --benchmark-duration 900 --benchmark-grace-period 300 --request-count 1000000 --random-seed 0 --use-server-token-count --artifact-dir 'results/capacity/sweep-hammer-jul20/N0038' --export-level 'raw' --export-http-trace --slice-duration 60 --server-metrics-formats 'json' 'csv' 'jsonl' --ui 'simple'"
+  },
+  "was_cancelled": false,
+  "error_summary": [],
+  "start_time": "2026-07-21T01:20:26.544587",
+  "end_time": "2026-07-21T01:35:43.943011",
+  "warmup_metrics": {
+    "usage_completion_tokens": {
+      "unit": "tokens",
+      "avg": 199.85111662531017,
+      "p1": 37.06,
+      "p5": 48.0,
+      "p10": 56.0,
+      "p25": 66.0,
+      "p50": 90.0,
+      "p75": 173.5,
+      "p90": 511.0000000000001,
+      "p95": 795.9,
+      "p99": 1400.4200000000014,
+      "min": 29.0,
+      "max": 1726.0,
+      "std": 279.90704353382205,
+      "count": 403,
+      "sum": 80540.0
+    },
+    "time_to_first_token": {
+      "unit": "ms",
+      "avg": 1113.04228864268,
+      "p1": 348.83775957999995,
+      "p5": 376.8067971,
+      "p10": 407.956843,
+      "p25": 567.5539014999999,
+      "p50": 723.7650659999999,
+      "p75": 1028.126747,
+      "p90": 2128.9870386000025,
+      "p95": 4304.4678281999995,
+      "p99": 4616.582786859999,
+      "min": 328.197112,
+      "max": 7133.865379999999,
+      "std": 1103.2989432177908,
+      "count": 403,
+      "sum": 448556.042323
+    },
+    "http_req_receiving": {
+      "unit": "ms",
+      "avg": 9982.05443166005,
+      "p1": 1264.0134534000001,
+      "p5": 2122.9340174,
+      "p10": 2822.5463014,
+      "p25": 3942.8484009999997,
+      "p50": 5932.35368,
+      "p75": 9842.117873,
+      "p90": 21969.0645728,
+      "p95": 34531.27642639999,
+      "p99": 60517.06729306001,
+      "min": 709.285396,
+      "max": 82652.654499,
+      "std": 11545.799638331106,
+      "count": 403,
+      "sum": 4022767.9359589997
+    },
+    "usage_prompt_cache_read_tokens": {
+      "unit": "tokens",
+      "avg": 12703.270886075949,
+      "p1": 7744.0,
+      "p5": 7744.0,
+      "p10": 8256.0,
+      "p25": 9728.0,
+      "p50": 11840.0,
+      "p75": 15392.0,
+      "p90": 18534.4,
+      "p95": 20755.199999999997,
+      "p99": 22223.36,
+      "min": 7744.0,
+      "max": 23104.0,
+      "std": 3910.7854855466703,
+      "count": 395,
+      "sum": 5017792.0
+    },
+    "request_latency": {
+      "unit": "ms",
+      "avg": 10993.145965655087,
+      "p1": 1758.8853366,
+      "p5": 3030.2009823,
+      "p10": 3848.4776318,
+      "p25": 5065.824301,
+      "p50": 6930.60247,
+      "p75": 11516.3221965,
+      "p90": 23627.081123,
+      "p95": 35433.80139589999,
+      "p99": 61138.918129460006,
+      "min": 1302.8343539999998,
+      "max": 84163.79482299999,
+      "std": 11543.44645163668,
+      "count": 403,
+      "sum": 4430237.824159
+    },
+    "usage_prompt_tokens": {
+      "unit": "tokens",
+      "avg": 13577.615384615385,
+      "p1": 8206.18,
+      "p5": 8439.3,
+      "p10": 8778.2,
+      "p25": 10365.5,
+      "p50": 12693.0,
+      "p75": 16362.5,
+      "p90": 19516.800000000003,
+      "p95": 21553.8,
+      "p99": 23142.780000000002,
+      "min": 8106.0,
+      "max": 26886.0,
+      "std": 3995.354427451259,
+      "count": 403,
+      "sum": 5471779.0
+    },
+    "output_sequence_length": {
+      "unit": "tokens",
+      "avg": 199.85111662531017,
+      "p1": 37.06,
+      "p5": 48.0,
+      "p10": 56.0,
+      "p25": 66.0,
+      "p50": 90.0,
+      "p75": 173.5,
+      "p90": 511.0000000000001,
+      "p95": 795.9,
+      "p99": 1400.4200000000014,
+      "min": 29.0,
+      "max": 1726.0,
+      "std": 279.90704353382205,
+      "count": 403,
+      "sum": 80540.0
+    },
+    "request_count": {
+      "unit": "requests",
+      "avg": 403.0
+    },
+    "time_to_second_token": {
+      "unit": "ms",
+      "avg": 179.00311202481387,
+      "p1": 0.772209,
+      "p5": 2.6299342,
+      "p10": 19.2018448,
+      "p25": 34.2477975,
+      "p50": 41.493106999999995,
+      "p75": 311.9487315,
+      "p90": 579.0816314000003,
+      "p95": 872.9650795999974,
+      "p99": 894.71884646,
+      "min": 0.0,
+      "max": 1606.391068,
+      "std": 245.67451370703904,
+      "count": 403,
+      "sum": 72138.25414599999
+    },
+    "http_req_data_sent": {
+      "unit": "KB",
+      "avg": 57.2249752830335,
+      "p1": 37.30884765625,
+      "p5": 38.298046875,
+      "p10": 39.6384765625,
+      "p25": 45.09326171875,
+      "p50": 53.142578125,
+      "p75": 66.91015625,
+      "p90": 78.95058593750001,
+      "p95": 88.43105468749998,
+      "p99": 94.52347656250001,
+      "min": 36.841796875,
+      "max": 97.9873046875,
+      "std": 15.1392635383503,
+      "count": 403,
+      "sum": 23061.6650390625
+    },
+    "http_req_blocked": {
+      "unit": "ms",
+      "avg": 0.0,
+      "p1": 0.0,
+      "p5": 0.0,
+      "p10": 0.0,
+      "p25": 0.0,
+      "p50": 0.0,
+      "p75": 0.0,
+      "p90": 0.0,
+      "p95": 0.0,
+      "p99": 0.0,
+      "min": 0.0,
+      "max": 0.0,
+      "std": 0.0,
+      "count": 403,
+      "sum": 0.0
+    },
+    "http_req_connection_reused": {
+      "unit": "ratio",
+      "avg": 0.8957816377171216,
+      "p1": 0.0,
+      "p5": 0.0,
+      "p10": 0.0,
+      "p25": 1.0,
+      "p50": 1.0,
+      "p75": 1.0,
+      "p90": 1.0,
+      "p95": 1.0,
+      "p99": 1.0,
+      "min": 0.0,
+      "max": 1.0,
+      "std": 0.3055436061284102,
+      "count": 403,
+      "sum": 361.0
+    },
+    "http_req_waiting": {
+      "unit": "ms",
+      "avg": 1073.9289892059553,
+      "p1": 345.91586151999996,
+      "p5": 369.40569739999995,
+      "p10": 394.77110839999995,
+      "p25": 551.657736,
+      "p50": 700.588406,
+      "p75": 953.4052664999999,
+      "p90": 1892.7084790000017,
+      "p95": 4302.3470696,
+      "p99": 4608.99182284,
+      "min": 327.837022,
+      "max": 7131.746576,
+      "std": 1104.6103297459179,
+      "count": 403,
+      "sum": 432793.38265
+    },
+    "http_req_dns_lookup": {
+      "unit": "ms",
+      "avg": 0.0,
+      "p1": 0.0,
+      "p5": 0.0,
+      "p10": 0.0,
+      "p25": 0.0,
+      "p50": 0.0,
+      "p75": 0.0,
+      "p90": 0.0,
+      "p95": 0.0,
+      "p99": 0.0,
+      "min": 0.0,
+      "max": 0.0,
+      "std": 0.0,
+      "count": 403,
+      "sum": 0.0
+    },
+    "http_req_connecting": {
+      "unit": "ms",
+      "avg": 0.401240429280397,
+      "p1": 0.0,
+      "p5": 0.0,
+      "p10": 0.0,
+      "p25": 0.0,
+      "p50": 0.0,
+      "p75": 0.0,
+      "p90": 0.544539600000001,
+      "p95": 2.4571782999999994,
+      "p99": 8.576521760000006,
+      "min": 0.0,
+      "max": 11.453522,
+      "std": 1.547819724892917,
+      "count": 403,
+      "sum": 161.699893
+    },
+    "http_req_sending": {
+      "unit": "ms",
+      "avg": 0.6737582406947891,
+      "p1": 0.18291382,
+      "p5": 0.2018072,
+      "p10": 0.21113579999999998,
+      "p25": 0.22816699999999998,
+      "p50": 0.25404099999999996,
+      "p75": 0.300855,
+      "p90": 0.7507652000000015,
+      "p95": 2.7119620999999996,
+      "p99": 9.580647160000009,
+      "min": 0.09967999999999999,
+      "max": 12.250055,
+      "std": 1.6212920791333696,
+      "count": 403,
+      "sum": 271.524571
+    },
+    "output_token_count": {
+      "unit": "tokens",
+      "avg": 199.85111662531017,
+      "p1": 37.06,
+      "p5": 48.0,
+      "p10": 56.0,
+      "p25": 66.0,
+      "p50": 90.0,
+      "p75": 173.5,
+      "p90": 511.0000000000001,
+      "p95": 795.9,
+      "p99": 1400.4200000000014,
+      "min": 29.0,
+      "max": 1726.0,
+      "std": 279.90704353382205,
+      "count": 403,
+      "sum": 80540.0
+    },
+    "http_req_data_received": {
+      "unit": "KB",
+      "avg": 26.818223674007445,
+      "p1": 5.102363281250001,
+      "p5": 6.591992187500001,
+      "p10": 7.8064453125,
+      "p25": 9.3232421875,
+      "p50": 12.4482421875,
+      "p75": 23.5400390625,
+      "p90": 66.96171875,
+      "p95": 106.95546875,
+      "p99": 181.15875000000008,
+      "min": 4.3095703125,
+      "max": 223.5751953125,
+      "std": 36.33429105578327,
+      "count": 403,
+      "sum": 10807.744140625
+    },
+    "input_sequence_length": {
+      "unit": "tokens",
+      "avg": 13577.615384615385,
+      "p1": 8206.18,
+      "p5": 8439.3,
+      "p10": 8778.2,
+      "p25": 10365.5,
+      "p50": 12693.0,
+      "p75": 16362.5,
+      "p90": 19516.800000000003,
+      "p95": 21553.8,
+      "p99": 23142.780000000002,
+      "min": 8106.0,
+      "max": 26886.0,
+      "std": 3995.354427451259,
+      "count": 403,
+      "sum": 5471779.0
+    },
+    "usage_total_tokens": {
+      "unit": "tokens",
+      "avg": 13777.466501240695,
+      "p1": 8268.1,
+      "p5": 8525.5,
+      "p10": 8846.2,
+      "p25": 10451.5,
+      "p50": 12861.0,
+      "p75": 16515.5,
+      "p90": 19791.800000000003,
+      "p95": 21699.699999999997,
+      "p99": 23359.72,
+      "min": 8190.0,
+      "max": 28046.0,
+      "std": 4090.956839393574,
+      "count": 403,
+      "sum": 5552319.0
+    },
+    "http_req_chunks_received": {
+      "unit": "count",
+      "avg": 108.08684863523573,
+      "p1": 19.04,
+      "p5": 25.1,
+      "p10": 29.0,
+      "p25": 35.0,
+      "p50": 48.0,
+      "p75": 94.0,
+      "p90": 277.0,
+      "p95": 443.9,
+      "p99": 760.6200000000003,
+      "min": 15.0,
+      "max": 935.0,
+      "std": 152.39106641246056,
+      "count": 403,
+      "sum": 43559.0
+    },
+    "time_to_first_output_token": {
+      "unit": "ms",
+      "avg": 6081.411146265822,
+      "p1": 403.0839858,
+      "p5": 519.6475859,
+      "p10": 689.2793514,
+      "p25": 1189.2223179999999,
+      "p50": 2449.897042,
+      "p75": 5908.6838705,
+      "p90": 15625.26247740001,
+      "p95": 28150.18890349999,
+      "p99": 43620.10087178001,
+      "min": 372.610217,
+      "max": 76259.01030699999,
+      "std": 9429.537219729886,
+      "count": 395,
+      "sum": 2402157.402775
+    },
+    "http_req_chunks_sent": {
+      "unit": "count",
+      "avg": 1.0,
+      "p1": 1.0,
+      "p5": 1.0,
+      "p10": 1.0,
+      "p25": 1.0,
+      "p50": 1.0,
+      "p75": 1.0,
+      "p90": 1.0,
+      "p95": 1.0,
+      "p99": 1.0,
+      "min": 1.0,
+      "max": 1.0,
+      "std": 0.0,
+      "count": 403,
+      "sum": 403.0
+    },
+    "http_req_duration": {
+      "unit": "ms",
+      "avg": 11056.657179106698,
+      "p1": 1783.2044162400002,
+      "p5": 3048.5731035,
+      "p10": 3862.2437004,
+      "p25": 5181.653335,
+      "p50": 7063.506315,
+      "p75": 11551.00225,
+      "p90": 24104.662629200007,
+      "p95": 35433.70333399999,
+      "p99": 61138.7617488,
+      "min": 1316.5329669999999,
+      "max": 84163.67156999999,
+      "std": 11534.936128224701,
+      "count": 403,
+      "sum": 4455832.84318
+    },
+    "inter_token_latency": {
+      "unit": "ms",
+      "avg": 58.678618826766545,
+      "p1": 16.383580906401416,
+      "p5": 26.01792389136116,
+      "p10": 33.18040442990153,
+      "p25": 42.129239630256414,
+      "p50": 59.02800360743801,
+      "p75": 72.78947436745717,
+      "p90": 84.099104375965,
+      "p95": 96.03347058135262,
+      "p99": 105.64387055312793,
+      "min": 14.162651781102362,
+      "max": 111.75416423404255,
+      "std": 20.469238509824745,
+      "count": 403,
+      "sum": 23647.48338718692
+    },
+    "e2e_output_token_throughput": {
+      "unit": "tokens/sec/user",
+      "avg": 16.462123957202174,
+      "p1": 8.291775816105691,
+      "p5": 9.533586934046074,
+      "p10": 10.522010715876773,
+      "p25": 11.821880397487307,
+      "p50": 13.332958808051348,
+      "p75": 20.168321535413273,
+      "p90": 24.215257855282186,
+      "p95": 28.280806470728447,
+      "p99": 47.826370843310656,
+      "min": 6.394611661636756,
+      "max": 67.1059528144612,
+      "std": 7.770441413651017,
+      "count": 403,
+      "sum": 6634.235954752477
+    },
+    "osl_mismatch_diff_pct": {
+      "unit": "%",
+      "avg": -83.80463670432609,
+      "p1": -96.69293478260869,
+      "p5": -95.84372114496767,
+      "p10": -94.96059496059496,
+      "p25": -94.06474820143885,
+      "p50": -91.88458070333634,
+      "p75": -85.33368230944637,
+      "p90": -62.25506114623186,
+      "p95": -31.85025817555946,
+      "p99": 0.0,
+      "min": -97.31481481481481,
+      "max": 0.0,
+      "std": 20.480105100830695,
+      "count": 403,
+      "sum": -33773.268591843414
+    },
+    "http_req_connection_overhead": {
+      "unit": "ms",
+      "avg": 0.401240429280397,
+      "p1": 0.0,
+      "p5": 0.0,
+      "p10": 0.0,
+      "p25": 0.0,
+      "p50": 0.0,
+      "p75": 0.0,
+      "p90": 0.544539600000001,
+      "p95": 2.4571782999999994,
+      "p99": 8.576521760000006,
+      "min": 0.0,
+      "max": 11.453522,
+      "std": 1.547819724892917,
+      "count": 403,
+      "sum": 161.699893
+    },
+    "http_req_total": {
+      "unit": "ms",
+      "avg": 11057.058419535979,
+      "p1": 1783.2044162400002,
+      "p5": 3048.5731035,
+      "p10": 3862.2437004,
+      "p25": 5181.653335,
+      "p50": 7063.506315,
+      "p75": 11551.00225,
+      "p90": 24104.662629200007,
+      "p95": 35433.70333399999,
+      "p99": 61138.7617488,
+      "min": 1317.907917,
+      "max": 84163.67156999999,
+      "std": 11534.903422399799,
+      "count": 403,
+      "sum": 4455994.543072999
+    },
+    "prefill_throughput_per_user": {
+      "unit": "tokens/sec/user",
+      "avg": 19390.209388769108,
+      "p1": 1889.6131530121422,
+      "p5": 2272.542838016498,
+      "p10": 6317.701040036634,
+      "p25": 12249.251725786193,
+      "p50": 17135.880976291366,
+      "p75": 25227.625228412147,
+      "p90": 34660.90376351837,
+      "p95": 42026.9283907021,
+      "p99": 55305.29490229913,
+      "min": 1354.2903100036733,
+      "max": 63566.221559767946,
+      "std": 11678.492317482647,
+      "count": 403,
+      "sum": 7814254.383673951
+    },
+    "output_token_throughput_per_user": {
+      "unit": "tokens/sec/user",
+      "avg": 19.91614743394274,
+      "p1": 9.465851695177074,
+      "p5": 10.41329680498713,
+      "p10": 11.89073327654203,
+      "p25": 13.738263350113733,
+      "p50": 16.941111656942294,
+      "p75": 23.73653278035166,
+      "p90": 30.13874641607534,
+      "p95": 38.438818085125526,
+      "p99": 61.03728821630048,
+      "min": 8.948212416547964,
+      "max": 70.60824593133957,
+      "std": 9.508023411066853,
+      "count": 403,
+      "sum": 8026.207415878924
+    },
+    "osl_mismatch_count": {
+      "unit": "requests",
+      "avg": 393.0
+    },
+    "inter_chunk_latency": {
+      "unit": "ms",
+      "avg": 92.73959523538454,
+      "p1": 1.23786879,
+      "p5": 2.8715979999999997,
+      "p10": 13.6028853,
+      "p25": 31.830987999999998,
+      "p50": 36.519991,
+      "p75": 49.597293,
+      "p90": 306.8501324,
+      "p95": 347.1155494999997,
+      "p99": 906.2951814599999,
+      "min": 0.0,
+      "max": 1792.5170759999999,
+      "std": 175.67271900850488,
+      "count": 42934,
+      "sum": 3981681.781836
+    },
+    "total_usage_completion_tokens": {
+      "unit": "tokens",
+      "avg": 80540.0
+    },
+    "total_output_tokens": {
+      "unit": "tokens",
+      "avg": 80540.0
+    },
+    "total_usage_prompt_cache_read_tokens": {
+      "unit": "tokens",
+      "avg": 5017792.0
+    },
+    "total_usage_prompt_tokens": {
+      "unit": "tokens",
+      "avg": 5471779.0
+    },
+    "total_isl": {
+      "unit": "tokens",
+      "avg": 5471779.0
+    },
+    "total_usage_total_tokens": {
+      "unit": "tokens",
+      "avg": 5552319.0
+    },
+    "total_osl": {
+      "unit": "tokens",
+      "avg": 80540.0
+    },
+    "benchmark_duration": {
+      "unit": "sec",
+      "avg": 137.808622592
+    },
+    "overall_usage_prompt_cache_read_pct": {
+      "unit": "%",
+      "avg": 91.70311885768778
+    },
+    "total_token_throughput": {
+      "unit": "tokens/sec",
+      "avg": 40284.36688102414
+    },
+    "output_token_throughput": {
+      "unit": "tokens/sec",
+      "avg": 584.3509547267879
+    },
+    "input_token_throughput": {
+      "unit": "tokens/sec",
+      "avg": 39700.015926297354
+    },
+    "request_throughput": {
+      "unit": "requests/sec",
+      "avg": 2.9239313975030488
+    },
+    "effective_concurrency": {
+      "unit": "requests",
+      "avg": 32.32931230441113,
+      "p50": 35.0,
+      "p90": 38.0,
+      "p95": 38.0,
+      "p99": 38.0,
+      "min": 0.0,
+      "max": 38.0,
+      "std": 8.939298630077571
+    },
+    "effective_decode_throughput": {
+      "unit": "tokens/sec",
+      "avg": 581.4269884023231,
+      "p50": 192.52718771691286,
+      "p90": 1720.127483652913,
+      "p95": 1872.1077943685716,
+      "p99": 2901.3318849373413,
+      "min": 0.0,
+      "max": 129312.33613437334,
+      "std": 1088.4304867167373
+    },
+    "effective_prefill_throughput": {
+      "unit": "tokens/sec",
+      "avg": 39700.01593609075,
+      "p50": 32976.21260865983,
+      "p90": 93206.24088112456,
+      "p95": 113676.20436510732,
+      "p99": 156918.60951932165,
+      "min": 0.0,
+      "max": 197692.06582276564,
+      "std": 38537.60257028053
+    },
+    "effective_decode_concurrency": {
+      "unit": "requests",
+      "avg": 29.074853031988155,
+      "p50": 33.0,
+      "p90": 36.0,
+      "p95": 36.0,
+      "p99": 38.0,
+      "min": 0.0,
+      "max": 38.0,
+      "std": 9.63849750461768
+    },
+    "effective_prefill_concurrency": {
+      "unit": "requests",
+      "avg": 3.254459272422978,
+      "p50": 1.0,
+      "p90": 6.0,
+      "p95": 10.0,
+      "p99": 38.0,
+      "min": 0.0,
+      "max": 38.0,
+      "std": 6.446772843090381
+    },
+    "effective_total_throughput": {
+      "unit": "tokens/sec",
+      "avg": 40281.44295956362,
+      "p50": 33696.45243458552,
+      "p90": 93586.13650643508,
+      "p95": 114141.12932013458,
+      "p99": 157310.96361764355,
+      "min": 0.0,
+      "max": 198083.7952601637,
+      "std": 38488.16777986818
+    },
+    "effective_decode_throughput_per_user": {
+      "unit": "tokens/sec/user",
+      "avg": 23.263772551139855,
+      "p50": 5.98844570608418,
+      "p90": 56.6928976957194,
+      "p95": 70.50862013974543,
+      "p99": 104.3649220630687,
+      "min": 0.0,
+      "max": 20147.679858500695,
+      "std": 59.682343403746316
+    },
+    "effective_prefill_throughput_per_user": {
+      "unit": "tokens/sec/user",
+      "avg": 14290.24066031416,
+      "p50": 13254.835990061298,
+      "p90": 32194.55520323828,
+      "p95": 39516.27951048514,
+      "p99": 53954.90880273203,
+      "min": 0.0,
+      "max": 63566.197475764064,
+      "std": 13634.453412049459
+    },
+    "tokens_in_flight": {
+      "unit": "tokens",
+      "avg": 470386.5949539142,
+      "p50": 520762.3024448732,
+      "p90": 605187.5152448368,
+      "p95": 627262.1089578733,
+      "p99": 642691.6789110727,
+      "min": 0.0,
+      "max": 648874.7489037912,
+      "std": 142993.972180403
+    },
+    "active_decode_throughput": {
+      "unit": "tokens/sec",
+      "avg": 594.6274141089538,
+      "p50": 193.30304668950333,
+      "p90": 1723.2119682288992,
+      "p95": 1879.1116099310452,
+      "p99": 2923.230090243508,
+      "min": 0.0,
+      "max": 129312.33613437334,
+      "std": 1097.145381823865
+    },
+    "active_prefill_throughput": {
+      "unit": "tokens/sec",
+      "avg": 57094.1593362463,
+      "p50": 49950.30831352553,
+      "p90": 102139.62545021603,
+      "p95": 122241.77923232979,
+      "p99": 160382.60451050114,
+      "min": 1397.4191185869706,
+      "max": 197692.06582276564,
+      "std": 33804.50020101434
+    },
+    "active_decode_throughput_per_user": {
+      "unit": "tokens/sec/user",
+      "avg": 23.791941534250046,
+      "p50": 6.030078522025211,
+      "p90": 57.36409756506997,
+      "p95": 70.60184476619726,
+      "p99": 104.53927597719635,
+      "min": 0.0,
+      "max": 20147.679858500695,
+      "std": 60.25184974533843
+    },
+    "active_prefill_throughput_per_user": {
+      "unit": "tokens/sec/user",
+      "avg": 20551.358934633787,
+      "p50": 18317.54260834961,
+      "p90": 35724.87348499276,
+      "p95": 44579.11136240358,
+      "p99": 57732.92120594103,
+      "min": 1375.8547276893428,
+      "max": 63566.197475764064,
+      "std": 11775.959394245423
+    },
+    "active_total_throughput": {
+      "unit": "tokens/sec",
+      "avg": 40287.14528580103,
+      "p50": 33696.45243458552,
+      "p90": 93586.13650643508,
+      "p95": 114141.12932013458,
+      "p99": 157310.96361764355,
+      "min": 70.60824625344341,
+      "max": 198083.7952601637,
+      "std": 38487.90758270795
+    },
+    "credit_to_start_latency": {
+      "unit": "ms",
+      "avg": 2.1421089032258065,
+      "p1": 0.909056,
+      "p5": 1.0214400000000001,
+      "p10": 1.0721792,
+      "p25": 1.17632,
+      "p50": 1.32096,
+      "p75": 1.4749439999999998,
+      "p90": 2.1597184000000014,
+      "p95": 10.150553599999995,
+      "p99": 14.339092480000001,
+      "min": 0.595456,
+      "max": 18.424832,
+      "std": 2.829957797561303,
+      "count": 403,
+      "sum": 863.269888
+    },
+    "effective_latency": {
+      "unit": "ms",
+      "avg": 11058.937875057072,
+      "p1": 1784.2363136000001,
+      "p5": 3050.022656,
+      "p10": 3863.5173888,
+      "p25": 5183.048448,
+      "p50": 7065.245696,
+      "p75": 11552.617856,
+      "p90": 24105.907097600008,
+      "p95": 35434.86187519999,
+      "p99": 61140.338524160004,
+      "min": 1318.05568,
+      "max": 84165.1904,
+      "std": 11534.838492779776,
+      "count": 403,
+      "sum": 4456751.963648
+    }
+  },
+  "usage_completion_tokens": {
+    "unit": "tokens",
+    "avg": 374.50354223433243,
+    "p1": 38.0,
+    "p5": 53.0,
+    "p10": 61.0,
+    "p25": 83.0,
+    "p50": 144.0,
+    "p75": 455.5,
+    "p90": 1134.6000000000001,
+    "p95": 1363.0999999999997,
+    "p99": 1770.8800000000015,
+    "min": 32.0,
+    "max": 3291.0,
+    "std": 457.14648093179557,
+    "count": 1835,
+    "sum": 687214.0
+  },
+  "http_req_receiving": {
+    "unit": "ms",
+    "avg": 15471.304560356402,
+    "p1": 1061.27513924,
+    "p5": 1712.0696082,
+    "p10": 2268.1674338000003,
+    "p25": 3734.9303459999996,
+    "p50": 6869.465448,
+    "p75": 20033.682321,
+    "p90": 45285.460378400014,
+    "p95": 55078.817743,
+    "p99": 72024.43144278004,
+    "min": 653.526723,
+    "max": 127194.086875,
+    "std": 18045.852238791864,
+    "count": 1835,
+    "sum": 28389843.868254
+  },
+  "usage_prompt_cache_read_tokens": {
+    "unit": "tokens",
+    "avg": 23179.2096332786,
+    "p1": 7744.0,
+    "p5": 8384.0,
+    "p10": 9958.400000000001,
+    "p25": 14144.0,
+    "p50": 22080.0,
+    "p75": 30656.0,
+    "p90": 38784.0,
+    "p95": 42624.0,
+    "p99": 48570.880000000005,
+    "min": 4032.0,
+    "max": 54784.0,
+    "std": 10681.797132434927,
+    "count": 1827,
+    "sum": 42348416.0
+  },
+  "usage_prompt_tokens": {
+    "unit": "tokens",
+    "avg": 23992.62288828338,
+    "p1": 8242.1,
+    "p5": 9208.1,
+    "p10": 10450.4,
+    "p25": 15396.5,
+    "p50": 22683.0,
+    "p75": 31444.5,
+    "p90": 39679.200000000004,
+    "p95": 43533.299999999996,
+    "p99": 49075.840000000004,
+    "min": 8085.0,
+    "max": 56709.0,
+    "std": 10676.691588236537,
+    "count": 1835,
+    "sum": 44026463.0
+  },
+  "http_req_data_sent": {
+    "unit": "KB",
+    "avg": 99.70101115463216,
+    "p1": 37.5925,
+    "p5": 41.147265625,
+    "p10": 45.2943359375,
+    "p25": 64.2431640625,
+    "p50": 93.3359375,
+    "p75": 130.083984375,
+    "p90": 163.91308593750003,
+    "p95": 180.07255859375,
+    "p99": 200.56433593750006,
+    "min": 36.7392578125,
+    "max": 227.02734375,
+    "std": 43.49147523985746,
+    "count": 1835,
+    "sum": 182951.35546875
+  },
+  "http_req_blocked": {
+    "unit": "ms",
+    "avg": 0.0,
+    "p1": 0.0,
+    "p5": 0.0,
+    "p10": 0.0,
+    "p25": 0.0,
+    "p50": 0.0,
+    "p75": 0.0,
+    "p90": 0.0,
+    "p95": 0.0,
+    "p99": 0.0,
+    "min": 0.0,
+    "max": 0.0,
+    "std": 0.0,
+    "count": 1835,
+    "sum": 0.0
+  },
+  "http_req_connection_reused": {
+    "unit": "ratio",
+    "avg": 0.9651226158038148,
+    "p1": 0.0,
+    "p5": 1.0,
+    "p10": 1.0,
+    "p25": 1.0,
+    "p50": 1.0,
+    "p75": 1.0,
+    "p90": 1.0,
+    "p95": 1.0,
+    "p99": 1.0,
+    "min": 0.0,
+    "max": 1.0,
+    "std": 0.18346921340600164,
+    "count": 1835,
+    "sum": 1771.0
+  },
+  "http_req_waiting": {
+    "unit": "ms",
+    "avg": 1374.7925420158037,
+    "p1": 347.22808874,
+    "p5": 375.82679169999994,
+    "p10": 400.4412776,
+    "p25": 487.407654,
+    "p50": 690.1002159999999,
+    "p75": 1084.3914885,
+    "p90": 3534.9972328,
+    "p95": 5004.8364821000005,
+    "p99": 10406.345406360042,
+    "min": 317.288427,
+    "max": 26271.376311,
+    "std": 1993.2878658649572,
+    "count": 1835,
+    "sum": 2522744.314599
+  },
+  "http_req_dns_lookup": {
+    "unit": "ms",
+    "avg": 0.0,
+    "p1": 0.0,
+    "p5": 0.0,
+    "p10": 0.0,
+    "p25": 0.0,
+    "p50": 0.0,
+    "p75": 0.0,
+    "p90": 0.0,
+    "p95": 0.0,
+    "p99": 0.0,
+    "min": 0.0,
+    "max": 0.0,
+    "std": 0.0,
+    "count": 1835,
+    "sum": 0.0
+  },
+  "http_req_connecting": {
+    "unit": "ms",
+    "avg": 0.10386728283378746,
+    "p1": 0.0,
+    "p5": 0.0,
+    "p10": 0.0,
+    "p25": 0.0,
+    "p50": 0.0,
+    "p75": 0.0,
+    "p90": 0.0,
+    "p95": 0.0,
+    "p99": 3.1572060200000047,
+    "min": 0.0,
+    "max": 12.665085,
+    "std": 0.7022971985663381,
+    "count": 1835,
+    "sum": 190.596464
+  },
+  "http_req_sending": {
+    "unit": "ms",
+    "avg": 0.37596579182561307,
+    "p1": 0.18966068,
+    "p5": 0.2001277,
+    "p10": 0.205541,
+    "p25": 0.2204805,
+    "p50": 0.249043,
+    "p75": 0.295497,
+    "p90": 0.3725060000000001,
+    "p95": 0.46944479999999983,
+    "p99": 3.4623150200000192,
+    "min": 0.130104,
+    "max": 13.244622,
+    "std": 0.7382943270339377,
+    "count": 1835,
+    "sum": 689.8972279999999
+  },
+  "http_req_data_received": {
+    "unit": "KB",
+    "avg": 49.13409453763624,
+    "p1": 5.5631640625,
+    "p5": 7.31787109375,
+    "p10": 8.4501953125,
+    "p25": 11.58837890625,
+    "p50": 20.0087890625,
+    "p75": 60.6552734375,
+    "p90": 146.2294921875,
+    "p95": 173.49433593749998,
+    "p99": 224.43308593750012,
+    "min": 4.5576171875,
+    "max": 414.7822265625,
+    "std": 58.32130128366572,
+    "count": 1835,
+    "sum": 90161.0634765625
+  },
+  "usage_total_tokens": {
+    "unit": "tokens",
+    "avg": 24367.12643051771,
+    "p1": 8327.1,
+    "p5": 9272.1,
+    "p10": 10555.0,
+    "p25": 15704.5,
+    "p50": 23002.0,
+    "p75": 32098.0,
+    "p90": 40191.8,
+    "p95": 44021.4,
+    "p99": 49679.18000000003,
+    "min": 8162.0,
+    "max": 57487.0,
+    "std": 10828.786503996556,
+    "count": 1835,
+    "sum": 44713677.0
+  },
+  "http_req_chunks_received": {
+    "unit": "count",
+    "avg": 200.79019073569484,
+    "p1": 20.0,
+    "p5": 28.0,
+    "p10": 31.0,
+    "p25": 44.0,
+    "p50": 78.0,
+    "p75": 247.0,
+    "p90": 610.2000000000003,
+    "p95": 725.0,
+    "p99": 941.6000000000008,
+    "min": 17.0,
+    "max": 1752.0,
+    "std": 244.29047370723507,
+    "count": 1835,
+    "sum": 368450.0
+  },
+  "time_to_first_output_token": {
+    "unit": "ms",
+    "avg": 7720.604664074726,
+    "p1": 398.57805425,
+    "p5": 447.90919974999997,
+    "p10": 535.373559,
+    "p25": 915.2559387499999,
+    "p50": 2891.770917,
+    "p75": 7990.94046525,
+    "p90": 23781.4965285,
+    "p95": 34222.6333115,
+    "p99": 57958.666193599995,
+    "min": 372.688184,
+    "max": 92473.964276,
+    "std": 11993.332040584373,
+    "count": 1646,
+    "sum": 12708115.277067
+  },
+  "http_req_chunks_sent": {
+    "unit": "count",
+    "avg": 1.0,
+    "p1": 1.0,
+    "p5": 1.0,
+    "p10": 1.0,
+    "p25": 1.0,
+    "p50": 1.0,
+    "p75": 1.0,
+    "p90": 1.0,
+    "p95": 1.0,
+    "p99": 1.0,
+    "min": 1.0,
+    "max": 1.0,
+    "std": 0.0,
+    "count": 1835,
+    "sum": 1835.0
+  },
+  "http_req_duration": {
+    "unit": "ms",
+    "avg": 16846.473068164032,
+    "p1": 1621.6934456,
+    "p5": 2375.0862654,
+    "p10": 3027.1878358,
+    "p25": 4787.377823,
+    "p50": 8718.806168,
+    "p75": 21916.8454585,
+    "p90": 46214.290177600014,
+    "p95": 56438.792606699986,
+    "p99": 74913.42193772005,
+    "min": 1092.93442,
+    "max": 127584.552929,
+    "std": 18045.66231838143,
+    "count": 1835,
+    "sum": 30913278.080080997
+  },
+  "e2e_output_token_throughput": {
+    "unit": "tokens/sec/user",
+    "avg": 20.965306631794135,
+    "p1": 5.692909177619335,
+    "p5": 8.627032520101935,
+    "p10": 10.279797181905769,
+    "p25": 17.29518989718783,
+    "p50": 22.402990203819083,
+    "p75": 25.309226746737743,
+    "p90": 27.81337600082561,
+    "p95": 29.73152721514002,
+    "p99": 35.32867430584938,
+    "min": 2.0972494868063536,
+    "max": 51.82333481328521,
+    "std": 6.748980912920451,
+    "count": 1835,
+    "sum": 38471.33766934224
+  },
+  "osl_mismatch_diff_pct": {
+    "unit": "%",
+    "avg": -71.08709183328173,
+    "p1": -96.81133118110694,
+    "p5": -95.51819223606932,
+    "p10": -94.79360248984179,
+    "p25": -92.93429649218325,
+    "p50": -87.87618228718831,
+    "p75": -63.83573352010336,
+    "p90": 0.0,
+    "p95": 0.0,
+    "p99": 0.0,
+    "min": -98.099891422367,
+    "max": 0.0,
+    "std": 32.78149168894645,
+    "count": 1835,
+    "sum": -130444.81351407198
+  },
+  "http_req_connection_overhead": {
+    "unit": "ms",
+    "avg": 0.10386728283378746,
+    "p1": 0.0,
+    "p5": 0.0,
+    "p10": 0.0,
+    "p25": 0.0,
+    "p50": 0.0,
+    "p75": 0.0,
+    "p90": 0.0,
+    "p95": 0.0,
+    "p99": 3.1572060200000047,
+    "min": 0.0,
+    "max": 12.665085,
+    "std": 0.7022971985663381,
+    "count": 1835,
+    "sum": 190.596464
+  },
+  "http_req_total": {
+    "unit": "ms",
+    "avg": 16846.576935446865,
+    "p1": 1622.6414907199999,
+    "p5": 2375.0862654,
+    "p10": 3027.1878358,
+    "p25": 4787.377823,
+    "p50": 8718.806168,
+    "p75": 21916.8454585,
+    "p90": 46214.290177600014,
+    "p95": 56438.792606699986,
+    "p99": 74913.42193772005,
+    "min": 1092.93442,
+    "max": 127584.552929,
+    "std": 18045.636292045245,
+    "count": 1835,
+    "sum": 30913468.676544998
+  },
+  "prefill_throughput_per_user": {
+    "unit": "tokens/sec/user",
+    "avg": 33684.595101985906,
+    "p1": 1608.7117597175622,
+    "p5": 2873.009347459143,
+    "p10": 5293.36461539215,
+    "p25": 14280.721445405572,
+    "p50": 29410.28652745074,
+    "p75": 48753.90286380819,
+    "p90": 69028.45096407167,
+    "p95": 80199.73172265416,
+    "p99": 97707.05466364679,
+    "min": 411.2022978802623,
+    "max": 118483.21062931366,
+    "std": 24022.734566819217,
+    "count": 1835,
+    "sum": 61811232.01214413
+  },
+  "osl_mismatch_count": {
+    "unit": "requests",
+    "avg": 1602.0
+  },
+  "total_usage_completion_tokens": {
+    "unit": "tokens",
+    "avg": 687214.0
+  },
+  "total_usage_prompt_cache_read_tokens": {
+    "unit": "tokens",
+    "avg": 42348416.0
+  },
+  "total_usage_prompt_tokens": {
+    "unit": "tokens",
+    "avg": 44026463.0
+  },
+  "total_usage_total_tokens": {
+    "unit": "tokens",
+    "avg": 44713677.0
+  },
+  "overall_usage_prompt_cache_read_pct": {
+    "unit": "%",
+    "avg": 96.18854914599885
+  },
+  "total_token_throughput": {
+    "unit": "tokens/sec",
+    "avg": 48739.64884227361
+  },
+  "input_token_throughput": {
+    "unit": "tokens/sec",
+    "avg": 47990.558825823064
+  },
+  "effective_concurrency": {
+    "unit": "requests",
+    "avg": 33.696944193375515,
+    "p50": 34.0,
+    "p90": 37.0,
+    "p95": 38.0,
+    "p99": 38.0,
+    "min": 0.0,
+    "max": 38.0,
+    "std": 3.8346715339703064
+  },
+  "effective_decode_throughput": {
+    "unit": "tokens/sec",
+    "avg": 747.0898352327906,
+    "p50": 202.07023743606265,
+    "p90": 1700.2289116867787,
+    "p95": 1787.4772413455107,
+    "p99": 3742.1188160039187,
+    "min": 0.0,
+    "max": 87579.93334707854,
+    "std": 1341.1233363882757
+  },
+  "effective_prefill_throughput": {
+    "unit": "tokens/sec",
+    "avg": 47990.55882592761,
+    "p50": 35842.58026391119,
+    "p90": 109962.49312265782,
+    "p95": 133498.1628039072,
+    "p99": 176015.2793973222,
+    "min": 0.0,
+    "max": 264160.5831714467,
+    "std": 44399.08463151915
+  },
+  "effective_decode_concurrency": {
+    "unit": "requests",
+    "avg": 30.87467134440136,
+    "p50": 31.0,
+    "p90": 34.0,
+    "p95": 35.0,
+    "p99": 36.0,
+    "min": 0.0,
+    "max": 37.0,
+    "std": 4.025370704028643
+  },
+  "effective_prefill_concurrency": {
+    "unit": "requests",
+    "avg": 2.8222728489741553,
+    "p50": 2.0,
+    "p90": 5.0,
+    "p95": 6.0,
+    "p99": 9.0,
+    "min": 0.0,
+    "max": 38.0,
+    "std": 2.7175684042507395
+  },
+  "effective_total_throughput": {
+    "unit": "tokens/sec",
+    "avg": 48737.648620934655,
+    "p50": 36533.94617472267,
+    "p90": 110647.51894181957,
+    "p95": 134301.92007415948,
+    "p99": 176765.5880055357,
+    "min": 0.0,
+    "max": 264815.5980500909,
+    "std": 44389.46754623195
+  },
+  "effective_decode_throughput_per_user": {
+    "unit": "tokens/sec/user",
+    "avg": 24.14289211186672,
+    "p50": 6.227085317554301,
+    "p90": 51.69093326198297,
+    "p95": 53.79521927615226,
+    "p99": 118.68508950118634,
+    "min": 0.0,
+    "max": 18104.823818765737,
+    "std": 44.87908797296593
+  },
+  "effective_prefill_throughput_per_user": {
+    "unit": "tokens/sec/user",
+    "avg": 16173.195846565517,
+    "p50": 12645.783824815828,
+    "p90": 33870.13148687821,
+    "p95": 39547.55269436477,
+    "p99": 53833.321926132776,
+    "min": 0.0,
+    "max": 89196.8767538202,
+    "std": 13287.231246578402
+  },
+  "tokens_in_flight": {
+    "unit": "tokens",
+    "avg": 916811.8692482281,
+    "p50": 948405.3059563945,
+    "p90": 1226890.6988251947,
+    "p95": 1252712.534723765,
+    "p99": 1291951.6368914235,
+    "min": 0.0,
+    "max": 1318476.1635234244,
+    "std": 266043.92684684834
+  },
+  "active_decode_throughput": {
+    "unit": "tokens/sec",
+    "avg": 749.4816921128436,
+    "p50": 202.33565154088282,
+    "p90": 1700.8246064728216,
+    "p95": 1787.8058279696552,
+    "p99": 3763.489487943855,
+    "min": 0.0,
+    "max": 87579.93334707854,
+    "std": 1342.6010291860866
+  },
+  "active_prefill_throughput": {
+    "unit": "tokens/sec",
+    "avg": 50318.088744575536,
+    "p50": 39616.735540726775,
+    "p90": 111744.81173507203,
+    "p95": 135191.17218394423,
+    "p99": 176310.72941736065,
+    "min": 885.9662920206716,
+    "max": 264160.5831714467,
+    "std": 44156.18163241444
+  },
+  "active_decode_throughput_per_user": {
+    "unit": "tokens/sec/user",
+    "avg": 24.220187157092646,
+    "p50": 6.232119975700407,
+    "p90": 51.69730233007223,
+    "p95": 53.81141914818903,
+    "p99": 118.78485903540154,
+    "min": 0.0,
+    "max": 18104.823818765737,
+    "std": 44.93004355251663
+  },
+  "active_prefill_throughput_per_user": {
+    "unit": "tokens/sec/user",
+    "avg": 16957.59173888207,
+    "p50": 13946.925011884186,
+    "p90": 34288.80850274482,
+    "p95": 39759.055702466685,
+    "p99": 55222.21215080472,
+    "min": 885.9662920206716,
+    "max": 89196.8767538202,
+    "std": 13107.696365512245
+  },
+  "active_total_throughput": {
+    "unit": "tokens/sec",
+    "avg": 48738.671611602396,
+    "p50": 36553.61219742007,
+    "p90": 110647.51894181957,
+    "p95": 134301.92007415948,
+    "p99": 176765.5880055357,
+    "min": 49.82208058964218,
+    "max": 264815.5980500909,
+    "std": 44389.37179709186
+  },
+  "credit_to_start_latency": {
+    "unit": "ms",
+    "avg": 1.7362522681198909,
+    "p1": 0.8740096,
+    "p5": 1.0265344,
+    "p10": 1.0968064,
+    "p25": 1.211904,
+    "p50": 1.349376,
+    "p75": 1.546624,
+    "p90": 1.9005440000000002,
+    "p95": 2.6452479999999987,
+    "p99": 15.33016576000003,
+    "min": 0.558848,
+    "max": 22.646784,
+    "std": 2.119021187487743,
+    "count": 1835,
+    "sum": 3186.022912
+  },
+  "effective_latency": {
+    "unit": "ms",
+    "avg": 16848.343049486648,
+    "p1": 1623.23836928,
+    "p5": 2376.3842304,
+    "p10": 3029.6739328,
+    "p25": 4788.525312,
+    "p50": 8719.873792,
+    "p75": 21918.475008,
+    "p90": 46215.88377600002,
+    "p95": 56440.576844799994,
+    "p99": 74915.12632320005,
+    "min": 1094.77632,
+    "max": 127586.082816,
+    "std": 18045.58656921296,
+    "count": 1835,
+    "sum": 30916709.495807998
+  }
+}