{
  "optimization_id": "opt-e4fdfbd4b187",
  "run_id": "run-1488b44fa409",
  "target": "balanced",
  "timestamp": "2026-08-14T03:44:05.169036Z",
  "workflow": "research-ops",
  "minimum_quality": 0.8,
  "status": "DEPLOYABLE",
  "selections": [
    {
      "stage_id": "classify",
      "target": "balanced",
      "selected": {
        "run_id": "run-1488b44fa409",
        "timestamp": "2026-08-12T22:22:52.341896Z",
        "git_commit": "b414b9e8d52ea72dc909b92d87ba8a0efbd9c2d4",
        "hostname": "runnervma9114",
        "architecture": "aarch64",
        "cpu": "aarch64",
        "cores": 4,
        "ram_gb": 15.57,
        "stage_id": "classify",
        "model": "large",
        "model_repo": "ggml-org/Qwen3-4B-GGUF",
        "model_hash": "ab27b9bfa375a178d6cba48f3ad892b94b7739659dcc7aae8058ce0ffed6b328",
        "quantization": "Q4_K_M",
        "threads": 1,
        "batch_size": 64,
        "context_size": 512,
        "prompt_tokens": 334,
        "generated_tokens": 36,
        "latency_ms": 6510.327356500037,
        "median_latency_ms": 6404.471022500047,
        "p95_latency_ms": 6745.650973250008,
        "ttft_ms": null,
        "prompt_tokens_per_second": 10.67890021994837,
        "generation_tokens_per_second": 4.6447409965507305,
        "total_tokens_per_second": 9.47212993907249,
        "requests_per_minute": 9.216126427205667,
        "peak_memory_mb": 4885.046875,
        "average_cpu_percent": 99.8595882990249,
        "peak_cpu_percent": 132.9,
        "model_size_bytes": 2497280640,
        "load_time_ms": 3315.4892599999357,
        "quality_score": 1.0,
        "measured_runs": 6,
        "warmup_runs": 1,
        "run_label": "VERIFIED ARM64 RUN",
        "verified_arm64": true,
        "dataset_hash": "a2beee5f8f7f65cd35792936cc247f241551d0e9c1155d23a9ca863f042f7736",
        "error": null,
        "raw_artifact": "\\home\\runner\\work\\a64forge\\a64forge\\benchmarks\\run-1488b44fa409\\classify\\large_Q4_K_M_t1_b64_c512.json",
        "performix_artifact": null,
        "metadata": {
          "command": [
            "/home/runner/work/a64forge/a64forge/.a64forge/llama.cpp/build/bin/llama-server",
            "-m",
            "/home/runner/work/a64forge/a64forge/models/Qwen3-4B-Q4_K_M.gguf",
            "--host",
            "127.0.0.1",
            "--port",
            "48835",
            "--threads",
            "1",
            "--threads-batch",
            "1",
            "--batch-size",
            "64",
            "--ctx-size",
            "512",
            "--metrics"
          ],
          "server_log": "/home/runner/work/a64forge/a64forge/benchmarks/run-1488b44fa409/classify/large_Q4_K_M_t1_b64_c512.server.log",
          "server_log_tail": "d  0 | task -1 | saving idle slot to prompt cache\n0.29.252.631 I slot prompt_clear: id  0 | task -1 | clearing prompt with 0 tokens\n0.29.252.634 I slot process_sing: id  1 | task -1 | saving idle slot to prompt cache\n0.29.252.634 I slot prompt_clear: id  1 | task -1 | clearing prompt with 0 tokens\n0.29.252.635 I slot process_sing: id  2 | task -1 | saving idle slot to prompt cache\n0.29.252.635 I slot prompt_clear: id  2 | task -1 | clearing prompt with 0 tokens\n0.35.653.756 I slot print_timing: id  3 | task 32 | prompt eval time =    5095.98 ms /    53 tokens (   96.15 ms per token,    10.40 tokens per second)\n0.35.653.774 I slot print_timing: id  3 | task 32 |        eval time =    1305.10 ms /     6 tokens (  217.52 ms per token,     4.60 tokens per second)\n0.35.653.774 I slot print_timing: id  3 | task 32 |       total time =    6401.08 ms /    59 tokens\n0.35.653.775 I slot print_timing: id  3 | task 32 |    graphs reused =         21\n0.35.653.810 I slot      release: id  3 | task 32 | stop processing: n_tokens = 58, truncated = 0\n0.35.653.824 I srv  update_slots: all slots are idle\n0.35.661.053 I srv  params_from_: Chat format: peg-native\n0.35.661.229 I slot get_availabl: id  3 | task -1 | selected slot by LCP similarity, sim_best = 0.576 (> 0.100 thold), f_keep = 0.586\n0.35.661.283 I slot launch_slot_: id  3 | task 40 | processing task, is_child = 0\n0.35.661.288 I slot process_sing: id  0 | task -1 | saving idle slot to prompt cache\n0.35.661.289 I slot prompt_clear: id  0 | task -1 | clearing prompt with 0 tokens\n0.35.661.292 I slot process_sing: id  1 | task -1 | saving idle slot to prompt cache\n0.35.661.292 I slot prompt_clear: id  1 | task -1 | clearing prompt with 0 tokens\n0.35.661.293 I slot process_sing: id  2 | task -1 | saving idle slot to prompt cache\n0.35.661.293 I slot prompt_clear: id  2 | task -1 | clearing prompt with 0 tokens\n0.42.385.680 I slot print_timing: id  3 | task 40 | prompt eval time =    5429.93 ms /    59 tokens (   92.03 ms per token,    10.87 tokens per second)\n0.42.385.702 I slot print_timing: id  3 | task 40 |        eval time =    1294.43 ms /     6 tokens (  215.74 ms per token,     4.64 tokens per second)\n0.42.385.703 I slot print_timing: id  3 | task 40 |       total time =    6724.36 ms /    65 tokens\n0.42.385.704 I slot print_timing: id  3 | task 40 |    graphs reused =         25\n0.42.385.751 I slot      release: id  3 | task 40 | stop processing: n_tokens = 64, truncated = 0\n0.42.385.765 I srv  update_slots: all slots are idle\n0.42.392.931 I srv  params_from_: Chat format: peg-native\n0.42.393.113 I slot get_availabl: id  3 | task -1 | selected slot by LCP similarity, sim_best = 0.618 (> 0.100 thold), f_keep = 0.531\n0.42.393.168 I slot launch_slot_: id  3 | task 48 | processing task, is_child = 0\n0.42.393.173 I slot process_sing: id  0 | task -1 | saving idle slot to prompt cache\n0.42.393.174 I slot prompt_clear: id  0 | task -1 | clearing prompt with 0 tokens\n0.42.393.177 I slot process_sing: id  1 | task -1 | saving idle slot to prompt cache\n0.42.393.178 I slot prompt_clear: id  1 | task -1 | clearing prompt with 0 tokens\n0.42.393.178 I slot process_sing: id  2 | task -1 | saving idle slot to prompt cache\n0.42.393.179 I slot prompt_clear: id  2 | task -1 | clearing prompt with 0 tokens\n0.48.775.891 I slot print_timing: id  3 | task 48 | prompt eval time =    5096.28 ms /    55 tokens (   92.66 ms per token,    10.79 tokens per second)\n0.48.775.911 I slot print_timing: id  3 | task 48 |        eval time =    1286.40 ms /     6 tokens (  214.40 ms per token,     4.66 tokens per second)\n0.48.775.911 I slot print_timing: id  3 | task 48 |       total time =    6382.69 ms /    61 tokens\n0.48.775.913 I slot print_timing: id  3 | task 48 |    graphs reused =         29\n0.48.775.946 I slot      release: id  3 | task 48 | stop processing: n_tokens = 60, truncated = 0\n0.48.775.961 I srv  update_slots: all slots are idle\n0.48.785.659 I srv    operator(): operator(): cleaning up before exit...\n",
          "baseline": true
        }
      },
      "pareto_keys": [
        "large:Q4_K_M:t1:b64:c512"
      ],
      "explanation": [
        "Meets quality floor 0.950 with 1.000.",
        "Selected from 1 Pareto-efficient candidates using the balanced preset.",
        "Median latency: 0.0% lower than baseline.",
        "Peak memory: 0.0% lower than baseline."
      ],
      "score": 1.0
    },
    {
      "stage_id": "extract",
      "target": "balanced",
      "selected": {
        "run_id": "run-1488b44fa409",
        "timestamp": "2026-08-12T22:24:01.542756Z",
        "git_commit": "b414b9e8d52ea72dc909b92d87ba8a0efbd9c2d4",
        "hostname": "runnervma9114",
        "architecture": "aarch64",
        "cpu": "aarch64",
        "cores": 4,
        "ram_gb": 15.57,
        "stage_id": "extract",
        "model": "medium",
        "model_repo": "ggml-org/Qwen3-1.7B-GGUF",
        "model_hash": "d2387ca2dbfee2ffabce7120d3770dadca0b293052bc2f0e138fdc940d9bc7b5",
        "quantization": "Q4_K_M",
        "threads": 4,
        "batch_size": 128,
        "context_size": 1024,
        "prompt_tokens": 228,
        "generated_tokens": 158,
        "latency_ms": 2122.910189749973,
        "median_latency_ms": 2124.376566999956,
        "p95_latency_ms": 2197.414479599905,
        "ttft_ms": null,
        "prompt_tokens_per_second": 101.5256225420951,
        "generation_tokens_per_second": 25.46229341967435,
        "total_tokens_per_second": 45.456468420534236,
        "requests_per_minute": 28.263089173389165,
        "peak_memory_mb": 2311.41796875,
        "average_cpu_percent": 339.6286956521739,
        "peak_cpu_percent": 418.8,
        "model_size_bytes": 1282439264,
        "load_time_ms": 1541.3667899999837,
        "quality_score": 0.8333333333333333,
        "measured_runs": 4,
        "warmup_runs": 1,
        "run_label": "VERIFIED ARM64 RUN",
        "verified_arm64": true,
        "dataset_hash": "737bfd42a8b1a6f18928b8f6db197d00c63cdf9e0359a6c2656f605b49d1244f",
        "error": null,
        "raw_artifact": "\\home\\runner\\work\\a64forge\\a64forge\\benchmarks\\run-1488b44fa409\\extract\\medium_Q4_K_M_t4_b128_c1024.json",
        "performix_artifact": null,
        "metadata": {
          "command": [
            "/home/runner/work/a64forge/a64forge/.a64forge/llama.cpp/build/bin/llama-server",
            "-m",
            "/home/runner/work/a64forge/a64forge/models/Qwen3-1.7B-Q4_K_M.gguf",
            "--host",
            "127.0.0.1",
            "--port",
            "33939",
            "--threads",
            "4",
            "--threads-batch",
            "4",
            "--batch-size",
            "128",
            "--ctx-size",
            "1024",
            "--metrics"
          ],
          "server_log": "/home/runner/work/a64forge/a64forge/benchmarks/run-1488b44fa409/extract/medium_Q4_K_M_t4_b128_c1024.server.log",
          "server_log_tail": "vailabl: updating prompt cache\n0.08.006.869 W srv   prompt_save:  - saving prompt with length 92, total state size = 10.064 MiB (draft: 0.000 MiB)\n0.08.009.801 I srv          load:  - looking for better prompt, base f_keep = 0.337, sim = 0.525\n0.08.009.816 I srv          load:  - found better prompt with f_keep = 0.596, sim = 1.000\n0.08.010.992 I srv        update:  - cache state: 1 prompts, 10.064 MiB (limits: 8192.000 MiB, 1024 tokens, 74885 est)\n0.08.011.004 I srv        update:    - prompt 0xaacf962570a0:      92 tokens, checkpoints:  0,    10.064 MiB\n0.08.011.005 I srv  get_availabl: prompt cache update took 4.15 ms\n0.08.011.055 I slot launch_slot_: id  3 | task 126 | processing task, is_child = 0\n0.08.011.060 I slot process_sing: id  0 | task -1 | saving idle slot to prompt cache\n0.08.011.061 I slot prompt_clear: id  0 | task -1 | clearing prompt with 0 tokens\n0.08.011.064 I slot process_sing: id  1 | task -1 | saving idle slot to prompt cache\n0.08.011.064 I slot prompt_clear: id  1 | task -1 | clearing prompt with 0 tokens\n0.08.011.065 I slot process_sing: id  2 | task -1 | saving idle slot to prompt cache\n0.08.011.066 I slot prompt_clear: id  2 | task -1 | clearing prompt with 0 tokens\n0.10.193.689 I slot print_timing: id  3 | task 126 | prompt eval time =     582.35 ms /    59 tokens (    9.87 ms per token,   101.31 tokens per second)\n0.10.193.704 I slot print_timing: id  3 | task 126 |        eval time =    1600.25 ms /    41 tokens (   39.03 ms per token,    25.62 tokens per second)\n0.10.193.704 I slot print_timing: id  3 | task 126 |       total time =    2182.60 ms /   100 tokens\n0.10.193.705 I slot print_timing: id  3 | task 126 |    graphs reused =        154\n0.10.193.736 I slot      release: id  3 | task 126 | stop processing: n_tokens = 99, truncated = 0\n0.10.193.751 I srv  update_slots: all slots are idle\n0.10.200.958 I srv  params_from_: Chat format: peg-native\n0.10.201.125 I slot get_availabl: id  3 | task -1 | selected slot by LCP similarity, sim_best = 0.564 (> 0.100 thold), f_keep = 0.313\n0.10.201.128 I srv  get_availabl: updating prompt cache\n0.10.201.145 W srv   prompt_save:  - saving prompt with length 99, total state size = 10.830 MiB (draft: 0.000 MiB)\n0.10.204.178 I srv          load:  - looking for better prompt, base f_keep = 0.313, sim = 0.564\n0.10.204.194 I srv          load:  - found better prompt with f_keep = 0.598, sim = 1.000\n0.10.205.352 I srv        update:  - cache state: 1 prompts, 10.830 MiB (limits: 8192.000 MiB, 1024 tokens, 74885 est)\n0.10.205.386 I srv        update:    - prompt 0xaacf96256320:      99 tokens, checkpoints:  0,    10.830 MiB\n0.10.205.388 I srv  get_availabl: prompt cache update took 4.26 ms\n0.10.205.442 I slot launch_slot_: id  3 | task 169 | processing task, is_child = 0\n0.10.205.447 I slot process_sing: id  0 | task -1 | saving idle slot to prompt cache\n0.10.205.448 I slot prompt_clear: id  0 | task -1 | clearing prompt with 0 tokens\n0.10.205.451 I slot process_sing: id  1 | task -1 | saving idle slot to prompt cache\n0.10.205.451 I slot prompt_clear: id  1 | task -1 | clearing prompt with 0 tokens\n0.10.205.452 I slot process_sing: id  2 | task -1 | saving idle slot to prompt cache\n0.10.205.452 I slot prompt_clear: id  2 | task -1 | clearing prompt with 0 tokens\n0.12.248.124 I slot print_timing: id  3 | task 169 | prompt eval time =     547.58 ms /    55 tokens (    9.96 ms per token,   100.44 tokens per second)\n0.12.248.138 I slot print_timing: id  3 | task 169 |        eval time =    1495.06 ms /    38 tokens (   39.34 ms per token,    25.42 tokens per second)\n0.12.248.138 I slot print_timing: id  3 | task 169 |       total time =    2042.64 ms /    93 tokens\n0.12.248.139 I slot print_timing: id  3 | task 169 |    graphs reused =        190\n0.12.248.387 I slot      release: id  3 | task 169 | stop processing: n_tokens = 92, truncated = 0\n0.12.248.402 I srv  update_slots: all slots are idle\n0.12.279.153 I srv    operator(): operator(): cleaning up before exit...\n",
          "baseline": false
        }
      },
      "pareto_keys": [
        "medium:Q4_K_M:t2:b128:c1024",
        "medium:Q4_K_M:t4:b128:c1024"
      ],
      "explanation": [
        "Meets quality floor 0.800 with 0.833.",
        "Selected from 2 Pareto-efficient candidates using the balanced preset.",
        "Median latency: 86.1% lower than baseline.",
        "Peak memory: 52.8% lower than baseline."
      ],
      "score": 0.8
    },
    {
      "stage_id": "tool_select",
      "target": "balanced",
      "selected": {
        "run_id": "run-1488b44fa409",
        "timestamp": "2026-08-12T22:28:15.263131Z",
        "git_commit": "b414b9e8d52ea72dc909b92d87ba8a0efbd9c2d4",
        "hostname": "runnervma9114",
        "architecture": "aarch64",
        "cpu": "aarch64",
        "cores": 4,
        "ram_gb": 15.57,
        "stage_id": "tool_select",
        "model": "large",
        "model_repo": "ggml-org/Qwen3-4B-GGUF",
        "model_hash": "ab27b9bfa375a178d6cba48f3ad892b94b7739659dcc7aae8058ce0ffed6b328",
        "quantization": "Q4_K_M",
        "threads": 1,
        "batch_size": 128,
        "context_size": 512,
        "prompt_tokens": 428,
        "generated_tokens": 50,
        "latency_ms": 6445.144349500026,
        "median_latency_ms": 6338.98487850007,
        "p95_latency_ms": 7075.522790650001,
        "ttft_ms": null,
        "prompt_tokens_per_second": 10.467867874825258,
        "generation_tokens_per_second": 4.715612250936199,
        "total_tokens_per_second": 9.270544887739408,
        "requests_per_minute": 9.309333778483088,
        "peak_memory_mb": 4885.359375,
        "average_cpu_percent": 99.74147386461011,
        "peak_cpu_percent": 115.0,
        "model_size_bytes": 2497280640,
        "load_time_ms": 3315.810873000032,
        "quality_score": 1.0,
        "measured_runs": 8,
        "warmup_runs": 1,
        "run_label": "VERIFIED ARM64 RUN",
        "verified_arm64": true,
        "dataset_hash": "5e97a2f29f08e1cc6be3cf033845890eb3eebdae6b58ec7f76c2dd04d22f1e72",
        "error": null,
        "raw_artifact": "\\home\\runner\\work\\a64forge\\a64forge\\benchmarks\\run-1488b44fa409\\tool_select\\large_Q4_K_M_t1_b128_c512.json",
        "performix_artifact": null,
        "metadata": {
          "command": [
            "/home/runner/work/a64forge/a64forge/.a64forge/llama.cpp/build/bin/llama-server",
            "-m",
            "/home/runner/work/a64forge/a64forge/models/Qwen3-4B-Q4_K_M.gguf",
            "--host",
            "127.0.0.1",
            "--port",
            "54357",
            "--threads",
            "1",
            "--threads-batch",
            "1",
            "--batch-size",
            "128",
            "--ctx-size",
            "512",
            "--metrics"
          ],
          "server_log": "/home/runner/work/a64forge/a64forge/benchmarks/run-1488b44fa409/tool_select/large_Q4_K_M_t1_b128_c512.server.log",
          "server_log_tail": "d  0 | task -1 | saving idle slot to prompt cache\n0.42.012.209 I slot prompt_clear: id  0 | task -1 | clearing prompt with 0 tokens\n0.42.012.213 I slot process_sing: id  1 | task -1 | saving idle slot to prompt cache\n0.42.012.213 I slot prompt_clear: id  1 | task -1 | clearing prompt with 0 tokens\n0.42.012.214 I slot process_sing: id  2 | task -1 | saving idle slot to prompt cache\n0.42.012.214 I slot prompt_clear: id  2 | task -1 | clearing prompt with 0 tokens\n0.49.080.512 I slot print_timing: id  3 | task 51 | prompt eval time =    5802.09 ms /    61 tokens (   95.12 ms per token,    10.51 tokens per second)\n0.49.080.526 I slot print_timing: id  3 | task 51 |        eval time =    1266.18 ms /     6 tokens (  211.03 ms per token,     4.74 tokens per second)\n0.49.080.527 I slot print_timing: id  3 | task 51 |       total time =    7068.27 ms /    67 tokens\n0.49.080.528 I slot print_timing: id  3 | task 51 |    graphs reused =         32\n0.49.080.560 I slot      release: id  3 | task 51 | stop processing: n_tokens = 66, truncated = 0\n0.49.080.574 I srv  update_slots: all slots are idle\n0.49.087.700 I srv  params_from_: Chat format: peg-native\n0.49.087.882 I slot get_availabl: id  3 | task -1 | selected slot by LCP similarity, sim_best = 0.680 (> 0.100 thold), f_keep = 0.515\n0.49.087.932 I slot launch_slot_: id  3 | task 59 | processing task, is_child = 0\n0.49.087.936 I slot process_sing: id  0 | task -1 | saving idle slot to prompt cache\n0.49.087.937 I slot prompt_clear: id  0 | task -1 | clearing prompt with 0 tokens\n0.49.087.941 I slot process_sing: id  1 | task -1 | saving idle slot to prompt cache\n0.49.087.942 I slot prompt_clear: id  1 | task -1 | clearing prompt with 0 tokens\n0.49.087.943 I slot process_sing: id  2 | task -1 | saving idle slot to prompt cache\n0.49.087.944 I slot prompt_clear: id  2 | task -1 | clearing prompt with 0 tokens\n0.55.301.562 I slot print_timing: id  3 | task 59 | prompt eval time =    4950.83 ms /    50 tokens (   99.02 ms per token,    10.10 tokens per second)\n0.55.301.578 I slot print_timing: id  3 | task 59 |        eval time =    1262.77 ms /     6 tokens (  210.46 ms per token,     4.75 tokens per second)\n0.55.301.579 I slot print_timing: id  3 | task 59 |       total time =    6213.59 ms /    56 tokens\n0.55.301.580 I slot print_timing: id  3 | task 59 |    graphs reused =         36\n0.55.301.614 I slot      release: id  3 | task 59 | stop processing: n_tokens = 55, truncated = 0\n0.55.301.628 I srv  update_slots: all slots are idle\n0.55.308.847 I srv  params_from_: Chat format: peg-native\n0.55.309.027 I slot get_availabl: id  3 | task -1 | selected slot by LCP similarity, sim_best = 0.667 (> 0.100 thold), f_keep = 0.618\n0.55.309.079 I slot launch_slot_: id  3 | task 67 | processing task, is_child = 0\n0.55.309.084 I slot process_sing: id  0 | task -1 | saving idle slot to prompt cache\n0.55.309.085 I slot prompt_clear: id  0 | task -1 | clearing prompt with 0 tokens\n0.55.309.089 I slot process_sing: id  1 | task -1 | saving idle slot to prompt cache\n0.55.309.089 I slot prompt_clear: id  1 | task -1 | clearing prompt with 0 tokens\n0.55.309.090 I slot process_sing: id  2 | task -1 | saving idle slot to prompt cache\n0.55.309.090 I slot prompt_clear: id  2 | task -1 | clearing prompt with 0 tokens\n1.01.331.997 I slot print_timing: id  3 | task 67 | prompt eval time =    4763.09 ms /    51 tokens (   93.39 ms per token,    10.71 tokens per second)\n1.01.332.015 I slot print_timing: id  3 | task 67 |        eval time =    1259.79 ms /     6 tokens (  209.97 ms per token,     4.76 tokens per second)\n1.01.332.015 I slot print_timing: id  3 | task 67 |       total time =    6022.88 ms /    57 tokens\n1.01.332.016 I slot print_timing: id  3 | task 67 |    graphs reused =         40\n1.01.332.049 I slot      release: id  3 | task 67 | stop processing: n_tokens = 56, truncated = 0\n1.01.332.063 I srv  update_slots: all slots are idle\n1.01.335.560 I srv    operator(): operator(): cleaning up before exit...\n",
          "baseline": false
        }
      },
      "pareto_keys": [
        "large:Q4_K_M:t1:b128:c512",
        "large:Q4_K_M:t1:b64:c512"
      ],
      "explanation": [
        "Meets quality floor 0.950 with 1.000.",
        "Selected from 2 Pareto-efficient candidates using the balanced preset.",
        "Median latency: 0.5% lower than baseline.",
        "Peak memory: 0.0% higher than baseline."
      ],
      "score": 0.8
    },
    {
      "stage_id": "reason",
      "target": "balanced",
      "selected": {
        "run_id": "run-1488b44fa409",
        "timestamp": "2026-08-12T22:30:32.216992Z",
        "git_commit": "b414b9e8d52ea72dc909b92d87ba8a0efbd9c2d4",
        "hostname": "runnervma9114",
        "architecture": "aarch64",
        "cpu": "aarch64",
        "cores": 4,
        "ram_gb": 15.57,
        "stage_id": "reason",
        "model": "small",
        "model_repo": "ggml-org/Qwen3-0.6B-GGUF",
        "model_hash": "da2572f16c06133561ce56accaa822216f2391ef4d37fba427801cd6736417d4",
        "quantization": "Q4_0",
        "threads": 1,
        "batch_size": 64,
        "context_size": 2048,
        "prompt_tokens": 334,
        "generated_tokens": 296,
        "latency_ms": 3245.383856249987,
        "median_latency_ms": 3244.1491644999587,
        "p95_latency_ms": 4901.045476050046,
        "ttft_ms": null,
        "prompt_tokens_per_second": 112.04492137207272,
        "generation_tokens_per_second": 31.27637630610107,
        "total_tokens_per_second": 48.53046880623542,
        "requests_per_minute": 18.487797640470635,
        "peak_memory_mb": 955.16015625,
        "average_cpu_percent": 98.44805555555557,
        "peak_cpu_percent": 133.5,
        "model_size_bytes": 428970080,
        "load_time_ms": 1035.544967000078,
        "quality_score": 0.8333333333333333,
        "measured_runs": 4,
        "warmup_runs": 1,
        "run_label": "VERIFIED ARM64 RUN",
        "verified_arm64": true,
        "dataset_hash": "74c11a25259eb389353c38803fd559026426a07a30d5523f65d3ee707dc12e46",
        "error": null,
        "raw_artifact": "\\home\\runner\\work\\a64forge\\a64forge\\benchmarks\\run-1488b44fa409\\reason\\small_Q4_0_t1_b64_c2048.json",
        "performix_artifact": null,
        "metadata": {
          "command": [
            "/home/runner/work/a64forge/a64forge/.a64forge/llama.cpp/build/bin/llama-server",
            "-m",
            "/home/runner/work/a64forge/a64forge/models/Qwen3-0.6B-Q4_0.gguf",
            "--host",
            "127.0.0.1",
            "--port",
            "43227",
            "--threads",
            "1",
            "--threads-batch",
            "1",
            "--batch-size",
            "64",
            "--ctx-size",
            "2048",
            "--metrics"
          ],
          "server_log": "/home/runner/work/a64forge/a64forge/benchmarks/run-1488b44fa409/reason/small_Q4_0_t1_b64_c2048.server.log",
          "server_log_tail": " 103, total state size = 11.267 MiB (draft: 0.000 MiB)\n0.12.416.300 I srv          load:  - looking for better prompt, base f_keep = 0.320, sim = 0.337\n0.12.416.313 I srv          load:  - found better prompt with f_keep = 0.467, sim = 1.000\n0.12.418.741 I srv        update:  - cache state: 1 prompts, 11.267 MiB (limits: 8192.000 MiB, 2048 tokens, 74886 est)\n0.12.418.747 I srv        update:    - prompt 0xab1d11438340:     103 tokens, checkpoints:  0,    11.267 MiB\n0.12.418.748 I srv  get_availabl: prompt cache update took 5.66 ms\n0.12.418.803 I slot launch_slot_: id  3 | task 269 | processing task, is_child = 0\n0.12.418.808 I slot process_sing: id  0 | task -1 | saving idle slot to prompt cache\n0.12.418.808 I slot prompt_clear: id  0 | task -1 | clearing prompt with 0 tokens\n0.12.418.812 I slot process_sing: id  1 | task -1 | saving idle slot to prompt cache\n0.12.418.812 I slot prompt_clear: id  1 | task -1 | clearing prompt with 0 tokens\n0.12.418.814 I slot process_sing: id  2 | task -1 | saving idle slot to prompt cache\n0.12.418.815 I slot prompt_clear: id  2 | task -1 | clearing prompt with 0 tokens\n0.16.854.965 I slot print_timing: id  3 | task 269 | n_decoded =    100, tg =  30.21 t/s\n0.17.307.570 I slot print_timing: id  3 | task 269 | prompt eval time =    1126.35 ms /    98 tokens (   11.49 ms per token,    87.01 tokens per second)\n0.17.307.581 I slot print_timing: id  3 | task 269 |        eval time =    3762.37 ms /   113 tokens (   33.30 ms per token,    30.03 tokens per second)\n0.17.307.582 I slot print_timing: id  3 | task 269 |       total time =    4888.72 ms /   211 tokens\n0.17.307.583 I slot print_timing: id  3 | task 269 |    graphs reused =        367\n0.17.307.616 I slot      release: id  3 | task 269 | stop processing: n_tokens = 210, truncated = 0\n0.17.307.629 I srv  update_slots: all slots are idle\n0.17.314.802 I srv  params_from_: Chat format: peg-native\n0.17.314.988 I slot get_availabl: id  3 | task -1 | selected slot by LCP similarity, sim_best = 0.478 (> 0.100 thold), f_keep = 0.157\n0.17.314.993 I srv  get_availabl: updating prompt cache\n0.17.315.013 W srv   prompt_save:  - saving prompt with length 210, total state size = 22.972 MiB (draft: 0.000 MiB)\n0.17.320.922 I srv          load:  - looking for better prompt, base f_keep = 0.157, sim = 0.478\n0.17.320.938 I srv          load:  - found better prompt with f_keep = 0.670, sim = 1.000\n0.17.322.140 I srv        update:  - cache state: 1 prompts, 22.972 MiB (limits: 8192.000 MiB, 2048 tokens, 74888 est)\n0.17.322.153 I srv        update:    - prompt 0xab1d1143b9a0:     210 tokens, checkpoints:  0,    22.972 MiB\n0.17.322.154 I srv  get_availabl: prompt cache update took 7.16 ms\n0.17.322.209 I slot launch_slot_: id  3 | task 385 | processing task, is_child = 0\n0.17.322.213 I slot process_sing: id  0 | task -1 | saving idle slot to prompt cache\n0.17.322.214 I slot prompt_clear: id  0 | task -1 | clearing prompt with 0 tokens\n0.17.322.217 I slot process_sing: id  1 | task -1 | saving idle slot to prompt cache\n0.17.322.218 I slot prompt_clear: id  1 | task -1 | clearing prompt with 0 tokens\n0.17.322.220 I slot process_sing: id  2 | task -1 | saving idle slot to prompt cache\n0.17.322.220 I slot prompt_clear: id  2 | task -1 | clearing prompt with 0 tokens\n0.18.899.433 I slot print_timing: id  3 | task 385 | prompt eval time =     503.71 ms /    69 tokens (    7.30 ms per token,   136.98 tokens per second)\n0.18.899.450 I slot print_timing: id  3 | task 385 |        eval time =    1073.48 ms /    35 tokens (   30.67 ms per token,    32.60 tokens per second)\n0.18.899.450 I slot print_timing: id  3 | task 385 |       total time =    1577.19 ms /   104 tokens\n0.18.899.451 I slot print_timing: id  3 | task 385 |    graphs reused =        400\n0.18.899.485 I slot      release: id  3 | task 385 | stop processing: n_tokens = 103, truncated = 0\n0.18.899.499 I srv  update_slots: all slots are idle\n0.18.922.092 I srv    operator(): operator(): cleaning up before exit...\n",
          "baseline": false
        }
      },
      "pareto_keys": [
        "medium:Q4_K_M:t2:b128:c1024",
        "medium:Q4_K_M:t4:b128:c1024",
        "small:Q4_0:t1:b64:c2048"
      ],
      "explanation": [
        "Meets quality floor 0.800 with 0.833.",
        "Selected from 3 Pareto-efficient candidates using the balanced preset.",
        "Median latency: 82.0% lower than baseline.",
        "Peak memory: 80.5% lower than baseline."
      ],
      "score": 0.75
    },
    {
      "stage_id": "summarize",
      "target": "balanced",
      "selected": {
        "run_id": "run-1488b44fa409",
        "timestamp": "2026-08-12T22:32:58.947825Z",
        "git_commit": "b414b9e8d52ea72dc909b92d87ba8a0efbd9c2d4",
        "hostname": "runnervma9114",
        "architecture": "aarch64",
        "cpu": "aarch64",
        "cores": 4,
        "ram_gb": 15.57,
        "stage_id": "summarize",
        "model": "small",
        "model_repo": "ggml-org/Qwen3-0.6B-GGUF",
        "model_hash": "da2572f16c06133561ce56accaa822216f2391ef4d37fba427801cd6736417d4",
        "quantization": "Q4_0",
        "threads": 1,
        "batch_size": 128,
        "context_size": 2048,
        "prompt_tokens": 260,
        "generated_tokens": 144,
        "latency_ms": 1626.5348535000044,
        "median_latency_ms": 1626.4850359999627,
        "p95_latency_ms": 1627.8431223500888,
        "ttft_ms": null,
        "prompt_tokens_per_second": 127.04278306989058,
        "generation_tokens_per_second": 32.59361941571392,
        "total_tokens_per_second": 62.09519567482157,
        "requests_per_minute": 36.888235054349444,
        "peak_memory_mb": 943.93359375,
        "average_cpu_percent": 99.40470588235294,
        "peak_cpu_percent": 114.7,
        "model_size_bytes": 428970080,
        "load_time_ms": 783.1988940001793,
        "quality_score": 1.0,
        "measured_runs": 4,
        "warmup_runs": 1,
        "run_label": "VERIFIED ARM64 RUN",
        "verified_arm64": true,
        "dataset_hash": "f524c6a0b252a3d0e01a44701914354f6130c685e7d862f2bfdd4d9d5686efdd",
        "error": null,
        "raw_artifact": "\\home\\runner\\work\\a64forge\\a64forge\\benchmarks\\run-1488b44fa409\\summarize\\small_Q4_0_t1_b128_c2048.json",
        "performix_artifact": null,
        "metadata": {
          "command": [
            "/home/runner/work/a64forge/a64forge/.a64forge/llama.cpp/build/bin/llama-server",
            "-m",
            "/home/runner/work/a64forge/a64forge/models/Qwen3-0.6B-Q4_0.gguf",
            "--host",
            "127.0.0.1",
            "--port",
            "34197",
            "--threads",
            "1",
            "--threads-batch",
            "1",
            "--batch-size",
            "128",
            "--ctx-size",
            "2048",
            "--metrics"
          ],
          "server_log": "/home/runner/work/a64forge/a64forge/benchmarks/run-1488b44fa409/summarize/small_Q4_0_t1_b128_c2048.server.log",
          "server_log_tail": "abl: updating prompt cache\n0.05.663.603 W srv   prompt_save:  - saving prompt with length 100, total state size = 10.939 MiB (draft: 0.000 MiB)\n0.05.666.674 I srv          load:  - looking for better prompt, base f_keep = 0.310, sim = 0.477\n0.05.666.686 I srv          load:  - found better prompt with f_keep = 0.650, sim = 1.000\n0.05.667.990 I srv        update:  - cache state: 1 prompts, 10.939 MiB (limits: 8192.000 MiB, 2048 tokens, 74885 est)\n0.05.668.001 I srv        update:    - prompt 0xab01c6ce2610:     100 tokens, checkpoints:  0,    10.939 MiB\n0.05.668.002 I srv  get_availabl: prompt cache update took 4.42 ms\n0.05.668.056 I slot launch_slot_: id  3 | task 114 | processing task, is_child = 0\n0.05.668.061 I slot process_sing: id  0 | task -1 | saving idle slot to prompt cache\n0.05.668.061 I slot prompt_clear: id  0 | task -1 | clearing prompt with 0 tokens\n0.05.668.065 I slot process_sing: id  1 | task -1 | saving idle slot to prompt cache\n0.05.668.066 I slot prompt_clear: id  1 | task -1 | clearing prompt with 0 tokens\n0.05.668.067 I slot process_sing: id  2 | task -1 | saving idle slot to prompt cache\n0.05.668.068 I slot prompt_clear: id  2 | task -1 | clearing prompt with 0 tokens\n0.07.282.239 I slot print_timing: id  3 | task 114 | prompt eval time =     511.85 ms /    65 tokens (    7.87 ms per token,   126.99 tokens per second)\n0.07.282.254 I slot print_timing: id  3 | task 114 |        eval time =    1102.29 ms /    36 tokens (   30.62 ms per token,    32.66 tokens per second)\n0.07.282.254 I slot print_timing: id  3 | task 114 |       total time =    1614.14 ms /   101 tokens\n0.07.282.255 I slot print_timing: id  3 | task 114 |    graphs reused =        137\n0.07.282.287 I slot      release: id  3 | task 114 | stop processing: n_tokens = 100, truncated = 0\n0.07.282.301 I srv  update_slots: all slots are idle\n0.07.289.423 I srv  params_from_: Chat format: peg-native\n0.07.289.602 I slot get_availabl: id  3 | task -1 | selected slot by LCP similarity, sim_best = 0.477 (> 0.100 thold), f_keep = 0.310\n0.07.289.607 I srv  get_availabl: updating prompt cache\n0.07.289.623 W srv   prompt_save:  - saving prompt with length 100, total state size = 10.939 MiB (draft: 0.000 MiB)\n0.07.292.568 I srv          load:  - looking for better prompt, base f_keep = 0.310, sim = 0.477\n0.07.292.578 I srv          load:  - found better prompt with f_keep = 0.650, sim = 1.000\n0.07.293.803 I srv        update:  - cache state: 1 prompts, 10.939 MiB (limits: 8192.000 MiB, 2048 tokens, 74885 est)\n0.07.293.813 I srv        update:    - prompt 0xab01c3a37620:     100 tokens, checkpoints:  0,    10.939 MiB\n0.07.293.814 I srv  get_availabl: prompt cache update took 4.21 ms\n0.07.293.866 I slot launch_slot_: id  3 | task 152 | processing task, is_child = 0\n0.07.293.870 I slot process_sing: id  0 | task -1 | saving idle slot to prompt cache\n0.07.293.871 I slot prompt_clear: id  0 | task -1 | clearing prompt with 0 tokens\n0.07.293.875 I slot process_sing: id  1 | task -1 | saving idle slot to prompt cache\n0.07.293.875 I slot prompt_clear: id  1 | task -1 | clearing prompt with 0 tokens\n0.07.293.877 I slot process_sing: id  2 | task -1 | saving idle slot to prompt cache\n0.07.293.877 I slot prompt_clear: id  2 | task -1 | clearing prompt with 0 tokens\n0.08.909.239 I slot print_timing: id  3 | task 152 | prompt eval time =     511.37 ms /    65 tokens (    7.87 ms per token,   127.11 tokens per second)\n0.08.909.253 I slot print_timing: id  3 | task 152 |        eval time =    1103.96 ms /    36 tokens (   30.67 ms per token,    32.61 tokens per second)\n0.08.909.254 I slot print_timing: id  3 | task 152 |       total time =    1615.33 ms /   101 tokens\n0.08.909.255 I slot print_timing: id  3 | task 152 |    graphs reused =        171\n0.08.909.287 I slot      release: id  3 | task 152 | stop processing: n_tokens = 100, truncated = 0\n0.08.909.301 I srv  update_slots: all slots are idle\n0.08.949.931 I srv    operator(): operator(): cleaning up before exit...\n",
          "baseline": false
        }
      },
      "pareto_keys": [
        "small:Q4_0:t1:b128:c2048",
        "small:Q4_0:t1:b64:c2048"
      ],
      "explanation": [
        "Meets quality floor 0.950 with 1.000.",
        "Selected from 2 Pareto-efficient candidates using the balanced preset.",
        "Median latency: 89.3% lower than baseline.",
        "Peak memory: 80.8% lower than baseline."
      ],
      "score": 0.8
    }
  ],
  "rejections": [],
  "baseline": [
    {
      "run_id": "run-1488b44fa409",
      "timestamp": "2026-08-12T22:22:52.341896Z",
      "git_commit": "b414b9e8d52ea72dc909b92d87ba8a0efbd9c2d4",
      "hostname": "runnervma9114",
      "architecture": "aarch64",
      "cpu": "aarch64",
      "cores": 4,
      "ram_gb": 15.57,
      "stage_id": "classify",
      "model": "large",
      "model_repo": "ggml-org/Qwen3-4B-GGUF",
      "model_hash": "ab27b9bfa375a178d6cba48f3ad892b94b7739659dcc7aae8058ce0ffed6b328",
      "quantization": "Q4_K_M",
      "threads": 1,
      "batch_size": 64,
      "context_size": 512,
      "prompt_tokens": 334,
      "generated_tokens": 36,
      "latency_ms": 6510.327356500037,
      "median_latency_ms": 6404.471022500047,
      "p95_latency_ms": 6745.650973250008,
      "ttft_ms": null,
      "prompt_tokens_per_second": 10.67890021994837,
      "generation_tokens_per_second": 4.6447409965507305,
      "total_tokens_per_second": 9.47212993907249,
      "requests_per_minute": 9.216126427205667,
      "peak_memory_mb": 4885.046875,
      "average_cpu_percent": 99.8595882990249,
      "peak_cpu_percent": 132.9,
      "model_size_bytes": 2497280640,
      "load_time_ms": 3315.4892599999357,
      "quality_score": 1.0,
      "measured_runs": 6,
      "warmup_runs": 1,
      "run_label": "VERIFIED ARM64 RUN",
      "verified_arm64": true,
      "dataset_hash": "a2beee5f8f7f65cd35792936cc247f241551d0e9c1155d23a9ca863f042f7736",
      "error": null,
      "raw_artifact": "\\home\\runner\\work\\a64forge\\a64forge\\benchmarks\\run-1488b44fa409\\classify\\large_Q4_K_M_t1_b64_c512.json",
      "performix_artifact": null,
      "metadata": {
        "command": [
          "/home/runner/work/a64forge/a64forge/.a64forge/llama.cpp/build/bin/llama-server",
          "-m",
          "/home/runner/work/a64forge/a64forge/models/Qwen3-4B-Q4_K_M.gguf",
          "--host",
          "127.0.0.1",
          "--port",
          "48835",
          "--threads",
          "1",
          "--threads-batch",
          "1",
          "--batch-size",
          "64",
          "--ctx-size",
          "512",
          "--metrics"
        ],
        "server_log": "/home/runner/work/a64forge/a64forge/benchmarks/run-1488b44fa409/classify/large_Q4_K_M_t1_b64_c512.server.log",
        "server_log_tail": "d  0 | task -1 | saving idle slot to prompt cache\n0.29.252.631 I slot prompt_clear: id  0 | task -1 | clearing prompt with 0 tokens\n0.29.252.634 I slot process_sing: id  1 | task -1 | saving idle slot to prompt cache\n0.29.252.634 I slot prompt_clear: id  1 | task -1 | clearing prompt with 0 tokens\n0.29.252.635 I slot process_sing: id  2 | task -1 | saving idle slot to prompt cache\n0.29.252.635 I slot prompt_clear: id  2 | task -1 | clearing prompt with 0 tokens\n0.35.653.756 I slot print_timing: id  3 | task 32 | prompt eval time =    5095.98 ms /    53 tokens (   96.15 ms per token,    10.40 tokens per second)\n0.35.653.774 I slot print_timing: id  3 | task 32 |        eval time =    1305.10 ms /     6 tokens (  217.52 ms per token,     4.60 tokens per second)\n0.35.653.774 I slot print_timing: id  3 | task 32 |       total time =    6401.08 ms /    59 tokens\n0.35.653.775 I slot print_timing: id  3 | task 32 |    graphs reused =         21\n0.35.653.810 I slot      release: id  3 | task 32 | stop processing: n_tokens = 58, truncated = 0\n0.35.653.824 I srv  update_slots: all slots are idle\n0.35.661.053 I srv  params_from_: Chat format: peg-native\n0.35.661.229 I slot get_availabl: id  3 | task -1 | selected slot by LCP similarity, sim_best = 0.576 (> 0.100 thold), f_keep = 0.586\n0.35.661.283 I slot launch_slot_: id  3 | task 40 | processing task, is_child = 0\n0.35.661.288 I slot process_sing: id  0 | task -1 | saving idle slot to prompt cache\n0.35.661.289 I slot prompt_clear: id  0 | task -1 | clearing prompt with 0 tokens\n0.35.661.292 I slot process_sing: id  1 | task -1 | saving idle slot to prompt cache\n0.35.661.292 I slot prompt_clear: id  1 | task -1 | clearing prompt with 0 tokens\n0.35.661.293 I slot process_sing: id  2 | task -1 | saving idle slot to prompt cache\n0.35.661.293 I slot prompt_clear: id  2 | task -1 | clearing prompt with 0 tokens\n0.42.385.680 I slot print_timing: id  3 | task 40 | prompt eval time =    5429.93 ms /    59 tokens (   92.03 ms per token,    10.87 tokens per second)\n0.42.385.702 I slot print_timing: id  3 | task 40 |        eval time =    1294.43 ms /     6 tokens (  215.74 ms per token,     4.64 tokens per second)\n0.42.385.703 I slot print_timing: id  3 | task 40 |       total time =    6724.36 ms /    65 tokens\n0.42.385.704 I slot print_timing: id  3 | task 40 |    graphs reused =         25\n0.42.385.751 I slot      release: id  3 | task 40 | stop processing: n_tokens = 64, truncated = 0\n0.42.385.765 I srv  update_slots: all slots are idle\n0.42.392.931 I srv  params_from_: Chat format: peg-native\n0.42.393.113 I slot get_availabl: id  3 | task -1 | selected slot by LCP similarity, sim_best = 0.618 (> 0.100 thold), f_keep = 0.531\n0.42.393.168 I slot launch_slot_: id  3 | task 48 | processing task, is_child = 0\n0.42.393.173 I slot process_sing: id  0 | task -1 | saving idle slot to prompt cache\n0.42.393.174 I slot prompt_clear: id  0 | task -1 | clearing prompt with 0 tokens\n0.42.393.177 I slot process_sing: id  1 | task -1 | saving idle slot to prompt cache\n0.42.393.178 I slot prompt_clear: id  1 | task -1 | clearing prompt with 0 tokens\n0.42.393.178 I slot process_sing: id  2 | task -1 | saving idle slot to prompt cache\n0.42.393.179 I slot prompt_clear: id  2 | task -1 | clearing prompt with 0 tokens\n0.48.775.891 I slot print_timing: id  3 | task 48 | prompt eval time =    5096.28 ms /    55 tokens (   92.66 ms per token,    10.79 tokens per second)\n0.48.775.911 I slot print_timing: id  3 | task 48 |        eval time =    1286.40 ms /     6 tokens (  214.40 ms per token,     4.66 tokens per second)\n0.48.775.911 I slot print_timing: id  3 | task 48 |       total time =    6382.69 ms /    61 tokens\n0.48.775.913 I slot print_timing: id  3 | task 48 |    graphs reused =         29\n0.48.775.946 I slot      release: id  3 | task 48 | stop processing: n_tokens = 60, truncated = 0\n0.48.775.961 I srv  update_slots: all slots are idle\n0.48.785.659 I srv    operator(): operator(): cleaning up before exit...\n",
        "baseline": true
      }
    },
    {
      "run_id": "run-1488b44fa409",
      "timestamp": "2026-08-12T22:25:23.293415Z",
      "git_commit": "b414b9e8d52ea72dc909b92d87ba8a0efbd9c2d4",
      "hostname": "runnervma9114",
      "architecture": "aarch64",
      "cpu": "aarch64",
      "cores": 4,
      "ram_gb": 15.57,
      "stage_id": "extract",
      "model": "large",
      "model_repo": "ggml-org/Qwen3-4B-GGUF",
      "model_hash": "ab27b9bfa375a178d6cba48f3ad892b94b7739659dcc7aae8058ce0ffed6b328",
      "quantization": "Q4_K_M",
      "threads": 1,
      "batch_size": 64,
      "context_size": 512,
      "prompt_tokens": 228,
      "generated_tokens": 160,
      "latency_ms": 15300.861782499966,
      "median_latency_ms": 15294.76051499995,
      "p95_latency_ms": 15761.63199965,
      "ttft_ms": null,
      "prompt_tokens_per_second": 10.801414693387148,
      "generation_tokens_per_second": 3.9950533782741164,
      "total_tokens_per_second": 6.33951220387741,
      "requests_per_minute": 3.921347754975718,
      "peak_memory_mb": 4898.98046875,
      "average_cpu_percent": 99.84252948885977,
      "peak_cpu_percent": 133.5,
      "model_size_bytes": 2497280640,
      "load_time_ms": 3313.5400689999415,
      "quality_score": 0.6666666666666666,
      "measured_runs": 4,
      "warmup_runs": 1,
      "run_label": "VERIFIED ARM64 RUN",
      "verified_arm64": true,
      "dataset_hash": "737bfd42a8b1a6f18928b8f6db197d00c63cdf9e0359a6c2656f605b49d1244f",
      "error": null,
      "raw_artifact": "\\home\\runner\\work\\a64forge\\a64forge\\benchmarks\\run-1488b44fa409\\extract\\large_Q4_K_M_t1_b64_c512.json",
      "performix_artifact": null,
      "metadata": {
        "command": [
          "/home/runner/work/a64forge/a64forge/.a64forge/llama.cpp/build/bin/llama-server",
          "-m",
          "/home/runner/work/a64forge/a64forge/models/Qwen3-4B-Q4_K_M.gguf",
          "--host",
          "127.0.0.1",
          "--port",
          "54525",
          "--threads",
          "1",
          "--threads-batch",
          "1",
          "--batch-size",
          "64",
          "--ctx-size",
          "512",
          "--metrics"
        ],
        "server_log": "/home/runner/work/a64forge/a64forge/benchmarks/run-1488b44fa409/extract/large_Q4_K_M_t1_b64_c512.server.log",
        "server_log_tail": "_availabl: updating prompt cache\n0.49.635.536 W srv   prompt_save:  - saving prompt with length 93, total state size = 13.080 MiB (draft: 0.000 MiB)\n0.49.639.317 I srv          load:  - looking for better prompt, base f_keep = 0.333, sim = 0.525\n0.49.639.329 I srv          load:  - found better prompt with f_keep = 0.596, sim = 1.000\n0.49.640.786 I srv        update:  - cache state: 1 prompts, 13.080 MiB (limits: 8192.000 MiB, 512 tokens, 58245 est)\n0.49.640.795 I srv        update:    - prompt 0xab435974ece0:      93 tokens, checkpoints:  0,    13.080 MiB\n0.49.640.796 I srv  get_availabl: prompt cache update took 5.28 ms\n0.49.640.862 I slot launch_slot_: id  3 | task 127 | processing task, is_child = 0\n0.49.640.867 I slot process_sing: id  0 | task -1 | saving idle slot to prompt cache\n0.49.640.868 I slot prompt_clear: id  0 | task -1 | clearing prompt with 0 tokens\n0.49.640.871 I slot process_sing: id  1 | task -1 | saving idle slot to prompt cache\n0.49.640.872 I slot prompt_clear: id  1 | task -1 | clearing prompt with 0 tokens\n0.49.640.873 I slot process_sing: id  2 | task -1 | saving idle slot to prompt cache\n0.49.640.873 I slot prompt_clear: id  2 | task -1 | clearing prompt with 0 tokens\n1.05.393.799 I slot print_timing: id  3 | task 127 | prompt eval time =    5455.44 ms /    59 tokens (   92.47 ms per token,    10.81 tokens per second)\n1.05.393.819 I slot print_timing: id  3 | task 127 |        eval time =   10297.46 ms /    41 tokens (  251.16 ms per token,     3.98 tokens per second)\n1.05.393.819 I slot print_timing: id  3 | task 127 |       total time =   15752.90 ms /   100 tokens\n1.05.393.820 I slot print_timing: id  3 | task 127 |    graphs reused =        155\n1.05.393.865 I slot      release: id  3 | task 127 | stop processing: n_tokens = 99, truncated = 0\n1.05.393.880 I srv  update_slots: all slots are idle\n1.05.400.998 I srv  params_from_: Chat format: peg-native\n1.05.401.172 I slot get_availabl: id  3 | task -1 | selected slot by LCP similarity, sim_best = 0.564 (> 0.100 thold), f_keep = 0.313\n1.05.401.178 I srv  get_availabl: updating prompt cache\n1.05.401.195 W srv   prompt_save:  - saving prompt with length 99, total state size = 13.924 MiB (draft: 0.000 MiB)\n1.05.404.979 I srv          load:  - looking for better prompt, base f_keep = 0.313, sim = 0.564\n1.05.404.990 I srv          load:  - found better prompt with f_keep = 0.591, sim = 1.000\n1.05.406.415 I srv        update:  - cache state: 1 prompts, 13.924 MiB (limits: 8192.000 MiB, 512 tokens, 58245 est)\n1.05.406.422 I srv        update:    - prompt 0xab4359ea2570:      99 tokens, checkpoints:  0,    13.924 MiB\n1.05.406.423 I srv  get_availabl: prompt cache update took 5.25 ms\n1.05.406.479 I slot launch_slot_: id  3 | task 170 | processing task, is_child = 0\n1.05.406.483 I slot process_sing: id  0 | task -1 | saving idle slot to prompt cache\n1.05.406.484 I slot prompt_clear: id  0 | task -1 | clearing prompt with 0 tokens\n1.05.406.486 I slot process_sing: id  1 | task -1 | saving idle slot to prompt cache\n1.05.406.487 I slot prompt_clear: id  1 | task -1 | clearing prompt with 0 tokens\n1.05.406.488 I slot process_sing: id  2 | task -1 | saving idle slot to prompt cache\n1.05.406.489 I slot prompt_clear: id  2 | task -1 | clearing prompt with 0 tokens\n1.20.242.239 I slot print_timing: id  3 | task 170 | prompt eval time =    5100.94 ms /    55 tokens (   92.74 ms per token,    10.78 tokens per second)\n1.20.242.255 I slot print_timing: id  3 | task 170 |        eval time =    9734.78 ms /    39 tokens (  249.61 ms per token,     4.01 tokens per second)\n1.20.242.256 I slot print_timing: id  3 | task 170 |       total time =   14835.72 ms /    94 tokens\n1.20.242.257 I slot print_timing: id  3 | task 170 |    graphs reused =        192\n1.20.242.290 I slot      release: id  3 | task 170 | stop processing: n_tokens = 93, truncated = 0\n1.20.242.304 I srv  update_slots: all slots are idle\n1.20.243.758 I srv    operator(): operator(): cleaning up before exit...\n",
        "baseline": true
      }
    },
    {
      "run_id": "run-1488b44fa409",
      "timestamp": "2026-08-12T22:27:05.677066Z",
      "git_commit": "b414b9e8d52ea72dc909b92d87ba8a0efbd9c2d4",
      "hostname": "runnervma9114",
      "architecture": "aarch64",
      "cpu": "aarch64",
      "cores": 4,
      "ram_gb": 15.57,
      "stage_id": "tool_select",
      "model": "large",
      "model_repo": "ggml-org/Qwen3-4B-GGUF",
      "model_hash": "ab27b9bfa375a178d6cba48f3ad892b94b7739659dcc7aae8058ce0ffed6b328",
      "quantization": "Q4_K_M",
      "threads": 1,
      "batch_size": 64,
      "context_size": 512,
      "prompt_tokens": 428,
      "generated_tokens": 50,
      "latency_ms": 6498.982136625045,
      "median_latency_ms": 6373.08005050005,
      "p95_latency_ms": 7149.208977150056,
      "ttft_ms": null,
      "prompt_tokens_per_second": 10.409228070463646,
      "generation_tokens_per_second": 4.630465533827807,
      "total_tokens_per_second": 9.193747381344316,
      "requests_per_minute": 9.232214943609355,
      "peak_memory_mb": 4885.0546875,
      "average_cpu_percent": 100.16652506372132,
      "peak_cpu_percent": 448.9,
      "model_size_bytes": 2497280640,
      "load_time_ms": 3317.842523999957,
      "quality_score": 1.0,
      "measured_runs": 8,
      "warmup_runs": 1,
      "run_label": "VERIFIED ARM64 RUN",
      "verified_arm64": true,
      "dataset_hash": "5e97a2f29f08e1cc6be3cf033845890eb3eebdae6b58ec7f76c2dd04d22f1e72",
      "error": null,
      "raw_artifact": "\\home\\runner\\work\\a64forge\\a64forge\\benchmarks\\run-1488b44fa409\\tool_select\\large_Q4_K_M_t1_b64_c512.json",
      "performix_artifact": null,
      "metadata": {
        "command": [
          "/home/runner/work/a64forge/a64forge/.a64forge/llama.cpp/build/bin/llama-server",
          "-m",
          "/home/runner/work/a64forge/a64forge/models/Qwen3-4B-Q4_K_M.gguf",
          "--host",
          "127.0.0.1",
          "--port",
          "47089",
          "--threads",
          "1",
          "--threads-batch",
          "1",
          "--batch-size",
          "64",
          "--ctx-size",
          "512",
          "--metrics"
        ],
        "server_log": "/home/runner/work/a64forge/a64forge/benchmarks/run-1488b44fa409/tool_select/large_Q4_K_M_t1_b64_c512.server.log",
        "server_log_tail": "d  0 | task -1 | saving idle slot to prompt cache\n0.42.358.287 I slot prompt_clear: id  0 | task -1 | clearing prompt with 0 tokens\n0.42.358.291 I slot process_sing: id  1 | task -1 | saving idle slot to prompt cache\n0.42.358.291 I slot prompt_clear: id  1 | task -1 | clearing prompt with 0 tokens\n0.42.358.292 I slot process_sing: id  2 | task -1 | saving idle slot to prompt cache\n0.42.358.292 I slot prompt_clear: id  2 | task -1 | clearing prompt with 0 tokens\n0.49.518.575 I slot print_timing: id  3 | task 51 | prompt eval time =    5871.21 ms /    61 tokens (   96.25 ms per token,    10.39 tokens per second)\n0.49.518.592 I slot print_timing: id  3 | task 51 |        eval time =    1289.05 ms /     6 tokens (  214.84 ms per token,     4.65 tokens per second)\n0.49.518.593 I slot print_timing: id  3 | task 51 |       total time =    7160.26 ms /    67 tokens\n0.49.518.593 I slot print_timing: id  3 | task 51 |    graphs reused =         32\n0.49.518.627 I slot      release: id  3 | task 51 | stop processing: n_tokens = 66, truncated = 0\n0.49.518.640 I srv  update_slots: all slots are idle\n0.49.525.563 I srv  params_from_: Chat format: peg-native\n0.49.525.749 I slot get_availabl: id  3 | task -1 | selected slot by LCP similarity, sim_best = 0.680 (> 0.100 thold), f_keep = 0.515\n0.49.525.804 I slot launch_slot_: id  3 | task 59 | processing task, is_child = 0\n0.49.525.808 I slot process_sing: id  0 | task -1 | saving idle slot to prompt cache\n0.49.525.810 I slot prompt_clear: id  0 | task -1 | clearing prompt with 0 tokens\n0.49.525.813 I slot process_sing: id  1 | task -1 | saving idle slot to prompt cache\n0.49.525.813 I slot prompt_clear: id  1 | task -1 | clearing prompt with 0 tokens\n0.49.525.814 I slot process_sing: id  2 | task -1 | saving idle slot to prompt cache\n0.49.525.814 I slot prompt_clear: id  2 | task -1 | clearing prompt with 0 tokens\n0.55.767.518 I slot print_timing: id  3 | task 59 | prompt eval time =    4957.69 ms /    50 tokens (   99.15 ms per token,    10.09 tokens per second)\n0.55.767.532 I slot print_timing: id  3 | task 59 |        eval time =    1283.98 ms /     6 tokens (  214.00 ms per token,     4.67 tokens per second)\n0.55.767.533 I slot print_timing: id  3 | task 59 |       total time =    6241.68 ms /    56 tokens\n0.55.767.534 I slot print_timing: id  3 | task 59 |    graphs reused =         36\n0.55.767.567 I slot      release: id  3 | task 59 | stop processing: n_tokens = 55, truncated = 0\n0.55.767.581 I srv  update_slots: all slots are idle\n0.55.774.770 I srv  params_from_: Chat format: peg-native\n0.55.774.948 I slot get_availabl: id  3 | task -1 | selected slot by LCP similarity, sim_best = 0.667 (> 0.100 thold), f_keep = 0.618\n0.55.775.002 I slot launch_slot_: id  3 | task 67 | processing task, is_child = 0\n0.55.775.008 I slot process_sing: id  0 | task -1 | saving idle slot to prompt cache\n0.55.775.008 I slot prompt_clear: id  0 | task -1 | clearing prompt with 0 tokens\n0.55.775.011 I slot process_sing: id  1 | task -1 | saving idle slot to prompt cache\n0.55.775.012 I slot prompt_clear: id  1 | task -1 | clearing prompt with 0 tokens\n0.55.775.013 I slot process_sing: id  2 | task -1 | saving idle slot to prompt cache\n0.55.775.013 I slot prompt_clear: id  2 | task -1 | clearing prompt with 0 tokens\n1.01.827.602 I slot print_timing: id  3 | task 67 | prompt eval time =    4766.45 ms /    51 tokens (   93.46 ms per token,    10.70 tokens per second)\n1.01.827.622 I slot print_timing: id  3 | task 67 |        eval time =    1286.12 ms /     6 tokens (  214.35 ms per token,     4.67 tokens per second)\n1.01.827.622 I slot print_timing: id  3 | task 67 |       total time =    6052.56 ms /    57 tokens\n1.01.827.623 I slot print_timing: id  3 | task 67 |    graphs reused =         40\n1.01.827.656 I slot      release: id  3 | task 67 | stop processing: n_tokens = 56, truncated = 0\n1.01.827.671 I srv  update_slots: all slots are idle\n1.01.848.066 I srv    operator(): operator(): cleaning up before exit...\n",
        "baseline": true
      }
    },
    {
      "run_id": "run-1488b44fa409",
      "timestamp": "2026-08-12T22:30:12.985095Z",
      "git_commit": "b414b9e8d52ea72dc909b92d87ba8a0efbd9c2d4",
      "hostname": "runnervma9114",
      "architecture": "aarch64",
      "cpu": "aarch64",
      "cores": 4,
      "ram_gb": 15.57,
      "stage_id": "reason",
      "model": "large",
      "model_repo": "ggml-org/Qwen3-4B-GGUF",
      "model_hash": "ab27b9bfa375a178d6cba48f3ad892b94b7739659dcc7aae8058ce0ffed6b328",
      "quantization": "Q4_K_M",
      "threads": 1,
      "batch_size": 64,
      "context_size": 512,
      "prompt_tokens": 334,
      "generated_tokens": 158,
      "latency_ms": 17991.955557750032,
      "median_latency_ms": 17979.77963050005,
      "p95_latency_ms": 20273.33615255003,
      "ttft_ms": null,
      "prompt_tokens_per_second": 10.597220332315299,
      "generation_tokens_per_second": 3.9546477386013397,
      "total_tokens_per_second": 6.836388607408368,
      "requests_per_minute": 3.3348237109309116,
      "peak_memory_mb": 4908.703125,
      "average_cpu_percent": 99.92554945054945,
      "peak_cpu_percent": 114.9,
      "model_size_bytes": 2497280640,
      "load_time_ms": 3314.5158639999863,
      "quality_score": 0.7083333333333333,
      "measured_runs": 4,
      "warmup_runs": 1,
      "run_label": "VERIFIED ARM64 RUN",
      "verified_arm64": true,
      "dataset_hash": "74c11a25259eb389353c38803fd559026426a07a30d5523f65d3ee707dc12e46",
      "error": null,
      "raw_artifact": "\\home\\runner\\work\\a64forge\\a64forge\\benchmarks\\run-1488b44fa409\\reason\\large_Q4_K_M_t1_b64_c512.json",
      "performix_artifact": null,
      "metadata": {
        "command": [
          "/home/runner/work/a64forge/a64forge/.a64forge/llama.cpp/build/bin/llama-server",
          "-m",
          "/home/runner/work/a64forge/a64forge/models/Qwen3-4B-Q4_K_M.gguf",
          "--host",
          "127.0.0.1",
          "--port",
          "53351",
          "--threads",
          "1",
          "--threads-batch",
          "1",
          "--batch-size",
          "64",
          "--ctx-size",
          "512",
          "--metrics"
        ],
        "server_log": "/home/runner/work/a64forge/a64forge/benchmarks/run-1488b44fa409/reason/large_Q4_K_M_t1_b64_c512.server.log",
        "server_log_tail": ".59.603.320 I srv          load:  - looking for better prompt, base f_keep = 0.311, sim = 0.337\n0.59.603.332 I srv          load:  - found better prompt with f_keep = 0.710, sim = 1.000\n0.59.605.361 I srv        update:  - cache state: 1 prompts, 14.908 MiB (limits: 8192.000 MiB, 512 tokens, 58246 est)\n0.59.605.369 I srv        update:    - prompt 0xaad59b617ce0:     106 tokens, checkpoints:  0,    14.908 MiB\n0.59.605.370 I srv  get_availabl: prompt cache update took 6.16 ms\n0.59.605.428 I slot launch_slot_: id  3 | task 128 | processing task, is_child = 0\n0.59.605.433 I slot process_sing: id  0 | task -1 | saving idle slot to prompt cache\n0.59.605.434 I slot prompt_clear: id  0 | task -1 | clearing prompt with 0 tokens\n0.59.605.437 I slot process_sing: id  1 | task -1 | saving idle slot to prompt cache\n0.59.605.437 I slot prompt_clear: id  1 | task -1 | clearing prompt with 0 tokens\n0.59.605.438 I slot process_sing: id  2 | task -1 | saving idle slot to prompt cache\n0.59.605.438 I slot prompt_clear: id  2 | task -1 | clearing prompt with 0 tokens\n1.08.627.982 I slot print_timing: id  3 | task 128 | prompt processing, n_tokens =     92, progress = 0.94, t =   9.02 s / 10.20 tokens per second\n1.19.822.055 I slot print_timing: id  3 | task 128 | prompt eval time =    9834.11 ms /    98 tokens (  100.35 ms per token,     9.97 tokens per second)\n1.19.822.076 I slot print_timing: id  3 | task 128 |        eval time =   10382.48 ms /    41 tokens (  253.23 ms per token,     3.95 tokens per second)\n1.19.822.077 I slot print_timing: id  3 | task 128 |       total time =   20216.59 ms /   139 tokens\n1.19.822.078 I slot print_timing: id  3 | task 128 |    graphs reused =        154\n1.19.822.116 I slot      release: id  3 | task 128 | stop processing: n_tokens = 138, truncated = 0\n1.19.822.130 I srv  update_slots: all slots are idle\n1.19.829.283 I srv  params_from_: Chat format: peg-native\n1.19.829.463 I slot get_availabl: id  3 | task -1 | selected slot by LCP similarity, sim_best = 0.478 (> 0.100 thold), f_keep = 0.239\n1.19.829.469 I srv  get_availabl: updating prompt cache\n1.19.829.486 W srv   prompt_save:  - saving prompt with length 138, total state size = 19.409 MiB (draft: 0.000 MiB)\n1.19.834.834 I srv          load:  - looking for better prompt, base f_keep = 0.239, sim = 0.478\n1.19.834.850 I srv          load:  - found better prompt with f_keep = 0.651, sim = 1.000\n1.19.836.363 I srv        update:  - cache state: 1 prompts, 19.409 MiB (limits: 8192.000 MiB, 512 tokens, 58246 est)\n1.19.836.379 I srv        update:    - prompt 0xaad59bd6b570:     138 tokens, checkpoints:  0,    19.409 MiB\n1.19.836.380 I srv  get_availabl: prompt cache update took 6.91 ms\n1.19.836.435 I slot launch_slot_: id  3 | task 172 | processing task, is_child = 0\n1.19.836.440 I slot process_sing: id  0 | task -1 | saving idle slot to prompt cache\n1.19.836.441 I slot prompt_clear: id  0 | task -1 | clearing prompt with 0 tokens\n1.19.836.444 I slot process_sing: id  1 | task -1 | saving idle slot to prompt cache\n1.19.836.444 I slot prompt_clear: id  1 | task -1 | clearing prompt with 0 tokens\n1.19.836.445 I slot process_sing: id  2 | task -1 | saving idle slot to prompt cache\n1.19.836.445 I slot prompt_clear: id  2 | task -1 | clearing prompt with 0 tokens\n1.35.551.254 I slot print_timing: id  3 | task 172 | prompt eval time =    6147.51 ms /    69 tokens (   89.09 ms per token,    11.22 tokens per second)\n1.35.551.274 I slot print_timing: id  3 | task 172 |        eval time =    9567.27 ms /    38 tokens (  251.77 ms per token,     3.97 tokens per second)\n1.35.551.275 I slot print_timing: id  3 | task 172 |       total time =   15714.78 ms /   107 tokens\n1.35.551.276 I slot print_timing: id  3 | task 172 |    graphs reused =        190\n1.35.551.312 I slot      release: id  3 | task 172 | stop processing: n_tokens = 106, truncated = 0\n1.35.551.326 I srv  update_slots: all slots are idle\n1.35.558.865 I srv    operator(): operator(): cleaning up before exit...\n",
        "baseline": true
      }
    },
    {
      "run_id": "run-1488b44fa409",
      "timestamp": "2026-08-12T22:32:40.170846Z",
      "git_commit": "b414b9e8d52ea72dc909b92d87ba8a0efbd9c2d4",
      "hostname": "runnervma9114",
      "architecture": "aarch64",
      "cpu": "aarch64",
      "cores": 4,
      "ram_gb": 15.57,
      "stage_id": "summarize",
      "model": "large",
      "model_repo": "ggml-org/Qwen3-4B-GGUF",
      "model_hash": "ab27b9bfa375a178d6cba48f3ad892b94b7739659dcc7aae8058ce0ffed6b328",
      "quantization": "Q4_K_M",
      "threads": 1,
      "batch_size": 64,
      "context_size": 512,
      "prompt_tokens": 260,
      "generated_tokens": 144,
      "latency_ms": 15189.06903324995,
      "median_latency_ms": 15190.18536149997,
      "p95_latency_ms": 15205.193758699954,
      "ttft_ms": null,
      "prompt_tokens_per_second": 10.584576821834933,
      "generation_tokens_per_second": 3.9836926777731194,
      "total_tokens_per_second": 6.649518794002702,
      "requests_per_minute": 3.9502091845560607,
      "peak_memory_mb": 4911.7578125,
      "average_cpu_percent": 99.84860927152317,
      "peak_cpu_percent": 114.9,
      "model_size_bytes": 2497280640,
      "load_time_ms": 3316.409096999905,
      "quality_score": 1.0,
      "measured_runs": 4,
      "warmup_runs": 1,
      "run_label": "VERIFIED ARM64 RUN",
      "verified_arm64": true,
      "dataset_hash": "f524c6a0b252a3d0e01a44701914354f6130c685e7d862f2bfdd4d9d5686efdd",
      "error": null,
      "raw_artifact": "\\home\\runner\\work\\a64forge\\a64forge\\benchmarks\\run-1488b44fa409\\summarize\\large_Q4_K_M_t1_b64_c512.json",
      "performix_artifact": null,
      "metadata": {
        "command": [
          "/home/runner/work/a64forge/a64forge/.a64forge/llama.cpp/build/bin/llama-server",
          "-m",
          "/home/runner/work/a64forge/a64forge/models/Qwen3-4B-Q4_K_M.gguf",
          "--host",
          "127.0.0.1",
          "--port",
          "38033",
          "--threads",
          "1",
          "--threads-batch",
          "1",
          "--batch-size",
          "64",
          "--ctx-size",
          "512",
          "--metrics"
        ],
        "server_log": "/home/runner/work/a64forge/a64forge/benchmarks/run-1488b44fa409/summarize/large_Q4_K_M_t1_b64_c512.server.log",
        "server_log_tail": "ilabl: updating prompt cache\n0.48.896.049 W srv   prompt_save:  - saving prompt with length 100, total state size = 14.064 MiB (draft: 0.000 MiB)\n0.48.900.028 I srv          load:  - looking for better prompt, base f_keep = 0.310, sim = 0.477\n0.48.900.043 I srv          load:  - found better prompt with f_keep = 0.650, sim = 1.000\n0.48.901.556 I srv        update:  - cache state: 1 prompts, 14.064 MiB (limits: 8192.000 MiB, 512 tokens, 58245 est)\n0.48.901.563 I srv        update:    - prompt 0xab187d3ce890:     100 tokens, checkpoints:  0,    14.064 MiB\n0.48.901.564 I srv  get_availabl: prompt cache update took 5.53 ms\n0.48.901.620 I slot launch_slot_: id  3 | task 114 | processing task, is_child = 0\n0.48.901.625 I slot process_sing: id  0 | task -1 | saving idle slot to prompt cache\n0.48.901.626 I slot prompt_clear: id  0 | task -1 | clearing prompt with 0 tokens\n0.48.901.629 I slot process_sing: id  1 | task -1 | saving idle slot to prompt cache\n0.48.901.629 I slot prompt_clear: id  1 | task -1 | clearing prompt with 0 tokens\n0.48.901.630 I slot process_sing: id  2 | task -1 | saving idle slot to prompt cache\n0.48.901.630 I slot prompt_clear: id  2 | task -1 | clearing prompt with 0 tokens\n1.04.075.465 I slot print_timing: id  3 | task 114 | prompt eval time =    6143.15 ms /    65 tokens (   94.51 ms per token,    10.58 tokens per second)\n1.04.075.479 I slot print_timing: id  3 | task 114 |        eval time =    9030.66 ms /    36 tokens (  250.85 ms per token,     3.99 tokens per second)\n1.04.075.480 I slot print_timing: id  3 | task 114 |       total time =   15173.81 ms /   101 tokens\n1.04.075.481 I slot print_timing: id  3 | task 114 |    graphs reused =        137\n1.04.075.514 I slot      release: id  3 | task 114 | stop processing: n_tokens = 100, truncated = 0\n1.04.075.528 I srv  update_slots: all slots are idle\n1.04.082.682 I srv  params_from_: Chat format: peg-native\n1.04.082.861 I slot get_availabl: id  3 | task -1 | selected slot by LCP similarity, sim_best = 0.477 (> 0.100 thold), f_keep = 0.310\n1.04.082.867 I srv  get_availabl: updating prompt cache\n1.04.082.883 W srv   prompt_save:  - saving prompt with length 100, total state size = 14.064 MiB (draft: 0.000 MiB)\n1.04.086.825 I srv          load:  - looking for better prompt, base f_keep = 0.310, sim = 0.477\n1.04.086.841 I srv          load:  - found better prompt with f_keep = 0.650, sim = 1.000\n1.04.088.292 I srv        update:  - cache state: 1 prompts, 14.064 MiB (limits: 8192.000 MiB, 512 tokens, 58245 est)\n1.04.088.304 I srv        update:    - prompt 0xab187d3cd1b0:     100 tokens, checkpoints:  0,    14.064 MiB\n1.04.088.305 I srv  get_availabl: prompt cache update took 5.44 ms\n1.04.088.361 I slot launch_slot_: id  3 | task 152 | processing task, is_child = 0\n1.04.088.366 I slot process_sing: id  0 | task -1 | saving idle slot to prompt cache\n1.04.088.367 I slot prompt_clear: id  0 | task -1 | clearing prompt with 0 tokens\n1.04.088.370 I slot process_sing: id  1 | task -1 | saving idle slot to prompt cache\n1.04.088.371 I slot prompt_clear: id  1 | task -1 | clearing prompt with 0 tokens\n1.04.088.371 I slot process_sing: id  2 | task -1 | saving idle slot to prompt cache\n1.04.088.373 I slot prompt_clear: id  2 | task -1 | clearing prompt with 0 tokens\n1.19.269.120 I slot print_timing: id  3 | task 152 | prompt eval time =    6139.97 ms /    65 tokens (   94.46 ms per token,    10.59 tokens per second)\n1.19.269.136 I slot print_timing: id  3 | task 152 |        eval time =    9040.75 ms /    36 tokens (  251.13 ms per token,     3.98 tokens per second)\n1.19.269.136 I slot print_timing: id  3 | task 152 |       total time =   15180.72 ms /   101 tokens\n1.19.269.137 I slot print_timing: id  3 | task 152 |    graphs reused =        171\n1.19.269.172 I slot      release: id  3 | task 152 | stop processing: n_tokens = 100, truncated = 0\n1.19.269.186 I srv  update_slots: all slots are idle\n1.19.295.687 I srv    operator(): operator(): cleaning up before exit...\n",
        "baseline": true
      }
    }
  ],
  "run_label": "VERIFIED ARM64 RUN"
}