{
  "experiment": "e0-build714-4df29be4",
  "sequence": "boot scripts/gguf-quickstart.sh (CTX_SIZE=131072 WITH_MTP=0 EXTRA_ARGS='-np 8' PORT=8080, mmproj on) -> health poll -> bench_client --concurrency 8 --max-tokens 256 -> bench_client --anchor-only (greedy anchor, gate = anchor_ok field)",
  "llama_commit": "4df29be4f4c3673f428170fda944a5b19f743bb8",
  "build_dir": "third_party/llama.cpp/build-714",
  "server_flags": {
    "llama_server_flags": [
      "--ctx-size",
      "131072",
      "-ngl",
      "99",
      "--jinja",
      "-np",
      "8"
    ],
    "mmproj": "models/Qwen3.8-27B-GGUF/mmproj-F16.gguf",
    "model": "models/Qwen3.8-27B-GGUF/Qwen3.8-27B-UD-Q4_K_XL.gguf",
    "port": 8080
  },
  "slot_info": {
    "n_slots": 8,
    "n_ctx_slot": 16384,
    "kv_unified": "false",
    "log_line": "n_slots = 8, n_ctx_slot = 16384, kv_unified = 'false'"
  },
  "load": {
    "vram_mib": 1630,
    "gtt_mib": 27154
  },
  "boot": {
    "ok": true,
    "health_wall_s": 6
  },
  "boot_log_excerpt": [
    "llama-server : /home/amd/Desktop/Qwen3.8-27B-ROCm/third_party/llama.cpp/build-714/bin/llama-server (version: 0.1.0-dev (build 1, commit 4df29be4f))",
    "model        : models/Qwen3.8-27B-GGUF/Qwen3.8-27B-UD-Q4_K_XL.gguf (17G)",
    "ctx-size     : 131072  (override: CTX_SIZE=<n>)",
    "gpu layers   : 99 (all)",
    "mmproj       : models/Qwen3.8-27B-GGUF/mmproj-F16.gguf",
    "speculative  : off (opt in: WITH_MTP=1)",
    "extra args   : -np 8",
    "Serving on http://127.0.0.1:8080 ...",
    "0.00.027.425 W srv  llama_server: -----------------",
    "0.00.027.426 W srv  llama_server: CORS is set to allow all origins ('*') and no API key is set",
    "0.00.027.427 W srv  llama_server: this can be a security risk (cross-origin attacks)",
    "0.00.027.427 W srv  llama_server: more info: https://github.com/ggml-org/llama.cpp/pull/25655",
    "0.00.027.427 W srv  llama_server: -----------------",
    "0.01.160.986 W model has unused tensor blk.64.attn_norm.weight (size = 20480 bytes) -- ignoring",
    "0.01.160.992 W model has unused tensor blk.64.post_attention_norm.weight (size = 20480 bytes) -- ignoring",
    "0.01.160.996 W model has unused tensor blk.64.attn_q.weight (size = 43253760 bytes) -- ignoring",
    "0.01.160.998 W model has unused tensor blk.64.attn_k.weight (size = 3604480 bytes) -- ignoring",
    "0.01.160.999 W model has unused tensor blk.64.attn_v.weight (size = 4300800 bytes) -- ignoring",
    "0.01.161.005 W model has unused tensor blk.64.attn_output.weight (size = 21626880 bytes) -- ignoring",
    "0.01.161.006 W model has unused tensor blk.64.attn_q_norm.weight (size = 1024 bytes) -- ignoring"
  ],
  "bench_summary": {
    "command": "python3 scripts/bench_client.py --base-url http://127.0.0.1:8080 --concurrency 8 --prompts scripts/prompt-sets/default.json --max-tokens 256 --label <label> --model default --no-thinking",
    "aggregate": {
      "tok_per_s": 17.8747467872088,
      "wall_s": 114.57504961499944,
      "ok_streams": 8,
      "failed_streams": 0
    },
    "streams": [
      {
        "ttft_ms": 43688.24374600081,
        "tpot_ms": 277.9867506040461,
        "completion_tokens": 256,
        "prompt_tokens": 1412,
        "finish_reason": "length",
        "ok": true,
        "error": null
      },
      {
        "ttft_ms": 43685.95054099569,
        "tpot_ms": 277.9912691844134,
        "completion_tokens": 256,
        "prompt_tokens": 1474,
        "finish_reason": "length",
        "ok": true,
        "error": null
      },
      {
        "ttft_ms": 43115.93018198619,
        "tpot_ms": 279.6343893490677,
        "completion_tokens": 256,
        "prompt_tokens": 1298,
        "finish_reason": "length",
        "ok": true,
        "error": null
      },
      {
        "ttft_ms": 43117.21871502232,
        "tpot_ms": 279.6309204194603,
        "completion_tokens": 256,
        "prompt_tokens": 1469,
        "finish_reason": "length",
        "ok": true,
        "error": null
      },
      {
        "ttft_ms": 43682.60845198529,
        "tpot_ms": 277.995047156982,
        "completion_tokens": 256,
        "prompt_tokens": 1336,
        "finish_reason": "length",
        "ok": true,
        "error": null
      },
      {
        "ttft_ms": 37593.585060036276,
        "tpot_ms": 300.5074855136484,
        "completion_tokens": 256,
        "prompt_tokens": 1516,
        "finish_reason": "length",
        "ok": true,
        "error": null
      },
      {
        "ttft_ms": 29787.263220991008,
        "tpot_ms": 330.2255491214786,
        "completion_tokens": 256,
        "prompt_tokens": 1462,
        "finish_reason": "length",
        "ok": true,
        "error": null
      },
      {
        "ttft_ms": 29787.01346303569,
        "tpot_ms": 330.22647033326837,
        "completion_tokens": 256,
        "prompt_tokens": 1408,
        "finish_reason": "length",
        "ok": true,
        "error": null
      }
    ],
    "rc": 0
  },
  "anchor": {
    "command": "python3 scripts/bench_client.py --base-url http://127.0.0.1:8080 --anchor-only --prompts scripts/prompt-sets/default.json --max-tokens 256 --label <label>-anchor --model default --no-thinking",
    "ok": false,
    "content_tail": "////////////////",
    "expect_exact_substring": "OK"
  },
  "verdict": "pit reproduced: greedy anchor FAILED after sustained 8-stream load",
  "started_utc": "2026-08-17T23:28:08Z",
  "wall_s": 128,
  "attempts": [
    {
      "attempt": 1,
      "host_state": "16-job HIP compile of the E1 master build running concurrently (bench TTFT ~2x the cell's; aggregate 8.2 tok/s)",
      "anchor_ok": true,
      "content_tail": "OK",
      "bench_shape": "6/8 streams capped at 256; 2 streams stopped at 2 tokens (finish_reason=stop)",
      "verdict": "pit NOT reproduced",
      "note": "run discarded as a reference: not an idle host; raw log /tmp/upstream-control-e0-build714-4df29be4.log was overwritten by attempt 2"
    },
    {
      "attempt": 2,
      "host_state": "idle host (compile finished)",
      "anchor_ok": false,
      "content_tail": "////////////////",
      "bench_shape": "8/8 streams capped at 256 (finish_reason=length) \u2014 the all-capped pattern of the five degraded matrix cells",
      "verdict": "pit reproduced (receipt body = this attempt)"
    }
  ],
  "reference": "canonical degraded cell docs/results/matrix-714/cells/gguf-udq4kxl-auto-base-c8-ctx131072.json",
  "note": "no rebuild; existing build-714 binary (fingerprinted scripts/05-build-llama.sh build). Verdict = attempt 2 (idle host); attempt 1 (anchor OK) ran under concurrent compile load with 2 early-stopping streams and is recorded for reproducibility honesty."
}