{ "kind": "done", "sentence": "the phase ran to its end", "inner": { "t": 1790984452.996, "utc": "2026-10-02T23:40:52Z", "phase": "done", "runs": 9 }, "facts": { "arm": "L-IQ3-t8", "phase": "speed", "attempt": 1, "arm_cfg": { "order": 1.5, "engine": "llama.cpp", "quant": "IQ3_S", "mtp": null, "contexts": [ "4K", "32K", "128K" ], "quality": false, "pair": "S-IQ3", "band": [ 20, 55 ], "registered": true, "ncmoe_from": "L-IQ3" }, "driver_started": "2026-10-02T23:03:15Z", "unit_prefix": "sb", "driver_unit": "sb-driver-L-IQ3-t8-speed", "prereg": { "prereg_sha256": "withheld-2026-09-29", "recon_sha256": "96e4a739267cc85c250300d714faa51cc135f25ce69f61cb19fd6c286f826df0", "registered": "2026-10-02T19:26:17Z", "block_verbatim": true, "why": null }, "files": { "1": { "path": "/data/strata-bench/models/ISTA-ed59f920/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf", "pinned_sha256": "4c1eb2ceb4915e1192f4f386021897bde56a97f40a0bb78bb86465e0f7d2aca3", "receipt_sha256": "4c1eb2ceb4915e1192f4f386021897bde56a97f40a0bb78bb86465e0f7d2aca3", "receipt_at": "2026-10-02T20:08:17Z", "bytes": 54817524224, "mtime": "2026-10-02T20:07:50Z", "receipt_bytes_match": true, "ok": true }, "2": { "path": "/data/strata-bench/models/ISTA-ed59f920/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf", "pinned_sha256": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113", "receipt_sha256": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113", "receipt_at": "2026-10-02T20:06:46Z", "bytes": 28800138432, "mtime": "2026-10-02T20:06:33Z", "receipt_bytes_match": true, "ok": true } }, "n_cpu_moe_frozen": { "arm": "L-IQ3", "n_cpu_moe": 34, "evidence": "L-IQ3/fit/a7/fit-001: max_used 22449 of 24463 MiB (2014 MiB left >= 1536); N=32 (a8) and N=30 (a9) failed to load (CUDA OOM)", "frozen": "2026-10-02T20:53:14Z", "read_for": "L-IQ3-t8", "frozen_file": "FROZEN-NCMOE-L-IQ3.json" }, "engine_identity": { "exe": "/data/strata-bench/build/llama.cpp-3cf03257/bin/llama-server", "exe_realpath": "/data/strata-bench/build/llama.cpp-3cf03257/bin/llama-server", "cmake_cache": "/data/strata-bench/build/llama.cpp-3cf03257/CMakeCache.txt", "cmake_home": "/data/strata-bench/src/llama.cpp-3cf03257", "src": null, "expected_commit": "3cf03257f219afbe7334045ff7c6a06ac68c627d", "git_head": "3cf03257f219afbe7334045ff7c6a06ac68c627d", "ok": true, "why": "the build tree names the source tree, whose HEAD is the arm's commit" }, "engine_commit": "3cf03257f219afbe7334045ff7c6a06ac68c627d", "server_argv": [ "/data/strata-bench/build/llama.cpp-3cf03257/bin/llama-server", "-m", "/data/strata-bench/models/ISTA-ed59f920/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf", "--host", "127.0.0.1", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "34", "-c", "131072", "-fa", "on", "-ctk", "q8_0", "-ctv", "q8_0", "-lm", "mmap", "--lazy-mode", "on", "-fit", "off", "-np", "1", "--cache-ram", "0", "--no-webui", "--jinja", "-lv", "4", "-t", "8", "-tb", "8" ], "host_os": { "os": { "pretty_name": "Ubuntu 26.04.1 LTS", "id": "ubuntu", "version_id": "26.04", "kernel": "7.0.0-38-generic", "arch": "x86_64", "source": "/etc/os-release" }, "nvidia_driver": "595.91.07", "nvidia_driver_source": "/proc/driver/nvidia/version" }, "data_device": { "mount": "/data", "source": "/dev/nvme0n1p1", "partition": "nvme0n1p1", "disk": "nvme0n1", "model": "Samsung SSD 990 PRO 2TB", "hwmon_temp": "/sys/block/nvme0n1/device/hwmon3/temp1_input", "how": "findmnt -no SOURCE, then /sys/class/block/'s parent and its hwmon" }, "powercap": { "readable": false, "path": "/sys/class/powercap/intel-rapl:0/energy_uj", "why": "root-only (mode 0400) on this box: recorded null; /run/rapl/energy is read instead" }, "ssd_temp_at_preflight_c": 45.85, "memory_plan": { "memory_max_gib": 50, "need_gib": 40.2, "swap_max": "0", "arithmetic": "llama.cpp IQ3_S: CPU experts 34/48 x 46.84 GiB = 33.18 GiB of mmapped pages (--n-cpu-moe 34) + 3.0 GiB host buffers = 36.18 GiB working set; MemoryMax = min(working set + 14 GiB of page-cache room for the GPU layers' pages and the n-gram rows, MemTotal - 6) = 50 GiB (a page-cache reclaim at the cap shows as SSD reads/s, recorded per run, never as a kill); refuse below MemAvailable 40.2 GiB (working set + 4 GiB)." }, "lock": { "path": "/workshop/CARD-LOCK", "sentence": "LOCK TAKEN: /workshop/CARD-LOCK now names strata-bench (token 4e0d74d53b5a)" }, "watcher": { "unit": "sb-watch-L-IQ3-t8-speed-a1", "rc": 0, "argv": [ "systemd-run", "--user", "--quiet", "--unit=sb-watch-L-IQ3-t8-speed-a1", "--property=KillMode=control-group", "--property=TimeoutStopSec=30", "--working-directory=/workshop/bench-archive/plans-2026-10-02/strata-bench/harness", "--", "/usr/bin/python3", "/workshop/bench-archive/plans-2026-10-02/strata-bench/harness/restore_watcher.py", "--timer", "scribe-embed.timer", "--timer-stopped", "yes", "--engine-unit", "sb-eng-L-IQ3-t8-speed-a1", "--lock", "/workshop/CARD-LOCK", "--lock-token", "4e0d74d53b5a", "--log", "/workshop/bench-archive/plans-2026-10-02/strata-bench/records/L-IQ3-t8/speed/watcher-a1-2026-10-02T230315Z.log", "--deadline-s", "21600", "--systemctl", "systemctl --user", "--driver-unit", "sb-driver-L-IQ3-t8-speed", "--done-file", "/workshop/bench-archive/plans-2026-10-02/strata-bench/records/L-IQ3-t8/speed/watcher-a1.done" ], "log": "/workshop/bench-archive/plans-2026-10-02/strata-bench/records/L-IQ3-t8/speed/watcher-a1-2026-10-02T230315Z.log", "read_after_arming": { "armed": true, "fired": false, "line": null } }, "park": { "timer": { "timer": "scribe-embed.timer", "before": "active", "stopped_by_us": true, "stop_rc": 0, "stop_out": "", "after": "inactive" }, "ollama": { "host_is_loopback": true, "ps_before": { "rc": 0, "text": "NAME ID SIZE PROCESSOR CONTEXT UNTIL" }, "loaded": [], "evicted": [], "loaded_after": [], "ps_after": { "rc": 0, "text": "NAME ID SIZE PROCESSOR CONTEXT UNTIL" } } }, "mem_available_after_park_gib": 51.314, "psi_settle": { "paths": [ "/proc/pressure/memory", "/sys/fs/cgroup/user.slice/user-1000.slice/user@1000.service/memory.pressure" ], "waited_s": 0.0, "psi_start": 0.0, "psi_end": 0.0, "settled": true, "pct": 5.0, "max_s": 180.0, "rule": "the larger of system-wide and user@'s memory.pressure full avg10, below pct within max_s" }, "vram_baseline_mib": 15.0, "unit": { "name": "sb-eng-L-IQ3-t8-speed-a1", "argv": [ "systemd-run", "--user", "--quiet", "--unit=sb-eng-L-IQ3-t8-speed-a1", "--property=KillMode=control-group", "--property=TimeoutStopSec=30", "--property=MemoryMax=50G", "--property=MemorySwapMax=0", "--property=IPAddressDeny=any", "--property=IPAddressAllow=localhost", "--working-directory=/workshop/bench-archive/plans-2026-10-02/strata-bench/records/L-IQ3-t8/speed/a1", "--property=StandardOutput=append:/workshop/bench-archive/plans-2026-10-02/strata-bench/records/L-IQ3-t8/speed/a1/unit.log", "--property=StandardError=append:/workshop/bench-archive/plans-2026-10-02/strata-bench/records/L-IQ3-t8/speed/a1/unit.log", "--setenv=PYTHONDONTWRITEBYTECODE=1", "--setenv=PYTHONUNBUFFERED=1", "--", "bwrap", "--unshare-net", "--unshare-pid", "--die-with-parent", "--ro-bind", "/", "/", "--dev-bind", "/dev", "/dev", "--proc", "/proc", "--tmpfs", "/run", "--tmpfs", "/tmp", "--bind", "/workshop/bench-archive/plans-2026-10-02/strata-bench/records", "/workshop/bench-archive/plans-2026-10-02/strata-bench/records", "--ro-bind", "/run/nvidia-persistenced", "/run/nvidia-persistenced", "--", "/usr/bin/python3", "/workshop/bench-archive/plans-2026-10-02/strata-bench/harness/inner.py", "--spec", "/workshop/bench-archive/plans-2026-10-02/strata-bench/records/L-IQ3-t8/speed/a1/spec.json" ], "sandbox_notes": [ "/var/run is a link to /run (or absent): the /run tmpfs covers it", "/run/nvidia-persistenced bound read-only (the driver's persistence socket)" ], "ip_filter": "IPAddressDeny=any + IPAddressAllow=localhost CONFIGURED; a user manager does not enforce them (it logs 'unit configures an IP firewall, but not running as root'); the enforced block is the sandbox, proved by the probe" }, "egress_probe": { "ifaces_count": 1, "only_lo": true, "mode": "netns", "dns": "failed (gaierror)", "outside_connect": "failed (ENETUNREACH)", "loopback": "connected", "verdict": "enforced", "why": "only lo; DNS fails; the canary connect fails; loopback answers" }, "server_env_names": [ "DBUS_SESSION_BUS_ADDRESS", "DEBUGINFOD_URLS", "DESKTOP_SESSION", "DISPLAY", "GDMSESSION", "GNOME_DESKTOP_SESSION_ID", "GNOME_SETUP_DISPLAY", "GPG_AGENT_INFO", "GTK_MODULES", "HOME", "IM_CONFIG_ENTRY", "INVOCATION_ID", "LANG", "LLAMA_API_KEY", "LOGNAME", "MANAGERPID", "MANAGERPIDFDID", "MEMORY_PRESSURE_WATCH", "MEMORY_PRESSURE_WRITE", "ORT_DISABLE_TELEMETRY", "PATH", "PWD", "PYTHONDONTWRITEBYTECODE", "PYTHONUNBUFFERED", "QT_ACCESSIBILITY", "QT_IM_MODULE", "QT_IM_MODULES", "SHELL", "SHLVL", "SSH_AUTH_SOCK", "SYSTEMD_EXEC_PID", "USER", "USERNAME", "WAYLAND_DISPLAY", "XAUTHORITY", "XDG_CONFIG_DIRS", "XDG_CURRENT_DESKTOP", "XDG_DATA_DIRS", "XDG_MENU_PREFIX", "XDG_RUNTIME_DIR", "XDG_SESSION_CLASS", "XDG_SESSION_DESKTOP", "XDG_SESSION_EXTRA_DEVICE_ACCESS", "XDG_SESSION_TYPE", "XMODIFIERS" ], "server_env_dropped": [], "g0": { "verdict": "pass", "devices": { "cuda": 1, "vulkan": 0, "names": [ "CUDA0" ], "rule": "src/llama.cpp L306 at b11081: one `using device` line per device the model is offloaded to" }, "offload": { "verdict": "ok", "line": "offloaded 49/49 layers to GPU", "offloaded": 49, "of": 49, "sentence": "all 49 layers offloaded" }, "why": "one CUDA device, no Vulkan; all 49 layers offloaded", "health": { "ready": true, "seconds": 17.02, "health": { "status": "ok" } } }, "engine_identity_log": { "build": 1, "commit_short": "3cf0325", "line": "build 1 (3cf0325)", "ok": true, "expected": "3cf03257f219afbe7334045ff7c6a06ac68c627d", "why": "the start banner names the arm's commit" }, "threads": { "engine": "llama.cpp", "n_threads": 8, "n_threads_batch": 8, "logical_cpus": 24, "line": "0.00.224.969 I cmn common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 24", "requested": { "t": 8, "tb": 8 }, "source": "server.log (common_params_print_info system_info)", "matches_request": true } }, "unit_show_at_end": { "ActiveState": "active", "SubState": "running", "Result": "success", "NRestarts": "0", "ExecMainPID": "3201036", "ExecMainStatus": "0", "ControlGroup": "/user.slice/user-1000.slice/user@1000.service/app.slice/sb-eng-L-IQ3-t8-speed-a1.service", "MemoryCurrent": "38907265024", "MemoryPeak": "40719974400", "MemoryMax": "53687091200", "MemorySwapMax": "0", "IPAddressAllow": "127.0.0.0/8 ::1/128", "IPAddressDeny": "0.0.0.0/0 ::/0" }, "unit_memory_at_end": { "memory.current": "38907265024", "memory.peak": "40719974400", "memory.max": "53687091200", "memory.swap.current": "0", "memory.events": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "memory.stat": { "anon": 1091866624, "file": 37580566528, "kernel": 233385984, "shmem": 10485760, "file_mapped": 24919965696, "unevictable": 0 } }, "rows": 9, "rows_file": "rows.jsonl", "telemetry_file": "telemetry.csv", "uuid_scrubbed": { "server.log": 0, "unit.log": 0 }, "sampler_notes": [], "driver_ended": "2026-10-02T23:40:55Z", "marker": "RESULT", "arm": "L-IQ3-t8", "phase": "speed", "attempt": 1, "written": "2026-10-02T23:40:55Z" }