{"row_id": "L-Q2/speed/a2/run-001", "arm": "L-Q2", "engine": "llama.cpp", "engine_commit": "3cf03257f219afbe7334045ff7c6a06ac68c627d", "quant": "Q2_0", "file_sha256": {"1": "69820c02ec7d0b45ef2ebb19d6620299db749fe2aded7f39f93c6b88b199b720", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/data/strata-bench/build/llama.cpp-3cf03257/bin/llama-server", "-m", "/data/strata-bench/models/ISTA-ed59f920/Q2_0/Qwen3.8-Flash-Next-GSQ-RCO-Q2_0-00001-of-00002.gguf", "--host", "127.0.0.1", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "24", "-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"], "ctx": "4K", "repeat": 0, "run": 1, "slice_sha256": "9c676dc1959897277fda62a2d77d01fc06c932a2e04641c753a24bde3e7d4a16", "prompt_sha256": "b42c6a181c3e02478e7e033585c5962f35eb046b680c0ae837ad8706a3c0674b", "nonce": "Ticket 85822155.", "requests_since_start": 1, "status": "ok", "fail_kind": null, "error": null, "decode_tok_s": 29.968, "decode_rule": "client clock: (completion_tokens - 1) / (t_finish - t_first_token) (the kit's one timing rule, seatlib.RULES['engine_timing']); WITHHELD (null, the computed value kept as decode_tok_s_withheld) when the token count fails its check", "decode_tok_s_server": 29.963349535867714, "decode_tok_s_withheld": null, "decode_tok_s_server_withheld": null, "count_check": {"verdict": "ok", "completion_tokens": 256, "token_frames": 257, "rule": "llama.cpp: token_frames == completion_tokens + 1 (one frame per token, plus the finish frame)"}, "token_frames": 257, "content_frames": 256, "frames": 259, "thinking_seen": false, "threads": {"n_threads": 8, "n_threads_batch": 8, "requested": {"t": 8, "tb": 8}, "line": "0.01.269.939 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 24"}, "prompt_tok_s_server": 78.2289235190609, "prompt_tok_s_client": 78.2, "ttft_ms": 51628.82, "prompt_tokens": 4038, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 4038, "prompt_ms": 51617.737, "prompt_per_token_ms": 12.782995789995047, "prompt_per_second": 78.2289235190609, "predicted_n": 256, "predicted_ms": 8510.397, "predicted_per_token_ms": 33.37410588235294, "predicted_per_second": 29.963349535867714}, "short_completion": false, "cache_flag": null, "completion_sha256": "aa5035f3cca257cf4c6a9eadda775dfe9613d8c1b93919a6463d767fd29ead6e", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791007678.291, "t1": 1791007738.429, "seconds": 60.138, "host_samples": 60, "gpu_samples": 60}, "board": {"joules": 2187.91, "mean_w": 37.06, "max_w": 78.32, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": 1792.92, "label": "RAPL package-0, the CPU package: NOT WALL", "source": "/run/rapl/energy", "span_s": 59.007}, "data_reads_per_s": {"mean": 36271.6, "max": 58792.0}, "data_read_mb_per_s": {"mean": 150.01}, "temps": {"gpu_max_c": 44.0, "ssd_max_c": 47.85}, "vram": {"max_used_mib": 22839.0, "baseline_mib": 15.0, "growth_mib": 22824.0, "rule": "max memory.used in the window minus the pre-load baseline (read before the unit started)"}, "pinned": {"vmlck_max_bytes": 0, "vmpin_max_bytes": 0, "vmrss_max_bytes": 35444555776, "rule": "sums over the engine unit's processes' /proc/<pid>/status, every 5 s"}, "load": {"load1_mean": 1.49, "claude_procs_max": 1}, "memory": {"mem_available_gib_min": 51.849, "user_psi_full_avg10_max": 10.15, "psi_full_avg10_max": 9.33, "unit_psi_full_avg10_max": 38.42, "psi_rule": "psi = system-wide /proc/pressure/memory; user_psi = user@<uid>.service memory.pressure (systemd-oomd's); the stop reads the larger of the two; unit_psi = the engine unit's own memory.pressure (a figure, printed in every memory STOP)", "unit_max_events_delta": 1175}, "power_limit": {"min_w": 150.0, "max_w": 150.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}}
{"row_id": "L-Q2/speed/a2/run-002", "arm": "L-Q2", "engine": "llama.cpp", "engine_commit": "3cf03257f219afbe7334045ff7c6a06ac68c627d", "quant": "Q2_0", "file_sha256": {"1": "69820c02ec7d0b45ef2ebb19d6620299db749fe2aded7f39f93c6b88b199b720", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/data/strata-bench/build/llama.cpp-3cf03257/bin/llama-server", "-m", "/data/strata-bench/models/ISTA-ed59f920/Q2_0/Qwen3.8-Flash-Next-GSQ-RCO-Q2_0-00001-of-00002.gguf", "--host", "127.0.0.1", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "24", "-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"], "ctx": "4K", "repeat": 1, "run": 2, "slice_sha256": "9c676dc1959897277fda62a2d77d01fc06c932a2e04641c753a24bde3e7d4a16", "prompt_sha256": "9744fee0d5b25fee2aef36ad0bc7c5c4611f73d11f4e6bd9e5b84b17d93d319b", "nonce": "Ticket 42865306.", "requests_since_start": 2, "status": "ok", "fail_kind": null, "error": null, "decode_tok_s": 32.47, "decode_rule": "client clock: (completion_tokens - 1) / (t_finish - t_first_token) (the kit's one timing rule, seatlib.RULES['engine_timing']); WITHHELD (null, the computed value kept as decode_tok_s_withheld) when the token count fails its check", "decode_tok_s_server": 32.47013288943054, "decode_tok_s_withheld": null, "decode_tok_s_server_withheld": null, "count_check": {"verdict": "ok", "completion_tokens": 256, "token_frames": 257, "rule": "llama.cpp: token_frames == completion_tokens + 1 (one frame per token, plus the finish frame)"}, "token_frames": 257, "content_frames": 256, "frames": 259, "thinking_seen": false, "threads": {"n_threads": 8, "n_threads_batch": 8, "requested": {"t": 8, "tb": 8}, "line": "0.01.269.939 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 24"}, "prompt_tok_s_server": 425.9554329096459, "prompt_tok_s_client": 425.7, "ttft_ms": 9486.666, "prompt_tokens": 4038, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 4038, "prompt_ms": 9479.865, "prompt_per_token_ms": 2.347663447251114, "prompt_per_second": 425.9554329096459, "predicted_n": 256, "predicted_ms": 7853.371, "predicted_per_token_ms": 30.797533333333334, "predicted_per_second": 32.47013288943054}, "short_completion": false, "cache_flag": null, "completion_sha256": "4348eebd7d5fc17aa6b9b046f99697cf6e04084f60a9369897d7bfa656cdecfc", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791007738.43, "t1": 1791007755.77, "seconds": 17.34, "host_samples": 18, "gpu_samples": 18}, "board": {"joules": 1300.33, "mean_w": 76.4, "max_w": 85.95, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": 696.88, "label": "RAPL package-0, the CPU package: NOT WALL", "source": "/run/rapl/energy", "span_s": 17.003}, "data_reads_per_s": {"mean": 758.1, "max": 2980.7}, "data_read_mb_per_s": {"mean": 3.1}, "temps": {"gpu_max_c": 47.0, "ssd_max_c": 45.85}, "vram": {"max_used_mib": 22839.0, "baseline_mib": 15.0, "growth_mib": 22824.0, "rule": "max memory.used in the window minus the pre-load baseline (read before the unit started)"}, "pinned": {"vmlck_max_bytes": 0, "vmpin_max_bytes": 0, "vmrss_max_bytes": 19008151552, "rule": "sums over the engine unit's processes' /proc/<pid>/status, every 5 s"}, "load": {"load1_mean": 2.46, "claude_procs_max": 1}, "memory": {"mem_available_gib_min": 51.849, "user_psi_full_avg10_max": 2.7, "psi_full_avg10_max": 2.59, "unit_psi_full_avg10_max": 10.38, "psi_rule": "psi = system-wide /proc/pressure/memory; user_psi = user@<uid>.service memory.pressure (systemd-oomd's); the stop reads the larger of the two; unit_psi = the engine unit's own memory.pressure (a figure, printed in every memory STOP)", "unit_max_events_delta": 5}, "power_limit": {"min_w": 150.0, "max_w": 150.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}}
{"row_id": "L-Q2/speed/a2/run-003", "arm": "L-Q2", "engine": "llama.cpp", "engine_commit": "3cf03257f219afbe7334045ff7c6a06ac68c627d", "quant": "Q2_0", "file_sha256": {"1": "69820c02ec7d0b45ef2ebb19d6620299db749fe2aded7f39f93c6b88b199b720", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/data/strata-bench/build/llama.cpp-3cf03257/bin/llama-server", "-m", "/data/strata-bench/models/ISTA-ed59f920/Q2_0/Qwen3.8-Flash-Next-GSQ-RCO-Q2_0-00001-of-00002.gguf", "--host", "127.0.0.1", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "24", "-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"], "ctx": "4K", "repeat": 2, "run": 3, "slice_sha256": "9c676dc1959897277fda62a2d77d01fc06c932a2e04641c753a24bde3e7d4a16", "prompt_sha256": "214ec62ff2b1550ea59fc2fa7cc264e79ceed8841f5fd91e1709a8b73be2fa8f", "nonce": "Ticket 90569497.", "requests_since_start": 3, "status": "ok", "fail_kind": null, "error": null, "decode_tok_s": 33.394, "decode_rule": "client clock: (completion_tokens - 1) / (t_finish - t_first_token) (the kit's one timing rule, seatlib.RULES['engine_timing']); WITHHELD (null, the computed value kept as decode_tok_s_withheld) when the token count fails its check", "decode_tok_s_server": 33.39505525310112, "decode_tok_s_withheld": null, "decode_tok_s_server_withheld": null, "count_check": {"verdict": "ok", "completion_tokens": 256, "token_frames": 257, "rule": "llama.cpp: token_frames == completion_tokens + 1 (one frame per token, plus the finish frame)"}, "token_frames": 257, "content_frames": 256, "frames": 259, "thinking_seen": false, "threads": {"n_threads": 8, "n_threads_batch": 8, "requested": {"t": 8, "tb": 8}, "line": "0.01.269.939 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 24"}, "prompt_tok_s_server": 442.02533424745076, "prompt_tok_s_client": 441.7, "ttft_ms": 9142.303, "prompt_tokens": 4038, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 4038, "prompt_ms": 9135.223, "prompt_per_token_ms": 2.26231376919267, "prompt_per_second": 442.02533424745076, "predicted_n": 256, "predicted_ms": 7635.861, "predicted_per_token_ms": 29.94455294117647, "predicted_per_second": 33.39505525310112}, "short_completion": false, "cache_flag": null, "completion_sha256": "3a2fc38df8f888aa0039c54b159ad39c6b063bf500b7c347bb89772e8df8ecf5", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791007755.771, "t1": 1791007772.55, "seconds": 16.778, "host_samples": 17, "gpu_samples": 16}, "board": {"joules": 1179.08, "mean_w": 78.53, "max_w": 84.0, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": 671.02, "label": "RAPL package-0, the CPU package: NOT WALL", "source": "/run/rapl/energy", "span_s": 16.002}, "data_reads_per_s": {"mean": 245.7, "max": 968.8}, "data_read_mb_per_s": {"mean": 1.01}, "temps": {"gpu_max_c": 48.0, "ssd_max_c": 45.85}, "vram": {"max_used_mib": 22839.0, "baseline_mib": 15.0, "growth_mib": 22824.0, "rule": "max memory.used in the window minus the pre-load baseline (read before the unit started)"}, "pinned": {"vmlck_max_bytes": 0, "vmpin_max_bytes": 0, "vmrss_max_bytes": 19026165760, "rule": "sums over the engine unit's processes' /proc/<pid>/status, every 5 s"}, "load": {"load1_mean": 2.36, "claude_procs_max": 1}, "memory": {"mem_available_gib_min": 51.797, "user_psi_full_avg10_max": 0.54, "psi_full_avg10_max": 0.53, "unit_psi_full_avg10_max": 3.44, "psi_rule": "psi = system-wide /proc/pressure/memory; user_psi = user@<uid>.service memory.pressure (systemd-oomd's); the stop reads the larger of the two; unit_psi = the engine unit's own memory.pressure (a figure, printed in every memory STOP)", "unit_max_events_delta": 1}, "power_limit": {"min_w": 150.0, "max_w": 150.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}}
{"row_id": "L-Q2/speed/a2/run-004", "arm": "L-Q2", "engine": "llama.cpp", "engine_commit": "3cf03257f219afbe7334045ff7c6a06ac68c627d", "quant": "Q2_0", "file_sha256": {"1": "69820c02ec7d0b45ef2ebb19d6620299db749fe2aded7f39f93c6b88b199b720", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/data/strata-bench/build/llama.cpp-3cf03257/bin/llama-server", "-m", "/data/strata-bench/models/ISTA-ed59f920/Q2_0/Qwen3.8-Flash-Next-GSQ-RCO-Q2_0-00001-of-00002.gguf", "--host", "127.0.0.1", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "24", "-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"], "ctx": "32K", "repeat": 0, "run": 4, "slice_sha256": "b72b07858a845e76331ec38d46abae3c09d3e1cb6e4d968e887fa8b132c64977", "prompt_sha256": "507b7bc9961d968eaf07467339f314a3c76a036b2542852f07b9a638adcd5ad5", "nonce": "Ticket 10426037.", "requests_since_start": 4, "status": "ok", "fail_kind": null, "error": null, "decode_tok_s": 26.561, "decode_rule": "client clock: (completion_tokens - 1) / (t_finish - t_first_token) (the kit's one timing rule, seatlib.RULES['engine_timing']); WITHHELD (null, the computed value kept as decode_tok_s_withheld) when the token count fails its check", "decode_tok_s_server": 26.56152607737716, "decode_tok_s_withheld": null, "decode_tok_s_server_withheld": null, "count_check": {"verdict": "ok", "completion_tokens": 256, "token_frames": 257, "rule": "llama.cpp: token_frames == completion_tokens + 1 (one frame per token, plus the finish frame)"}, "token_frames": 257, "content_frames": 256, "frames": 259, "thinking_seen": false, "threads": {"n_threads": 8, "n_threads_batch": 8, "requested": {"t": 8, "tb": 8}, "line": "0.01.269.939 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 24"}, "prompt_tok_s_server": 357.18524647220204, "prompt_tok_s_client": 357.1, "ttft_ms": 89725.089, "prompt_tokens": 32038, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 32038, "prompt_ms": 89695.754, "prompt_per_token_ms": 2.7996677070978215, "prompt_per_second": 357.18524647220204, "predicted_n": 256, "predicted_ms": 9600.352, "predicted_per_token_ms": 37.64843921568628, "predicted_per_second": 26.56152607737716}, "short_completion": false, "cache_flag": null, "completion_sha256": "d853325f1f704f5ffdffea85214b3984cf71de74a5156a69b9344d07dbd9be13", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791007772.551, "t1": 1791007871.876, "seconds": 99.326, "host_samples": 99, "gpu_samples": 100}, "board": {"joules": 7354.28, "mean_w": 74.21, "max_w": 92.56, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": 2988.27, "label": "RAPL package-0, the CPU package: NOT WALL", "source": "/run/rapl/energy", "span_s": 98.017}, "data_reads_per_s": {"mean": 3906.2, "max": 17228.5}, "data_read_mb_per_s": {"mean": 16.01}, "temps": {"gpu_max_c": 54.0, "ssd_max_c": 47.85}, "vram": {"max_used_mib": 22841.0, "baseline_mib": 15.0, "growth_mib": 22826.0, "rule": "max memory.used in the window minus the pre-load baseline (read before the unit started)"}, "pinned": {"vmlck_max_bytes": 0, "vmpin_max_bytes": 0, "vmrss_max_bytes": 20634652672, "rule": "sums over the engine unit's processes' /proc/<pid>/status, every 5 s"}, "load": {"load1_mean": 1.72, "claude_procs_max": 1}, "memory": {"mem_available_gib_min": 51.643, "user_psi_full_avg10_max": 1.81, "psi_full_avg10_max": 1.76, "unit_psi_full_avg10_max": 12.08, "psi_rule": "psi = system-wide /proc/pressure/memory; user_psi = user@<uid>.service memory.pressure (systemd-oomd's); the stop reads the larger of the two; unit_psi = the engine unit's own memory.pressure (a figure, printed in every memory STOP)", "unit_max_events_delta": 193}, "power_limit": {"min_w": 150.0, "max_w": 150.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}}
{"row_id": "L-Q2/speed/a2/run-005", "arm": "L-Q2", "engine": "llama.cpp", "engine_commit": "3cf03257f219afbe7334045ff7c6a06ac68c627d", "quant": "Q2_0", "file_sha256": {"1": "69820c02ec7d0b45ef2ebb19d6620299db749fe2aded7f39f93c6b88b199b720", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/data/strata-bench/build/llama.cpp-3cf03257/bin/llama-server", "-m", "/data/strata-bench/models/ISTA-ed59f920/Q2_0/Qwen3.8-Flash-Next-GSQ-RCO-Q2_0-00001-of-00002.gguf", "--host", "127.0.0.1", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "24", "-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"], "ctx": "32K", "repeat": 1, "run": 5, "slice_sha256": "b72b07858a845e76331ec38d46abae3c09d3e1cb6e4d968e887fa8b132c64977", "prompt_sha256": "c8a1ccb2149bdf9595fe8d97b7183fd9efd61c32259e907606ddbac7710955b2", "nonce": "Ticket 97542637.", "requests_since_start": 5, "status": "ok", "fail_kind": null, "error": null, "decode_tok_s": 27.604, "decode_rule": "client clock: (completion_tokens - 1) / (t_finish - t_first_token) (the kit's one timing rule, seatlib.RULES['engine_timing']); WITHHELD (null, the computed value kept as decode_tok_s_withheld) when the token count fails its check", "decode_tok_s_server": 27.604674218295163, "decode_tok_s_withheld": null, "decode_tok_s_server_withheld": null, "count_check": {"verdict": "ok", "completion_tokens": 256, "token_frames": 257, "rule": "llama.cpp: token_frames == completion_tokens + 1 (one frame per token, plus the finish frame)"}, "token_frames": 257, "content_frames": 256, "frames": 259, "thinking_seen": false, "threads": {"n_threads": 8, "n_threads_batch": 8, "requested": {"t": 8, "tb": 8}, "line": "0.01.269.939 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 24"}, "prompt_tok_s_server": 437.1740153579103, "prompt_tok_s_client": 437.0, "ttft_ms": 73314.172, "prompt_tokens": 32038, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 32038, "prompt_ms": 73284.319, "prompt_per_token_ms": 2.287418659092328, "prompt_per_second": 437.1740153579103, "predicted_n": 256, "predicted_ms": 9237.566, "predicted_per_token_ms": 36.225749019607846, "predicted_per_second": 27.604674218295163}, "short_completion": false, "cache_flag": null, "completion_sha256": "5205f9b0dac7c6fecdebbc23f508f604d2bfa59468de68d0ab9109ac900163e1", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791007871.878, "t1": 1791007954.43, "seconds": 82.552, "host_samples": 82, "gpu_samples": 82}, "board": {"joules": 6914.29, "mean_w": 85.26, "max_w": 97.28, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": 2727.49, "label": "RAPL package-0, the CPU package: NOT WALL", "source": "/run/rapl/energy", "span_s": 81.012}, "data_reads_per_s": {"mean": 55.4, "max": 998.9}, "data_read_mb_per_s": {"mean": 0.23}, "temps": {"gpu_max_c": 56.0, "ssd_max_c": 47.85}, "vram": {"max_used_mib": 22841.0, "baseline_mib": 15.0, "growth_mib": 22826.0, "rule": "max memory.used in the window minus the pre-load baseline (read before the unit started)"}, "pinned": {"vmlck_max_bytes": 0, "vmpin_max_bytes": 0, "vmrss_max_bytes": 20658319360, "rule": "sums over the engine unit's processes' /proc/<pid>/status, every 5 s"}, "load": {"load1_mean": 1.53, "claude_procs_max": 1}, "memory": {"mem_available_gib_min": 51.609, "user_psi_full_avg10_max": 0.1, "psi_full_avg10_max": 0.07, "unit_psi_full_avg10_max": 1.39, "psi_rule": "psi = system-wide /proc/pressure/memory; user_psi = user@<uid>.service memory.pressure (systemd-oomd's); the stop reads the larger of the two; unit_psi = the engine unit's own memory.pressure (a figure, printed in every memory STOP)", "unit_max_events_delta": 2}, "power_limit": {"min_w": 150.0, "max_w": 150.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}}
{"row_id": "L-Q2/speed/a2/run-006", "arm": "L-Q2", "engine": "llama.cpp", "engine_commit": "3cf03257f219afbe7334045ff7c6a06ac68c627d", "quant": "Q2_0", "file_sha256": {"1": "69820c02ec7d0b45ef2ebb19d6620299db749fe2aded7f39f93c6b88b199b720", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/data/strata-bench/build/llama.cpp-3cf03257/bin/llama-server", "-m", "/data/strata-bench/models/ISTA-ed59f920/Q2_0/Qwen3.8-Flash-Next-GSQ-RCO-Q2_0-00001-of-00002.gguf", "--host", "127.0.0.1", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "24", "-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"], "ctx": "32K", "repeat": 2, "run": 6, "slice_sha256": "b72b07858a845e76331ec38d46abae3c09d3e1cb6e4d968e887fa8b132c64977", "prompt_sha256": "d6c964051c51baa9ef5cb009be0a921422c2b28350d2dbea610a6038552bdc18", "nonce": "Ticket 24502846.", "requests_since_start": 6, "status": "ok", "fail_kind": null, "error": null, "decode_tok_s": 26.103, "decode_rule": "client clock: (completion_tokens - 1) / (t_finish - t_first_token) (the kit's one timing rule, seatlib.RULES['engine_timing']); WITHHELD (null, the computed value kept as decode_tok_s_withheld) when the token count fails its check", "decode_tok_s_server": 26.10413050263555, "decode_tok_s_withheld": null, "decode_tok_s_server_withheld": null, "count_check": {"verdict": "ok", "completion_tokens": 256, "token_frames": 257, "rule": "llama.cpp: token_frames == completion_tokens + 1 (one frame per token, plus the finish frame)"}, "token_frames": 257, "content_frames": 256, "frames": 259, "thinking_seen": false, "threads": {"n_threads": 8, "n_threads_batch": 8, "requested": {"t": 8, "tb": 8}, "line": "0.01.269.939 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 24"}, "prompt_tok_s_server": 437.66355691213624, "prompt_tok_s_client": 437.5, "ttft_ms": 73231.556, "prompt_tokens": 32038, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 32038, "prompt_ms": 73202.348, "prompt_per_token_ms": 2.284860103626943, "prompt_per_second": 437.66355691213624, "predicted_n": 256, "predicted_ms": 9768.569, "predicted_per_token_ms": 38.308113725490195, "predicted_per_second": 26.10413050263555}, "short_completion": false, "cache_flag": null, "completion_sha256": "116012403271ad223a5a4b6b038002eebbfd5c2e4b8f97798342125a6a97d620", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791007954.431, "t1": 1791008037.432, "seconds": 83.001, "host_samples": 83, "gpu_samples": 83}, "board": {"joules": 6968.58, "mean_w": 84.89, "max_w": 91.73, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": 2738.9, "label": "RAPL package-0, the CPU package: NOT WALL", "source": "/run/rapl/energy", "span_s": 82.014}, "data_reads_per_s": {"mean": 39.5, "max": 439.9}, "data_read_mb_per_s": {"mean": 0.16}, "temps": {"gpu_max_c": 57.0, "ssd_max_c": 47.85}, "vram": {"max_used_mib": 22841.0, "baseline_mib": 15.0, "growth_mib": 22826.0, "rule": "max memory.used in the window minus the pre-load baseline (read before the unit started)"}, "pinned": {"vmlck_max_bytes": 0, "vmpin_max_bytes": 0, "vmrss_max_bytes": 20679655424, "rule": "sums over the engine unit's processes' /proc/<pid>/status, every 5 s"}, "load": {"load1_mean": 1.25, "claude_procs_max": 1}, "memory": {"mem_available_gib_min": 51.585, "user_psi_full_avg10_max": 0.0, "psi_full_avg10_max": 0.0, "unit_psi_full_avg10_max": 0.0, "psi_rule": "psi = system-wide /proc/pressure/memory; user_psi = user@<uid>.service memory.pressure (systemd-oomd's); the stop reads the larger of the two; unit_psi = the engine unit's own memory.pressure (a figure, printed in every memory STOP)", "unit_max_events_delta": 1}, "power_limit": {"min_w": 150.0, "max_w": 150.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}}
{"row_id": "L-Q2/speed/a2/run-007", "arm": "L-Q2", "engine": "llama.cpp", "engine_commit": "3cf03257f219afbe7334045ff7c6a06ac68c627d", "quant": "Q2_0", "file_sha256": {"1": "69820c02ec7d0b45ef2ebb19d6620299db749fe2aded7f39f93c6b88b199b720", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/data/strata-bench/build/llama.cpp-3cf03257/bin/llama-server", "-m", "/data/strata-bench/models/ISTA-ed59f920/Q2_0/Qwen3.8-Flash-Next-GSQ-RCO-Q2_0-00001-of-00002.gguf", "--host", "127.0.0.1", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "24", "-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"], "ctx": "128K", "repeat": 0, "run": 7, "slice_sha256": "286827657f5df11459f4ee550246f7e0e06ce7af3238fe74a53f0aa4c12676ac", "prompt_sha256": "2491e963527b8a39f7fff1aa2d04dab5eb4dd3ca12c7417c451f6216c665972c", "nonce": "Ticket 14239051.", "requests_since_start": 7, "status": "ok", "fail_kind": null, "error": null, "decode_tok_s": 17.898, "decode_rule": "client clock: (completion_tokens - 1) / (t_finish - t_first_token) (the kit's one timing rule, seatlib.RULES['engine_timing']); WITHHELD (null, the computed value kept as decode_tok_s_withheld) when the token count fails its check", "decode_tok_s_server": 17.90090354635151, "decode_tok_s_withheld": null, "decode_tok_s_server_withheld": null, "count_check": {"verdict": "ok", "completion_tokens": 256, "token_frames": 257, "rule": "llama.cpp: token_frames == completion_tokens + 1 (one frame per token, plus the finish frame)"}, "token_frames": 257, "content_frames": 256, "frames": 259, "thinking_seen": false, "threads": {"n_threads": 8, "n_threads_batch": 8, "requested": {"t": 8, "tb": 8}, "line": "0.01.269.939 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 24"}, "prompt_tok_s_server": 373.6080748786251, "prompt_tok_s_client": 373.5, "ttft_ms": 342805.275, "prompt_tokens": 128037, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 128037, "prompt_ms": 342704.049, "prompt_per_token_ms": 2.676601677640057, "prompt_per_second": 373.6080748786251, "predicted_n": 256, "predicted_ms": 14245.091, "predicted_per_token_ms": 55.86310196078431, "predicted_per_second": 17.90090354635151}, "short_completion": false, "cache_flag": null, "completion_sha256": "2bf1b64c18e033a67c9b29bca9e48d7582b5c0b60c0aecfcce12aa91a00172b5", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791008037.435, "t1": 1791008394.487, "seconds": 357.052, "host_samples": 357, "gpu_samples": 357}, "board": {"joules": 28766.52, "mean_w": 80.71, "max_w": 121.8, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": 12742.21, "label": "RAPL package-0, the CPU package: NOT WALL", "source": "/run/rapl/energy", "span_s": 356.056}, "data_reads_per_s": {"mean": 1890.3, "max": 7654.5}, "data_read_mb_per_s": {"mean": 8.26}, "temps": {"gpu_max_c": 58.0, "ssd_max_c": 49.85}, "vram": {"max_used_mib": 22841.0, "baseline_mib": 15.0, "growth_mib": 22826.0, "rule": "max memory.used in the window minus the pre-load baseline (read before the unit started)"}, "pinned": {"vmlck_max_bytes": 0, "vmpin_max_bytes": 0, "vmrss_max_bytes": 23410728960, "rule": "sums over the engine unit's processes' /proc/<pid>/status, every 5 s"}, "load": {"load1_mean": 1.51, "claude_procs_max": 1}, "memory": {"mem_available_gib_min": 51.61, "user_psi_full_avg10_max": 0.0, "psi_full_avg10_max": 0.0, "unit_psi_full_avg10_max": 1.11, "psi_rule": "psi = system-wide /proc/pressure/memory; user_psi = user@<uid>.service memory.pressure (systemd-oomd's); the stop reads the larger of the two; unit_psi = the engine unit's own memory.pressure (a figure, printed in every memory STOP)", "unit_max_events_delta": 325}, "power_limit": {"min_w": 150.0, "max_w": 150.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}}
{"row_id": "L-Q2/speed/a2/run-008", "arm": "L-Q2", "engine": "llama.cpp", "engine_commit": "3cf03257f219afbe7334045ff7c6a06ac68c627d", "quant": "Q2_0", "file_sha256": {"1": "69820c02ec7d0b45ef2ebb19d6620299db749fe2aded7f39f93c6b88b199b720", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/data/strata-bench/build/llama.cpp-3cf03257/bin/llama-server", "-m", "/data/strata-bench/models/ISTA-ed59f920/Q2_0/Qwen3.8-Flash-Next-GSQ-RCO-Q2_0-00001-of-00002.gguf", "--host", "127.0.0.1", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "24", "-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"], "ctx": "128K", "repeat": 1, "run": 8, "slice_sha256": "286827657f5df11459f4ee550246f7e0e06ce7af3238fe74a53f0aa4c12676ac", "prompt_sha256": "94613594c1c300728a39aa8436592de0bdf88c05edeceb88cd9c6794de1fa097", "nonce": "Ticket 81078402.", "requests_since_start": 8, "status": "ok", "fail_kind": null, "error": null, "decode_tok_s": 17.484, "decode_rule": "client clock: (completion_tokens - 1) / (t_finish - t_first_token) (the kit's one timing rule, seatlib.RULES['engine_timing']); WITHHELD (null, the computed value kept as decode_tok_s_withheld) when the token count fails its check", "decode_tok_s_server": 17.486412543147722, "decode_tok_s_withheld": null, "decode_tok_s_server_withheld": null, "count_check": {"verdict": "ok", "completion_tokens": 256, "token_frames": 257, "rule": "llama.cpp: token_frames == completion_tokens + 1 (one frame per token, plus the finish frame)"}, "token_frames": 257, "content_frames": 256, "frames": 259, "thinking_seen": false, "threads": {"n_threads": 8, "n_threads_batch": 8, "requested": {"t": 8, "tb": 8}, "line": "0.01.269.939 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 24"}, "prompt_tok_s_server": 399.18934604144545, "prompt_tok_s_client": 399.1, "ttft_ms": 320842.939, "prompt_tokens": 128037, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 128037, "prompt_ms": 320742.528, "prompt_per_token_ms": 2.505076876215469, "prompt_per_second": 399.18934604144545, "predicted_n": 256, "predicted_ms": 14582.751, "predicted_per_token_ms": 57.18725882352941, "predicted_per_second": 17.486412543147722}, "short_completion": false, "cache_flag": null, "completion_sha256": "2d1ae904834b4061df2210a592cc00a4dbce4fe91870d8ca0be6934bf78b7095", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791008394.491, "t1": 1791008729.919, "seconds": 335.428, "host_samples": 336, "gpu_samples": 335}, "board": {"joules": 28308.93, "mean_w": 84.66, "max_w": 120.2, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": 15950.04, "label": "RAPL package-0, the CPU package: NOT WALL", "source": "/run/rapl/energy", "span_s": 335.053}, "data_reads_per_s": {"mean": 5.2, "max": 197.0}, "data_read_mb_per_s": {"mean": 0.02}, "temps": {"gpu_max_c": 61.0, "ssd_max_c": 50.85}, "vram": {"max_used_mib": 22841.0, "baseline_mib": 15.0, "growth_mib": 22826.0, "rule": "max memory.used in the window minus the pre-load baseline (read before the unit started)"}, "pinned": {"vmlck_max_bytes": 0, "vmpin_max_bytes": 0, "vmrss_max_bytes": 23428472832, "rule": "sums over the engine unit's processes' /proc/<pid>/status, every 5 s"}, "load": {"load1_mean": 1.95, "claude_procs_max": 1}, "memory": {"mem_available_gib_min": 50.238, "user_psi_full_avg10_max": 0.0, "psi_full_avg10_max": 0.0, "unit_psi_full_avg10_max": 0.0, "psi_rule": "psi = system-wide /proc/pressure/memory; user_psi = user@<uid>.service memory.pressure (systemd-oomd's); the stop reads the larger of the two; unit_psi = the engine unit's own memory.pressure (a figure, printed in every memory STOP)", "unit_max_events_delta": 0}, "power_limit": {"min_w": 137.35, "max_w": 150.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}}
{"row_id": "L-Q2/speed/a2/run-009", "arm": "L-Q2", "engine": "llama.cpp", "engine_commit": "3cf03257f219afbe7334045ff7c6a06ac68c627d", "quant": "Q2_0", "file_sha256": {"1": "69820c02ec7d0b45ef2ebb19d6620299db749fe2aded7f39f93c6b88b199b720", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/data/strata-bench/build/llama.cpp-3cf03257/bin/llama-server", "-m", "/data/strata-bench/models/ISTA-ed59f920/Q2_0/Qwen3.8-Flash-Next-GSQ-RCO-Q2_0-00001-of-00002.gguf", "--host", "127.0.0.1", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "24", "-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"], "ctx": "128K", "repeat": 2, "run": 9, "slice_sha256": "286827657f5df11459f4ee550246f7e0e06ce7af3238fe74a53f0aa4c12676ac", "prompt_sha256": "38d3c9a3177694d5b9006f1c813fddc9e6c5745048a59c545c41d199f780b364", "nonce": "Ticket 81461625.", "requests_since_start": 9, "status": "ok", "fail_kind": null, "error": null, "decode_tok_s": 16.985, "decode_rule": "client clock: (completion_tokens - 1) / (t_finish - t_first_token) (the kit's one timing rule, seatlib.RULES['engine_timing']); WITHHELD (null, the computed value kept as decode_tok_s_withheld) when the token count fails its check", "decode_tok_s_server": 16.986319218501418, "decode_tok_s_withheld": null, "decode_tok_s_server_withheld": null, "count_check": {"verdict": "ok", "completion_tokens": 256, "token_frames": 257, "rule": "llama.cpp: token_frames == completion_tokens + 1 (one frame per token, plus the finish frame)"}, "token_frames": 257, "content_frames": 256, "frames": 259, "thinking_seen": false, "threads": {"n_threads": 8, "n_threads_batch": 8, "requested": {"t": 8, "tb": 8}, "line": "0.01.269.939 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 24"}, "prompt_tok_s_server": 397.95661955506, "prompt_tok_s_client": 397.8, "ttft_ms": 321839.175, "prompt_tokens": 128037, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 128037, "prompt_ms": 321736.073, "prompt_per_token_ms": 2.512836703452908, "prompt_per_second": 397.95661955506, "predicted_n": 256, "predicted_ms": 15012.081, "predicted_per_token_ms": 58.87090588235294, "predicted_per_second": 16.986319218501418}, "short_completion": false, "cache_flag": null, "completion_sha256": "90eac92fab80835de6f3510bc75b51d19d5f91cf57c09eeee453b6e6f36aa766", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791008729.92, "t1": 1791009066.773, "seconds": 336.853, "host_samples": 337, "gpu_samples": 336}, "board": {"joules": 28374.31, "mean_w": 84.6, "max_w": 117.23, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": 15658.1, "label": "RAPL package-0, the CPU package: NOT WALL", "source": "/run/rapl/energy", "span_s": 336.054}, "data_reads_per_s": {"mean": 4.8, "max": 362.9}, "data_read_mb_per_s": {"mean": 0.04}, "temps": {"gpu_max_c": 62.0, "ssd_max_c": 51.85}, "vram": {"max_used_mib": 22841.0, "baseline_mib": 15.0, "growth_mib": 22826.0, "rule": "max memory.used in the window minus the pre-load baseline (read before the unit started)"}, "pinned": {"vmlck_max_bytes": 0, "vmpin_max_bytes": 0, "vmrss_max_bytes": 23436500992, "rule": "sums over the engine unit's processes' /proc/<pid>/status, every 5 s"}, "load": {"load1_mean": 2.3, "claude_procs_max": 1}, "memory": {"mem_available_gib_min": 50.562, "user_psi_full_avg10_max": 0.0, "psi_full_avg10_max": 0.0, "unit_psi_full_avg10_max": 0.0, "psi_rule": "psi = system-wide /proc/pressure/memory; user_psi = user@<uid>.service memory.pressure (systemd-oomd's); the stop reads the larger of the two; unit_psi = the engine unit's own memory.pressure (a figure, printed in every memory STOP)", "unit_max_events_delta": 0}, "power_limit": {"min_w": 135.0, "max_w": 150.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}}