{"row_id": "L-Q2/speed/a1/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": 30.551, "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": 30.54957609168611, "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": "0.00.332.788 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 24"}, "prompt_tok_s_server": 353.09172112714435, "prompt_tok_s_client": 352.8, "ttft_ms": 11445.51, "prompt_tokens": 4038, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 4038, "prompt_ms": 11436.122, "prompt_per_token_ms": 2.8321253095591876, "prompt_per_second": 353.09172112714435, "predicted_n": 256, "predicted_ms": 8347.088, "predicted_per_token_ms": 32.733678431372546, "predicted_per_second": 30.54957609168611}, "short_completion": false, "cache_flag": null, "completion_sha256": "165a91c839b0b4ad3880bc16ec7c2f9e53dccc290e1723119829b414d03f6d6b", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1790986603.924, "t1": 1790986623.717, "seconds": 19.792, "host_samples": 20, "gpu_samples": 20}, "board": {"joules": 1308.56, "mean_w": 68.86, "max_w": 79.86, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": 725.45, "label": "RAPL package-0, the CPU package: NOT WALL", "source": "/run/rapl/energy", "span_s": 19.003}, "data_reads_per_s": {"mean": 2573.8, "max": 7167.4}, "data_read_mb_per_s": {"mean": 11.22}, "temps": {"gpu_max_c": 44.0, "ssd_max_c": 41.85}, "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": 20406407168, "rule": "sums over the engine unit's processes' /proc/<pid>/status, every 5 s"}, "load": {"load1_mean": 0.99, "claude_procs_max": 1}, "memory": {"mem_available_gib_min": 51.848, "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": 150.0, "max_w": 150.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}}
{"row_id": "L-Q2/speed/a1/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": 29.788, "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": 29.78795183650316, "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": "0.00.332.788 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 24"}, "prompt_tok_s_server": 455.68754595103337, "prompt_tok_s_client": 455.3, "ttft_ms": 8868.161, "prompt_tokens": 4038, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 4038, "prompt_ms": 8861.335, "prompt_per_token_ms": 2.19448613174839, "prompt_per_second": 455.68754595103337, "predicted_n": 256, "predicted_ms": 8560.508, "predicted_per_token_ms": 33.570619607843135, "predicted_per_second": 29.78795183650316}, "short_completion": false, "cache_flag": null, "completion_sha256": "4348eebd7d5fc17aa6b9b046f99697cf6e04084f60a9369897d7bfa656cdecfc", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1790986623.718, "t1": 1790986641.146, "seconds": 17.429, "host_samples": 17, "gpu_samples": 17}, "board": {"joules": 1227.93, "mean_w": 76.73, "max_w": 87.59, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": 668.83, "label": "RAPL package-0, the CPU package: NOT WALL", "source": "/run/rapl/energy", "span_s": 16.002}, "data_reads_per_s": {"mean": 25.0, "max": 264.0}, "data_read_mb_per_s": {"mean": 0.1}, "temps": {"gpu_max_c": 47.0, "ssd_max_c": 41.85}, "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": 20414668800, "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": 51.752, "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": 150.0, "max_w": 150.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}}
{"row_id": "L-Q2/speed/a1/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": 31.501, "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": 31.50212120459664, "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": "0.00.332.788 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 24"}, "prompt_tok_s_server": 438.01535006245354, "prompt_tok_s_client": 437.7, "ttft_ms": 9225.493, "prompt_tokens": 4038, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 4038, "prompt_ms": 9218.855, "prompt_per_token_ms": 2.2830250123823674, "prompt_per_second": 438.01535006245354, "predicted_n": 256, "predicted_ms": 8094.693, "predicted_per_token_ms": 31.74389411764706, "predicted_per_second": 31.50212120459664}, "short_completion": false, "cache_flag": null, "completion_sha256": "3a2fc38df8f888aa0039c54b159ad39c6b063bf500b7c347bb89772e8df8ecf5", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1790986641.148, "t1": 1790986658.468, "seconds": 17.321, "host_samples": 18, "gpu_samples": 18}, "board": {"joules": 1298.99, "mean_w": 76.39, "max_w": 83.95, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": 844.46, "label": "RAPL package-0, the CPU package: NOT WALL", "source": "/run/rapl/energy", "span_s": 17.003}, "data_reads_per_s": {"mean": 11.8, "max": 131.0}, "data_read_mb_per_s": {"mean": 0.05}, "temps": {"gpu_max_c": 51.0, "ssd_max_c": 41.85}, "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": 20416331776, "rule": "sums over the engine unit's processes' /proc/<pid>/status, every 5 s"}, "load": {"load1_mean": 2.11, "claude_procs_max": 1}, "memory": {"mem_available_gib_min": 51.715, "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": 150.0, "max_w": 150.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}}
{"row_id": "L-Q2/speed/a1/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": 24.842, "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": 24.843018911967352, "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": "0.00.332.788 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 24"}, "prompt_tok_s_server": 382.67638642551134, "prompt_tok_s_client": 382.5, "ttft_ms": 83750.082, "prompt_tokens": 32038, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 32038, "prompt_ms": 83720.87, "prompt_per_token_ms": 2.613174043323553, "prompt_per_second": 382.67638642551134, "predicted_n": 256, "predicted_ms": 10264.453, "predicted_per_token_ms": 40.252756862745095, "predicted_per_second": 24.843018911967352}, "short_completion": false, "cache_flag": null, "completion_sha256": "5d8c92c0f1c9d55d0c34b28314cd638b8e7d6e3de01f7215f5a1c27ad4dc14a2", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1790986658.469, "t1": 1790986752.484, "seconds": 94.015, "host_samples": 94, "gpu_samples": 94}, "board": {"joules": 7375.18, "mean_w": 79.22, "max_w": 90.66, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": 3514.62, "label": "RAPL package-0, the CPU package: NOT WALL", "source": "/run/rapl/energy", "span_s": 93.013}, "data_reads_per_s": {"mean": 1850.0, "max": 6457.5}, "data_read_mb_per_s": {"mean": 7.59}, "temps": {"gpu_max_c": 55.0, "ssd_max_c": 43.85}, "vram": {"max_used_mib": 22843.0, "baseline_mib": 15.0, "growth_mib": 22828.0, "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": 21149396992, "rule": "sums over the engine unit's processes' /proc/<pid>/status, every 5 s"}, "load": {"load1_mean": 2.24, "claude_procs_max": 1}, "memory": {"mem_available_gib_min": 51.629, "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": 150.0, "max_w": 150.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}}
{"row_id": "L-Q2/speed/a1/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": 26.257, "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": 26.25817865035021, "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": "0.00.332.788 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 24"}, "prompt_tok_s_server": 423.11232381534427, "prompt_tok_s_client": 422.9, "ttft_ms": 75751.519, "prompt_tokens": 32038, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 32038, "prompt_ms": 75719.846, "prompt_per_token_ms": 2.363438604157563, "prompt_per_second": 423.11232381534427, "predicted_n": 256, "predicted_ms": 9711.26, "predicted_per_token_ms": 38.08337254901961, "predicted_per_second": 26.25817865035021}, "short_completion": false, "cache_flag": null, "completion_sha256": "4393af15255fbe7069fa372044fa87f06d8303f5ce54f132d703a9bb9b49e816", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1790986752.485, "t1": 1790986837.949, "seconds": 85.463, "host_samples": 85, "gpu_samples": 85}, "board": {"joules": 6982.3, "mean_w": 83.01, "max_w": 97.87, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": 3167.75, "label": "RAPL package-0, the CPU package: NOT WALL", "source": "/run/rapl/energy", "span_s": 84.012}, "data_reads_per_s": {"mean": 14.9, "max": 291.0}, "data_read_mb_per_s": {"mean": 0.06}, "temps": {"gpu_max_c": 58.0, "ssd_max_c": 44.85}, "vram": {"max_used_mib": 22843.0, "baseline_mib": 15.0, "growth_mib": 22828.0, "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": 21162237952, "rule": "sums over the engine unit's processes' /proc/<pid>/status, every 5 s"}, "load": {"load1_mean": 2.34, "claude_procs_max": 1}, "memory": {"mem_available_gib_min": 50.71, "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": 150.0, "max_w": 150.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}}
{"row_id": "L-Q2/speed/a1/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": 25.305, "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": 25.306669696850907, "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": "0.00.332.788 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 24"}, "prompt_tok_s_server": 418.12736521317623, "prompt_tok_s_client": 418.0, "ttft_ms": 76652.188, "prompt_tokens": 32038, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 32038, "prompt_ms": 76622.586, "prompt_per_token_ms": 2.3916157687745803, "prompt_per_second": 418.12736521317623, "predicted_n": 256, "predicted_ms": 10076.395, "predicted_per_token_ms": 39.51527450980392, "predicted_per_second": 25.306669696850907}, "short_completion": false, "cache_flag": null, "completion_sha256": "623f03ebb4bbd44b9c42fdabc25c430e09f43c931d82a65eb884d85a9920d868", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1790986837.95, "t1": 1790986924.679, "seconds": 86.729, "host_samples": 87, "gpu_samples": 87}, "board": {"joules": 7139.51, "mean_w": 82.91, "max_w": 96.55, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": 3440.77, "label": "RAPL package-0, the CPU package: NOT WALL", "source": "/run/rapl/energy", "span_s": 86.012}, "data_reads_per_s": {"mean": 20.9, "max": 334.9}, "data_read_mb_per_s": {"mean": 0.1}, "temps": {"gpu_max_c": 59.0, "ssd_max_c": 46.85}, "vram": {"max_used_mib": 22843.0, "baseline_mib": 15.0, "growth_mib": 22828.0, "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": 21175504896, "rule": "sums over the engine unit's processes' /proc/<pid>/status, every 5 s"}, "load": {"load1_mean": 2.51, "claude_procs_max": 1}, "memory": {"mem_available_gib_min": 50.127, "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": 150.0, "max_w": 150.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}}
{"row_id": "L-Q2/speed/a1/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": 16.878, "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": 16.879969909964224, "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": "0.00.332.788 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 24"}, "prompt_tok_s_server": 369.2149493929658, "prompt_tok_s_client": 369.1, "ttft_ms": 346884.075, "prompt_tokens": 128037, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 128037, "prompt_ms": 346781.733, "prompt_per_token_ms": 2.7084493779141967, "prompt_per_second": 369.2149493929658, "predicted_n": 256, "predicted_ms": 15106.662, "predicted_per_token_ms": 59.241811764705886, "predicted_per_second": 16.879969909964224}, "short_completion": false, "cache_flag": null, "completion_sha256": "ea59f7b77999fd11095fb213e7a9128b2622f91cba5f4256117a207d59daa9f8", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1790986924.682, "t1": 1790987286.675, "seconds": 361.993, "host_samples": 362, "gpu_samples": 361}, "board": {"joules": 28977.91, "mean_w": 80.41, "max_w": 120.46, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": 13309.89, "label": "RAPL package-0, the CPU package: NOT WALL", "source": "/run/rapl/energy", "span_s": 361.051}, "data_reads_per_s": {"mean": 1736.4, "max": 6452.8}, "data_read_mb_per_s": {"mean": 7.33}, "temps": {"gpu_max_c": 60.0, "ssd_max_c": 48.85}, "vram": {"max_used_mib": 22843.0, "baseline_mib": 15.0, "growth_mib": 22828.0, "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": 23775404032, "rule": "sums over the engine unit's processes' /proc/<pid>/status, every 5 s"}, "load": {"load1_mean": 2.28, "claude_procs_max": 1}, "memory": {"mem_available_gib_min": 49.709, "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"}}}
{"row_id": "L-Q2/speed/a1/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": 17.181, "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": 17.1830533585019, "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": "0.00.332.788 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 24"}, "prompt_tok_s_server": 405.4884277655795, "prompt_tok_s_client": 405.4, "ttft_ms": 315862.815, "prompt_tokens": 128037, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 128037, "prompt_ms": 315759.936, "prompt_per_token_ms": 2.4661616251552285, "prompt_per_second": 405.4884277655795, "predicted_n": 256, "predicted_ms": 14840.203, "predicted_per_token_ms": 58.19687450980392, "predicted_per_second": 17.1830533585019}, "short_completion": false, "cache_flag": null, "completion_sha256": "e67af08d48e0631db16a65280e8208303fa63fb555265d354967754c7dea8cab", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1790987286.678, "t1": 1790987617.383, "seconds": 330.705, "host_samples": 331, "gpu_samples": 331}, "board": {"joules": 28494.15, "mean_w": 86.25, "max_w": 119.22, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": 11856.98, "label": "RAPL package-0, the CPU package: NOT WALL", "source": "/run/rapl/energy", "span_s": 330.047}, "data_reads_per_s": {"mean": 52.3, "max": 7675.3}, "data_read_mb_per_s": {"mean": 0.62}, "temps": {"gpu_max_c": 61.0, "ssd_max_c": 49.85}, "vram": {"max_used_mib": 22843.0, "baseline_mib": 15.0, "growth_mib": 22828.0, "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": 23790469120, "rule": "sums over the engine unit's processes' /proc/<pid>/status, every 5 s"}, "load": {"load1_mean": 2.24, "claude_procs_max": 1}, "memory": {"mem_available_gib_min": 51.405, "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": 150.0, "max_w": 150.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}}
{"row_id": "L-Q2/speed/a1/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": 18.126, "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": 18.128318637546663, "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": "0.00.332.788 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 24"}, "prompt_tok_s_server": 405.14279249605187, "prompt_tok_s_client": 405.0, "ttft_ms": 316130.379, "prompt_tokens": 128037, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 128037, "prompt_ms": 316029.317, "prompt_per_token_ms": 2.4682655560502043, "prompt_per_second": 405.14279249605187, "predicted_n": 256, "predicted_ms": 14066.39, "predicted_per_token_ms": 55.16231372549019, "predicted_per_second": 18.128318637546663}, "short_completion": false, "cache_flag": null, "completion_sha256": "38598b13db0bc17c5843ef1f459098aabae7de635fa6043d51dc0bf0f96e754f", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1790987617.385, "t1": 1790987947.583, "seconds": 330.199, "host_samples": 330, "gpu_samples": 329}, "board": {"joules": 28431.3, "mean_w": 86.58, "max_w": 129.1, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": 11850.32, "label": "RAPL package-0, the CPU package: NOT WALL", "source": "/run/rapl/energy", "span_s": 329.047}, "data_reads_per_s": {"mean": 3.0, "max": 156.0}, "data_read_mb_per_s": {"mean": 0.01}, "temps": {"gpu_max_c": 61.0, "ssd_max_c": 49.85}, "vram": {"max_used_mib": 22843.0, "baseline_mib": 15.0, "growth_mib": 22828.0, "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": 23799640064, "rule": "sums over the engine unit's processes' /proc/<pid>/status, every 5 s"}, "load": {"load1_mean": 1.83, "claude_procs_max": 1}, "memory": {"mem_available_gib_min": 51.4, "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": 150.0, "max_w": 150.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}}