{"row_id": "R-LS-IQ3/speed/a1/run-001", "arm": "R-LS-IQ3", "engine": "llama.cpp", "engine_commit": "3cf03257f219afbe7334045ff7c6a06ac68c627d", "quant": "IQ3_S", "file_sha256": {"1": "4c1eb2ceb4915e1192f4f386021897bde56a97f40a0bb78bb86465e0f7d2aca3", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/workshop/bench-flashnext/llama.cpp-3cf03257/build/bin/llama-server", "-m", "/workshop/bench-flashnext/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf", "--host", "", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "0", "-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": 101.261, "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": 101.26099724212696, "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": "0.00.424.671 I cmn common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 64"}, "prompt_tok_s_server": 1154.5627243791726, "prompt_tok_s_client": 1146.9, "ttft_ms": 3520.728, "prompt_tokens": 4038, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 4038, "prompt_ms": 3497.428, "prompt_per_token_ms": 0.8661287766220901, "prompt_per_second": 1154.5627243791726, "predicted_n": 256, "predicted_ms": 2518.245, "predicted_per_token_ms": 9.875470588235293, "predicted_per_second": 101.26099724212696}, "short_completion": false, "cache_flag": null, "completion_sha256": "22b195e3994ccf3b3c8284dab6f071fa86621a5119d0201278f6cfaf1b9fc2f7", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791272304.915, "t1": 1791272310.955, "seconds": 6.039, "host_samples": 0, "gpu_samples": 6}, "board": {"joules": 1203.67, "mean_w": 234.95, "max_w": 308.56, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": null, "label": "RAPL package-0: NOT WALL", "why": "fewer than 2 readable samples of /run/rapl/energy in the window"}, "data_reads_per_s": {"mean": null, "max": null}, "data_read_mb_per_s": {"mean": null}, "temps": {"gpu_max_c": 44.0, "ssd_max_c": null}, "vram": {"max_used_mib": 55648.0, "baseline_mib": 2.0, "growth_mib": 55646.0, "rule": "max memory.used in the window minus the pre-load baseline (read before the unit started)"}, "pinned": {"vmlck_max_bytes": null, "vmpin_max_bytes": null, "vmrss_max_bytes": null, "rule": "sums over the engine unit's processes' /proc//status, every 5 s"}, "load": {"load1_mean": null, "claude_procs_max": null}, "memory": {"mem_available_gib_min": null, "user_psi_full_avg10_max": null, "psi_full_avg10_max": null, "unit_psi_full_avg10_max": null, "psi_rule": "psi = system-wide /proc/pressure/memory; user_psi = user@.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": null}, "power_limit": {"min_w": 300.0, "max_w": 300.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}} {"row_id": "R-LS-IQ3/speed/a1/run-002", "arm": "R-LS-IQ3", "engine": "llama.cpp", "engine_commit": "3cf03257f219afbe7334045ff7c6a06ac68c627d", "quant": "IQ3_S", "file_sha256": {"1": "4c1eb2ceb4915e1192f4f386021897bde56a97f40a0bb78bb86465e0f7d2aca3", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/workshop/bench-flashnext/llama.cpp-3cf03257/build/bin/llama-server", "-m", "/workshop/bench-flashnext/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf", "--host", "", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "0", "-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": 102.889, "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": 102.89568637071986, "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": "0.00.424.671 I cmn common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 64"}, "prompt_tok_s_server": 1623.5482852829248, "prompt_tok_s_client": 1608.9, "ttft_ms": 2509.844, "prompt_tokens": 4038, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 4038, "prompt_ms": 2487.145, "prompt_per_token_ms": 0.6159348687469044, "prompt_per_second": 1623.5482852829248, "predicted_n": 256, "predicted_ms": 2478.238, "predicted_per_token_ms": 9.718580392156863, "predicted_per_second": 102.89568637071986}, "short_completion": false, "cache_flag": null, "completion_sha256": "ac942ab2f1e99e9adc63f7e9c63bf7390fefb510a01d0d1b0e5b53af15a94e99", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791272310.957, "t1": 1791272315.946, "seconds": 4.989, "host_samples": 0, "gpu_samples": 5}, "board": {"joules": 1221.51, "mean_w": 299.1, "max_w": 301.5, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": null, "label": "RAPL package-0: NOT WALL", "why": "fewer than 2 readable samples of /run/rapl/energy in the window"}, "data_reads_per_s": {"mean": null, "max": null}, "data_read_mb_per_s": {"mean": null}, "temps": {"gpu_max_c": 46.0, "ssd_max_c": null}, "vram": {"max_used_mib": 55648.0, "baseline_mib": 2.0, "growth_mib": 55646.0, "rule": "max memory.used in the window minus the pre-load baseline (read before the unit started)"}, "pinned": {"vmlck_max_bytes": null, "vmpin_max_bytes": null, "vmrss_max_bytes": null, "rule": "sums over the engine unit's processes' /proc//status, every 5 s"}, "load": {"load1_mean": null, "claude_procs_max": null}, "memory": {"mem_available_gib_min": null, "user_psi_full_avg10_max": null, "psi_full_avg10_max": null, "unit_psi_full_avg10_max": null, "psi_rule": "psi = system-wide /proc/pressure/memory; user_psi = user@.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": null}, "power_limit": {"min_w": 300.0, "max_w": 300.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}} {"row_id": "R-LS-IQ3/speed/a1/run-003", "arm": "R-LS-IQ3", "engine": "llama.cpp", "engine_commit": "3cf03257f219afbe7334045ff7c6a06ac68c627d", "quant": "IQ3_S", "file_sha256": {"1": "4c1eb2ceb4915e1192f4f386021897bde56a97f40a0bb78bb86465e0f7d2aca3", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/workshop/bench-flashnext/llama.cpp-3cf03257/build/bin/llama-server", "-m", "/workshop/bench-flashnext/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf", "--host", "", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "0", "-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": 103.383, "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": 103.37932835058486, "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": "0.00.424.671 I cmn common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 64"}, "prompt_tok_s_server": 1619.4085188588103, "prompt_tok_s_client": 1604.7, "ttft_ms": 2516.41, "prompt_tokens": 4038, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 4038, "prompt_ms": 2493.503, "prompt_per_token_ms": 0.6175094105993066, "prompt_per_second": 1619.4085188588103, "predicted_n": 256, "predicted_ms": 2466.644, "predicted_per_token_ms": 9.673113725490195, "predicted_per_second": 103.37932835058486}, "short_completion": false, "cache_flag": null, "completion_sha256": "28cc264d3a756d1e95e8b101b8d566a1e603199d08d3047e5ed306eb9bb04407", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791272315.951, "t1": 1791272320.934, "seconds": 4.983, "host_samples": 0, "gpu_samples": 5}, "board": {"joules": 1224.36, "mean_w": 297.83, "max_w": 306.78, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": null, "label": "RAPL package-0: NOT WALL", "why": "fewer than 2 readable samples of /run/rapl/energy in the window"}, "data_reads_per_s": {"mean": null, "max": null}, "data_read_mb_per_s": {"mean": null}, "temps": {"gpu_max_c": 48.0, "ssd_max_c": null}, "vram": {"max_used_mib": 55648.0, "baseline_mib": 2.0, "growth_mib": 55646.0, "rule": "max memory.used in the window minus the pre-load baseline (read before the unit started)"}, "pinned": {"vmlck_max_bytes": null, "vmpin_max_bytes": null, "vmrss_max_bytes": null, "rule": "sums over the engine unit's processes' /proc//status, every 5 s"}, "load": {"load1_mean": null, "claude_procs_max": null}, "memory": {"mem_available_gib_min": null, "user_psi_full_avg10_max": null, "psi_full_avg10_max": null, "unit_psi_full_avg10_max": null, "psi_rule": "psi = system-wide /proc/pressure/memory; user_psi = user@.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": null}, "power_limit": {"min_w": 300.0, "max_w": 300.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}} {"row_id": "R-LS-IQ3/speed/a1/run-004", "arm": "R-LS-IQ3", "engine": "llama.cpp", "engine_commit": "3cf03257f219afbe7334045ff7c6a06ac68c627d", "quant": "IQ3_S", "file_sha256": {"1": "4c1eb2ceb4915e1192f4f386021897bde56a97f40a0bb78bb86465e0f7d2aca3", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/workshop/bench-flashnext/llama.cpp-3cf03257/build/bin/llama-server", "-m", "/workshop/bench-flashnext/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf", "--host", "", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "0", "-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": 76.337, "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": 76.35891926077379, "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": "0.00.424.671 I cmn common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 64"}, "prompt_tok_s_server": 1421.834600667647, "prompt_tok_s_client": 1416.1, "ttft_ms": 22624.162, "prompt_tokens": 32038, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 32038, "prompt_ms": 22532.86, "prompt_per_token_ms": 0.7033166864348587, "prompt_per_second": 1421.834600667647, "predicted_n": 256, "predicted_ms": 3339.492, "predicted_per_token_ms": 13.09604705882353, "predicted_per_second": 76.35891926077379}, "short_completion": false, "cache_flag": null, "completion_sha256": "cf1b35da5070898545b1d6dee55cf6602fd8cb912a3b997a7ebad816ec214c11", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791272320.94, "t1": 1791272346.905, "seconds": 25.965, "host_samples": 0, "gpu_samples": 25}, "board": {"joules": 7286.1, "mean_w": 295.92, "max_w": 316.75, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": null, "label": "RAPL package-0: NOT WALL", "why": "fewer than 2 readable samples of /run/rapl/energy in the window"}, "data_reads_per_s": {"mean": null, "max": null}, "data_read_mb_per_s": {"mean": null}, "temps": {"gpu_max_c": 53.0, "ssd_max_c": null}, "vram": {"max_used_mib": 55650.0, "baseline_mib": 2.0, "growth_mib": 55648.0, "rule": "max memory.used in the window minus the pre-load baseline (read before the unit started)"}, "pinned": {"vmlck_max_bytes": null, "vmpin_max_bytes": null, "vmrss_max_bytes": null, "rule": "sums over the engine unit's processes' /proc//status, every 5 s"}, "load": {"load1_mean": null, "claude_procs_max": null}, "memory": {"mem_available_gib_min": null, "user_psi_full_avg10_max": null, "psi_full_avg10_max": null, "unit_psi_full_avg10_max": null, "psi_rule": "psi = system-wide /proc/pressure/memory; user_psi = user@.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": null}, "power_limit": {"min_w": 300.0, "max_w": 300.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}} {"row_id": "R-LS-IQ3/speed/a1/run-005", "arm": "R-LS-IQ3", "engine": "llama.cpp", "engine_commit": "3cf03257f219afbe7334045ff7c6a06ac68c627d", "quant": "IQ3_S", "file_sha256": {"1": "4c1eb2ceb4915e1192f4f386021897bde56a97f40a0bb78bb86465e0f7d2aca3", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/workshop/bench-flashnext/llama.cpp-3cf03257/build/bin/llama-server", "-m", "/workshop/bench-flashnext/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf", "--host", "", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "0", "-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": 76.778, "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": 76.79885506459838, "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": "0.00.424.671 I cmn common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 64"}, "prompt_tok_s_server": 1552.8069974510372, "prompt_tok_s_client": 1546.0, "ttft_ms": 20723.005, "prompt_tokens": 32038, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 32038, "prompt_ms": 20632.313, "prompt_per_token_ms": 0.6439950371433921, "prompt_per_second": 1552.8069974510372, "predicted_n": 256, "predicted_ms": 3320.362, "predicted_per_token_ms": 13.021027450980393, "predicted_per_second": 76.79885506459838}, "short_completion": false, "cache_flag": null, "completion_sha256": "e3cdde9fd2287da6e2ead8c329b64adbfc0655de9cf614bf07aec53e25b9827e", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791272346.91, "t1": 1791272370.955, "seconds": 24.045, "host_samples": 0, "gpu_samples": 24}, "board": {"joules": 7079.55, "mean_w": 300.08, "max_w": 304.59, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": null, "label": "RAPL package-0: NOT WALL", "why": "fewer than 2 readable samples of /run/rapl/energy in the window"}, "data_reads_per_s": {"mean": null, "max": null}, "data_read_mb_per_s": {"mean": null}, "temps": {"gpu_max_c": 56.0, "ssd_max_c": null}, "vram": {"max_used_mib": 55650.0, "baseline_mib": 2.0, "growth_mib": 55648.0, "rule": "max memory.used in the window minus the pre-load baseline (read before the unit started)"}, "pinned": {"vmlck_max_bytes": null, "vmpin_max_bytes": null, "vmrss_max_bytes": null, "rule": "sums over the engine unit's processes' /proc//status, every 5 s"}, "load": {"load1_mean": null, "claude_procs_max": null}, "memory": {"mem_available_gib_min": null, "user_psi_full_avg10_max": null, "psi_full_avg10_max": null, "unit_psi_full_avg10_max": null, "psi_rule": "psi = system-wide /proc/pressure/memory; user_psi = user@.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": null}, "power_limit": {"min_w": 300.0, "max_w": 300.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}} {"row_id": "R-LS-IQ3/speed/a1/run-006", "arm": "R-LS-IQ3", "engine": "llama.cpp", "engine_commit": "3cf03257f219afbe7334045ff7c6a06ac68c627d", "quant": "IQ3_S", "file_sha256": {"1": "4c1eb2ceb4915e1192f4f386021897bde56a97f40a0bb78bb86465e0f7d2aca3", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/workshop/bench-flashnext/llama.cpp-3cf03257/build/bin/llama-server", "-m", "/workshop/bench-flashnext/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf", "--host", "", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "0", "-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": 75.92, "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": 75.93959615620568, "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": "0.00.424.671 I cmn common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 64"}, "prompt_tok_s_server": 1539.0793962819594, "prompt_tok_s_client": 1532.5, "ttft_ms": 20905.323, "prompt_tokens": 32038, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 32038, "prompt_ms": 20816.34, "prompt_per_token_ms": 0.6497390598664087, "prompt_per_second": 1539.0793962819594, "predicted_n": 256, "predicted_ms": 3357.932, "predicted_per_token_ms": 13.168360784313725, "predicted_per_second": 75.93959615620568}, "short_completion": false, "cache_flag": null, "completion_sha256": "8ae9f3fad508efa4b3b0badf8328bbffbfc5ce11c5ea9fa91b96cf4e2416b41b", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791272370.96, "t1": 1791272395.225, "seconds": 24.264, "host_samples": 0, "gpu_samples": 24}, "board": {"joules": 7059.34, "mean_w": 299.82, "max_w": 309.39, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": null, "label": "RAPL package-0: NOT WALL", "why": "fewer than 2 readable samples of /run/rapl/energy in the window"}, "data_reads_per_s": {"mean": null, "max": null}, "data_read_mb_per_s": {"mean": null}, "temps": {"gpu_max_c": 59.0, "ssd_max_c": null}, "vram": {"max_used_mib": 55650.0, "baseline_mib": 2.0, "growth_mib": 55648.0, "rule": "max memory.used in the window minus the pre-load baseline (read before the unit started)"}, "pinned": {"vmlck_max_bytes": null, "vmpin_max_bytes": null, "vmrss_max_bytes": null, "rule": "sums over the engine unit's processes' /proc//status, every 5 s"}, "load": {"load1_mean": null, "claude_procs_max": null}, "memory": {"mem_available_gib_min": null, "user_psi_full_avg10_max": null, "psi_full_avg10_max": null, "unit_psi_full_avg10_max": null, "psi_rule": "psi = system-wide /proc/pressure/memory; user_psi = user@.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": null}, "power_limit": {"min_w": 300.0, "max_w": 300.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}} {"row_id": "R-LS-IQ3/speed/a1/run-007", "arm": "R-LS-IQ3", "engine": "llama.cpp", "engine_commit": "3cf03257f219afbe7334045ff7c6a06ac68c627d", "quant": "IQ3_S", "file_sha256": {"1": "4c1eb2ceb4915e1192f4f386021897bde56a97f40a0bb78bb86465e0f7d2aca3", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/workshop/bench-flashnext/llama.cpp-3cf03257/build/bin/llama-server", "-m", "/workshop/bench-flashnext/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf", "--host", "", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "0", "-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": 38.712, "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": 38.733967365341364, "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": "0.00.424.671 I cmn common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 64"}, "prompt_tok_s_server": 1292.225431767975, "prompt_tok_s_client": 1288.7, "ttft_ms": 99349.854, "prompt_tokens": 128037, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 128037, "prompt_ms": 99082.557, "prompt_per_token_ms": 0.7738587830080368, "prompt_per_second": 1292.225431767975, "predicted_n": 256, "predicted_ms": 6583.369, "predicted_per_token_ms": 25.81713333333333, "predicted_per_second": 38.733967365341364}, "short_completion": false, "cache_flag": null, "completion_sha256": "223fc2e836275bcf1eb0336ed537365f93d15fde07a79612b990c08223014b68", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791272395.237, "t1": 1791272501.174, "seconds": 105.937, "host_samples": 0, "gpu_samples": 103}, "board": {"joules": 31062.04, "mean_w": 296.75, "max_w": 314.28, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": null, "label": "RAPL package-0: NOT WALL", "why": "fewer than 2 readable samples of /run/rapl/energy in the window"}, "data_reads_per_s": {"mean": null, "max": null}, "data_read_mb_per_s": {"mean": null}, "temps": {"gpu_max_c": 66.0, "ssd_max_c": null}, "vram": {"max_used_mib": 55654.0, "baseline_mib": 2.0, "growth_mib": 55652.0, "rule": "max memory.used in the window minus the pre-load baseline (read before the unit started)"}, "pinned": {"vmlck_max_bytes": null, "vmpin_max_bytes": null, "vmrss_max_bytes": null, "rule": "sums over the engine unit's processes' /proc//status, every 5 s"}, "load": {"load1_mean": null, "claude_procs_max": null}, "memory": {"mem_available_gib_min": null, "user_psi_full_avg10_max": null, "psi_full_avg10_max": null, "unit_psi_full_avg10_max": null, "psi_rule": "psi = system-wide /proc/pressure/memory; user_psi = user@.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": null}, "power_limit": {"min_w": 300.0, "max_w": 300.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}} {"row_id": "R-LS-IQ3/speed/a1/run-008", "arm": "R-LS-IQ3", "engine": "llama.cpp", "engine_commit": "3cf03257f219afbe7334045ff7c6a06ac68c627d", "quant": "IQ3_S", "file_sha256": {"1": "4c1eb2ceb4915e1192f4f386021897bde56a97f40a0bb78bb86465e0f7d2aca3", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/workshop/bench-flashnext/llama.cpp-3cf03257/build/bin/llama-server", "-m", "/workshop/bench-flashnext/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf", "--host", "", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "0", "-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": 38.576, "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": 38.598385164657685, "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": "0.00.424.671 I cmn common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 64"}, "prompt_tok_s_server": 1369.4908495780567, "prompt_tok_s_client": 1365.6, "ttft_ms": 93759.31, "prompt_tokens": 128037, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 128037, "prompt_ms": 93492.41, "prompt_per_token_ms": 0.7301983801557362, "prompt_per_second": 1369.4908495780567, "predicted_n": 256, "predicted_ms": 6606.494, "predicted_per_token_ms": 25.907819607843138, "predicted_per_second": 38.598385164657685}, "short_completion": false, "cache_flag": null, "completion_sha256": "063c0076e26dd131b5444e2c27bcfc47fbfe9f59e762a1b991d72accc52ad606", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791272501.187, "t1": 1791272601.557, "seconds": 100.37, "host_samples": 0, "gpu_samples": 98}, "board": {"joules": 29836.75, "mean_w": 299.7, "max_w": 309.66, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": null, "label": "RAPL package-0: NOT WALL", "why": "fewer than 2 readable samples of /run/rapl/energy in the window"}, "data_reads_per_s": {"mean": null, "max": null}, "data_read_mb_per_s": {"mean": null}, "temps": {"gpu_max_c": 67.0, "ssd_max_c": null}, "vram": {"max_used_mib": 55654.0, "baseline_mib": 2.0, "growth_mib": 55652.0, "rule": "max memory.used in the window minus the pre-load baseline (read before the unit started)"}, "pinned": {"vmlck_max_bytes": null, "vmpin_max_bytes": null, "vmrss_max_bytes": null, "rule": "sums over the engine unit's processes' /proc//status, every 5 s"}, "load": {"load1_mean": null, "claude_procs_max": null}, "memory": {"mem_available_gib_min": null, "user_psi_full_avg10_max": null, "psi_full_avg10_max": null, "unit_psi_full_avg10_max": null, "psi_rule": "psi = system-wide /proc/pressure/memory; user_psi = user@.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": null}, "power_limit": {"min_w": 300.0, "max_w": 300.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}} {"row_id": "R-LS-IQ3/speed/a1/run-009", "arm": "R-LS-IQ3", "engine": "llama.cpp", "engine_commit": "3cf03257f219afbe7334045ff7c6a06ac68c627d", "quant": "IQ3_S", "file_sha256": {"1": "4c1eb2ceb4915e1192f4f386021897bde56a97f40a0bb78bb86465e0f7d2aca3", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/workshop/bench-flashnext/llama.cpp-3cf03257/build/bin/llama-server", "-m", "/workshop/bench-flashnext/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf", "--host", "", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "0", "-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": 38.415, "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": 38.43714855445421, "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": "0.00.424.671 I cmn common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 64"}, "prompt_tok_s_server": 1362.7540273629838, "prompt_tok_s_client": 1358.9, "ttft_ms": 94222.055, "prompt_tokens": 128037, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 128037, "prompt_ms": 93954.593, "prompt_per_token_ms": 0.7338081413966275, "prompt_per_second": 1362.7540273629838, "predicted_n": 256, "predicted_ms": 6634.207, "predicted_per_token_ms": 26.016498039215687, "predicted_per_second": 38.43714855445421}, "short_completion": false, "cache_flag": null, "completion_sha256": "8cd8500b8fe12004e642a87d895b58f90679532d01b4aa565b638a13c5b2cbb3", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791272601.564, "t1": 1791272702.424, "seconds": 100.86, "host_samples": 0, "gpu_samples": 98}, "board": {"joules": 29821.87, "mean_w": 299.65, "max_w": 307.99, "rule": "trapezoid over nvidia-smi power.draw at 1 Hz"}, "rapl_package0": {"joules": null, "label": "RAPL package-0: NOT WALL", "why": "fewer than 2 readable samples of /run/rapl/energy in the window"}, "data_reads_per_s": {"mean": null, "max": null}, "data_read_mb_per_s": {"mean": null}, "temps": {"gpu_max_c": 69.0, "ssd_max_c": null}, "vram": {"max_used_mib": 55654.0, "baseline_mib": 2.0, "growth_mib": 55652.0, "rule": "max memory.used in the window minus the pre-load baseline (read before the unit started)"}, "pinned": {"vmlck_max_bytes": null, "vmpin_max_bytes": null, "vmrss_max_bytes": null, "rule": "sums over the engine unit's processes' /proc//status, every 5 s"}, "load": {"load1_mean": null, "claude_procs_max": null}, "memory": {"mem_available_gib_min": null, "user_psi_full_avg10_max": null, "psi_full_avg10_max": null, "unit_psi_full_avg10_max": null, "psi_rule": "psi = system-wide /proc/pressure/memory; user_psi = user@.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": null}, "power_limit": {"min_w": 300.0, "max_w": 300.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}}