{"row_id": "R-LS-B11514-IQ3-24G/speed/a1/run-001", "arm": "R-LS-B11514-IQ3-24G", "engine": "llama.cpp", "engine_commit": "de7fa0a3c6a2e1b4cd9f22eb8d6bf5b12dbdb63b", "quant": "IQ3_S", "file_sha256": {"1": "4c1eb2ceb4915e1192f4f386021897bde56a97f40a0bb78bb86465e0f7d2aca3", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/workshop/card24-strata/build/llama.cpp-b11514-sm86/bin/llama-server", "-m", "/workshop/bench-flashnext/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf", "--host", "<loopback>", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "34", "-c", "131072", "-fa", "on", "-ctk", "q8_0", "-ctv", "q8_0", "-lm", "mmap", "--lazy-mode", "on", "-fit", "off", "-np", "1", "--cache-ram", "0", "--no-webui", "--jinja", "-lv", "4", "-t", "8", "-tb", "8"], "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": 22.657, "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": 22.656937274313584, "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": "0.00.275.839 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 64"}, "prompt_tok_s_server": 253.22526714074183, "prompt_tok_s_client": 253.0, "ttft_ms": 15958.066, "prompt_tokens": 4038, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 4038, "prompt_ms": 15946.276, "prompt_per_token_ms": 3.949052996532937, "prompt_per_second": 253.22526714074183, "predicted_n": 256, "predicted_ms": 11254.831, "predicted_per_token_ms": 44.13659215686275, "predicted_per_second": 22.656937274313584}, "short_completion": false, "cache_flag": null, "completion_sha256": "28133cc9e75467086e132961adbd62202167fb0e15f3cc13041362b69efdadb0", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791540964.197, "t1": 1791540991.41, "seconds": 27.213, "host_samples": 0, "gpu_samples": 28}, "board": {"joules": 4627.22, "mean_w": 171.35, "max_w": 189.3, "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": 58.0, "ssd_max_c": null}, "vram": {"max_used_mib": 22520.0, "baseline_mib": 1.0, "growth_mib": 22519.0, "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/<pid>/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@<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": 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-B11514-IQ3-24G/speed/a1/run-002", "arm": "R-LS-B11514-IQ3-24G", "engine": "llama.cpp", "engine_commit": "de7fa0a3c6a2e1b4cd9f22eb8d6bf5b12dbdb63b", "quant": "IQ3_S", "file_sha256": {"1": "4c1eb2ceb4915e1192f4f386021897bde56a97f40a0bb78bb86465e0f7d2aca3", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/workshop/card24-strata/build/llama.cpp-b11514-sm86/bin/llama-server", "-m", "/workshop/bench-flashnext/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf", "--host", "<loopback>", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "34", "-c", "131072", "-fa", "on", "-ctk", "q8_0", "-ctv", "q8_0", "-lm", "mmap", "--lazy-mode", "on", "-fit", "off", "-np", "1", "--cache-ram", "0", "--no-webui", "--jinja", "-lv", "4", "-t", "8", "-tb", "8"], "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": 22.694, "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": 22.69388722793367, "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": "0.00.275.839 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 64"}, "prompt_tok_s_server": 256.6182468664993, "prompt_tok_s_client": 256.2, "ttft_ms": 15758.368, "prompt_tokens": 4038, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 4038, "prompt_ms": 15735.436, "prompt_per_token_ms": 3.896839029222387, "prompt_per_second": 256.6182468664993, "predicted_n": 256, "predicted_ms": 11236.506, "predicted_per_token_ms": 44.0647294117647, "predicted_per_second": 22.69388722793367}, "short_completion": false, "cache_flag": null, "completion_sha256": "b6eb1d51b13e8a311acbe4f3c781b52e33a1ea908f79050eb56fe277cabd398c", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791540991.416, "t1": 1791541018.411, "seconds": 26.995, "host_samples": 0, "gpu_samples": 27}, "board": {"joules": 4450.99, "mean_w": 171.17, "max_w": 189.69, "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": 55.0, "ssd_max_c": null}, "vram": {"max_used_mib": 22520.0, "baseline_mib": 1.0, "growth_mib": 22519.0, "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/<pid>/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@<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": 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-B11514-IQ3-24G/speed/a1/run-003", "arm": "R-LS-B11514-IQ3-24G", "engine": "llama.cpp", "engine_commit": "de7fa0a3c6a2e1b4cd9f22eb8d6bf5b12dbdb63b", "quant": "IQ3_S", "file_sha256": {"1": "4c1eb2ceb4915e1192f4f386021897bde56a97f40a0bb78bb86465e0f7d2aca3", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/workshop/card24-strata/build/llama.cpp-b11514-sm86/bin/llama-server", "-m", "/workshop/bench-flashnext/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf", "--host", "<loopback>", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "34", "-c", "131072", "-fa", "on", "-ctk", "q8_0", "-ctv", "q8_0", "-lm", "mmap", "--lazy-mode", "on", "-fit", "off", "-np", "1", "--cache-ram", "0", "--no-webui", "--jinja", "-lv", "4", "-t", "8", "-tb", "8"], "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": 22.642, "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": 22.642556872109246, "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": "0.00.275.839 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 64"}, "prompt_tok_s_server": 255.15844372558877, "prompt_tok_s_client": 255.0, "ttft_ms": 15837.0, "prompt_tokens": 4038, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 4038, "prompt_ms": 15825.461, "prompt_per_token_ms": 3.9191334819217434, "prompt_per_second": 255.15844372558877, "predicted_n": 256, "predicted_ms": 11261.979, "predicted_per_token_ms": 44.16462352941176, "predicted_per_second": 22.642556872109246}, "short_completion": false, "cache_flag": null, "completion_sha256": "2f42cb414309477fc9d218ba3a726ae2762280c21580e79f45f37104fed104e0", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791541018.414, "t1": 1791541045.513, "seconds": 27.099, "host_samples": 0, "gpu_samples": 27}, "board": {"joules": 4369.54, "mean_w": 168.03, "max_w": 188.06, "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": 22520.0, "baseline_mib": 1.0, "growth_mib": 22519.0, "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/<pid>/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@<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": 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-B11514-IQ3-24G/speed/a1/run-004", "arm": "R-LS-B11514-IQ3-24G", "engine": "llama.cpp", "engine_commit": "de7fa0a3c6a2e1b4cd9f22eb8d6bf5b12dbdb63b", "quant": "IQ3_S", "file_sha256": {"1": "4c1eb2ceb4915e1192f4f386021897bde56a97f40a0bb78bb86465e0f7d2aca3", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/workshop/card24-strata/build/llama.cpp-b11514-sm86/bin/llama-server", "-m", "/workshop/bench-flashnext/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf", "--host", "<loopback>", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "34", "-c", "131072", "-fa", "on", "-ctk", "q8_0", "-ctv", "q8_0", "-lm", "mmap", "--lazy-mode", "on", "-fit", "off", "-np", "1", "--cache-ram", "0", "--no-webui", "--jinja", "-lv", "4", "-t", "8", "-tb", "8"], "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": 21.218, "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": 21.219970205497518, "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": "0.00.275.839 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 64"}, "prompt_tok_s_server": 244.7824746933635, "prompt_tok_s_client": 244.7, "ttft_ms": 130944.813, "prompt_tokens": 32038, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 32038, "prompt_ms": 130883.553, "prompt_per_token_ms": 4.08525978525501, "prompt_per_second": 244.7824746933635, "predicted_n": 256, "predicted_ms": 12016.982, "predicted_per_token_ms": 47.125419607843135, "predicted_per_second": 21.219970205497518}, "short_completion": false, "cache_flag": null, "completion_sha256": "cb3b179b4566e06666d02793b8c87956bc9d33e65a7ee48496ea6b07b62e3581", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791541045.518, "t1": 1791541188.481, "seconds": 142.963, "host_samples": 0, "gpu_samples": 143}, "board": {"joules": 25789.97, "mean_w": 181.6, "max_w": 197.61, "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": 22520.0, "baseline_mib": 1.0, "growth_mib": 22519.0, "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/<pid>/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@<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": 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-B11514-IQ3-24G/speed/a1/run-005", "arm": "R-LS-B11514-IQ3-24G", "engine": "llama.cpp", "engine_commit": "de7fa0a3c6a2e1b4cd9f22eb8d6bf5b12dbdb63b", "quant": "IQ3_S", "file_sha256": {"1": "4c1eb2ceb4915e1192f4f386021897bde56a97f40a0bb78bb86465e0f7d2aca3", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/workshop/card24-strata/build/llama.cpp-b11514-sm86/bin/llama-server", "-m", "/workshop/bench-flashnext/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf", "--host", "<loopback>", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "34", "-c", "131072", "-fa", "on", "-ctk", "q8_0", "-ctv", "q8_0", "-lm", "mmap", "--lazy-mode", "on", "-fit", "off", "-np", "1", "--cache-ram", "0", "--no-webui", "--jinja", "-lv", "4", "-t", "8", "-tb", "8"], "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": 21.316, "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": 21.317623053262125, "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": "0.00.275.839 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 64"}, "prompt_tok_s_server": 246.11936601808452, "prompt_tok_s_client": 246.0, "ttft_ms": 130233.163, "prompt_tokens": 32038, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 32038, "prompt_ms": 130172.609, "prompt_per_token_ms": 4.063069136650228, "prompt_per_second": 246.11936601808452, "predicted_n": 256, "predicted_ms": 11961.934, "predicted_per_token_ms": 46.90954509803921, "predicted_per_second": 21.317623053262125}, "short_completion": false, "cache_flag": null, "completion_sha256": "91c42bf6971b24a4e6034b461d889bddd8bcc818128e57814109854c7f6a3208", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791541188.484, "t1": 1791541330.68, "seconds": 142.196, "host_samples": 0, "gpu_samples": 142}, "board": {"joules": 25680.79, "mean_w": 182.11, "max_w": 200.37, "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": 22520.0, "baseline_mib": 1.0, "growth_mib": 22519.0, "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/<pid>/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@<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": 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-B11514-IQ3-24G/speed/a1/run-006", "arm": "R-LS-B11514-IQ3-24G", "engine": "llama.cpp", "engine_commit": "de7fa0a3c6a2e1b4cd9f22eb8d6bf5b12dbdb63b", "quant": "IQ3_S", "file_sha256": {"1": "4c1eb2ceb4915e1192f4f386021897bde56a97f40a0bb78bb86465e0f7d2aca3", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/workshop/card24-strata/build/llama.cpp-b11514-sm86/bin/llama-server", "-m", "/workshop/bench-flashnext/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf", "--host", "<loopback>", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "34", "-c", "131072", "-fa", "on", "-ctk", "q8_0", "-ctv", "q8_0", "-lm", "mmap", "--lazy-mode", "on", "-fit", "off", "-np", "1", "--cache-ram", "0", "--no-webui", "--jinja", "-lv", "4", "-t", "8", "-tb", "8"], "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": 21.257, "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": 21.2581914897874, "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": "0.00.275.839 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 64"}, "prompt_tok_s_server": 244.85728247424865, "prompt_tok_s_client": 244.7, "ttft_ms": 130904.528, "prompt_tokens": 32038, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 32038, "prompt_ms": 130843.566, "prompt_per_token_ms": 4.084011673637556, "prompt_per_second": 244.85728247424865, "predicted_n": 256, "predicted_ms": 11995.376, "predicted_per_token_ms": 47.04069019607843, "predicted_per_second": 21.2581914897874}, "short_completion": false, "cache_flag": null, "completion_sha256": "db751d20c11fde0fac3064236303fc8a9ac28e90e69e760b65f8c63e1f64c591", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791541330.685, "t1": 1791541473.586, "seconds": 142.901, "host_samples": 0, "gpu_samples": 143}, "board": {"joules": 25648.61, "mean_w": 180.6, "max_w": 201.74, "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": 57.0, "ssd_max_c": null}, "vram": {"max_used_mib": 22520.0, "baseline_mib": 1.0, "growth_mib": 22519.0, "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/<pid>/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@<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": 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-B11514-IQ3-24G/speed/a1/run-007", "arm": "R-LS-B11514-IQ3-24G", "engine": "llama.cpp", "engine_commit": "de7fa0a3c6a2e1b4cd9f22eb8d6bf5b12dbdb63b", "quant": "IQ3_S", "file_sha256": {"1": "4c1eb2ceb4915e1192f4f386021897bde56a97f40a0bb78bb86465e0f7d2aca3", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/workshop/card24-strata/build/llama.cpp-b11514-sm86/bin/llama-server", "-m", "/workshop/bench-flashnext/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf", "--host", "<loopback>", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "34", "-c", "131072", "-fa", "on", "-ctk", "q8_0", "-ctv", "q8_0", "-lm", "mmap", "--lazy-mode", "on", "-fit", "off", "-np", "1", "--cache-ram", "0", "--no-webui", "--jinja", "-lv", "4", "-t", "8", "-tb", "8"], "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": 17.681, "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": 17.685779052325213, "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": "0.00.275.839 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 64"}, "prompt_tok_s_server": 241.87810210762885, "prompt_tok_s_client": 241.8, "ttft_ms": 529589.884, "prompt_tokens": 128037, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 128037, "prompt_ms": 529345.149, "prompt_per_token_ms": 4.134313901450362, "prompt_per_second": 241.87810210762885, "predicted_n": 256, "predicted_ms": 14418.364, "predicted_per_token_ms": 56.54260392156863, "predicted_per_second": 17.685779052325213}, "short_completion": false, "cache_flag": null, "completion_sha256": "0ccb3cd4a365b8330e13d9a085e1c3a59aae0b3d5706c1a442e3066d258c3a9f", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791541473.598, "t1": 1791542017.61, "seconds": 544.012, "host_samples": 0, "gpu_samples": 544}, "board": {"joules": 99824.46, "mean_w": 183.82, "max_w": 250.32, "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": 57.0, "ssd_max_c": null}, "vram": {"max_used_mib": 22524.0, "baseline_mib": 1.0, "growth_mib": 22523.0, "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/<pid>/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@<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": 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-B11514-IQ3-24G/speed/a1/run-008", "arm": "R-LS-B11514-IQ3-24G", "engine": "llama.cpp", "engine_commit": "de7fa0a3c6a2e1b4cd9f22eb8d6bf5b12dbdb63b", "quant": "IQ3_S", "file_sha256": {"1": "4c1eb2ceb4915e1192f4f386021897bde56a97f40a0bb78bb86465e0f7d2aca3", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/workshop/card24-strata/build/llama.cpp-b11514-sm86/bin/llama-server", "-m", "/workshop/bench-flashnext/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf", "--host", "<loopback>", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "34", "-c", "131072", "-fa", "on", "-ctk", "q8_0", "-ctv", "q8_0", "-lm", "mmap", "--lazy-mode", "on", "-fit", "off", "-np", "1", "--cache-ram", "0", "--no-webui", "--jinja", "-lv", "4", "-t", "8", "-tb", "8"], "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": 17.683, "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": 17.687517336368096, "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": "0.00.275.839 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 64"}, "prompt_tok_s_server": 242.40629026574007, "prompt_tok_s_client": 242.3, "ttft_ms": 528432.687, "prompt_tokens": 128037, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 128037, "prompt_ms": 528191.739, "prompt_per_token_ms": 4.125305489819349, "prompt_per_second": 242.40629026574007, "predicted_n": 256, "predicted_ms": 14416.947, "predicted_per_token_ms": 56.53704705882353, "predicted_per_second": 17.687517336368096}, "short_completion": false, "cache_flag": null, "completion_sha256": "da85e3546e2d0b8edb1f33fe74430e8fffa7d6b291c7525ebb1ddd59572009d3", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791542017.616, "t1": 1791542560.47, "seconds": 542.854, "host_samples": 0, "gpu_samples": 542}, "board": {"joules": 99348.02, "mean_w": 183.61, "max_w": 250.25, "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": 22524.0, "baseline_mib": 1.0, "growth_mib": 22523.0, "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/<pid>/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@<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": 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-B11514-IQ3-24G/speed/a1/run-009", "arm": "R-LS-B11514-IQ3-24G", "engine": "llama.cpp", "engine_commit": "de7fa0a3c6a2e1b4cd9f22eb8d6bf5b12dbdb63b", "quant": "IQ3_S", "file_sha256": {"1": "4c1eb2ceb4915e1192f4f386021897bde56a97f40a0bb78bb86465e0f7d2aca3", "2": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113"}, "command_line": ["/workshop/card24-strata/build/llama.cpp-b11514-sm86/bin/llama-server", "-m", "/workshop/bench-flashnext/models/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf", "--host", "<loopback>", "--port", "withheld-2026-10-13", "-ngl", "99", "--n-cpu-moe", "34", "-c", "131072", "-fa", "on", "-ctk", "q8_0", "-ctv", "q8_0", "-lm", "mmap", "--lazy-mode", "on", "-fit", "off", "-np", "1", "--cache-ram", "0", "--no-webui", "--jinja", "-lv", "4", "-t", "8", "-tb", "8"], "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": 17.674, "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": 17.67830967074252, "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": "0.00.275.839 I cmn  common_param: system_info: n_threads = 8 (n_threads_batch = 8) / 64"}, "prompt_tok_s_server": 242.28285847517918, "prompt_tok_s_client": 242.2, "ttft_ms": 528721.288, "prompt_tokens": 128037, "completion_tokens": 256, "cached_tokens": 0, "finish_reason": "length", "server_timings": {"cache_n": 0, "prompt_n": 128037, "prompt_ms": 528460.828, "prompt_per_token_ms": 4.127407140123558, "prompt_per_second": 242.28285847517918, "predicted_n": 256, "predicted_ms": 14424.456, "predicted_per_token_ms": 56.56649411764706, "predicted_per_second": 17.67830967074252}, "short_completion": false, "cache_flag": null, "completion_sha256": "ab2dc8d666065e7be87c2c4279ec4f65879b9dfcba8f19950ea904eda26002df", "completion_head": "withheld-2026-10-13", "strata": null, "telemetry": {"window": {"t0": 1791542560.477, "t1": 1791543103.627, "seconds": 543.15, "host_samples": 0, "gpu_samples": 543}, "board": {"joules": 99537.43, "mean_w": 183.62, "max_w": 249.51, "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": 57.0, "ssd_max_c": null}, "vram": {"max_used_mib": 22524.0, "baseline_mib": 1.0, "growth_mib": 22523.0, "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/<pid>/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@<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": null}, "power_limit": {"min_w": 300.0, "max_w": 300.0, "rule": "enforced.power.limit READ (Dynamic Boost floats it), never one number"}}}