{"cases":[{"case_wall_seconds_including_cooldown_load_warmup_cleanup":134.83513087104075,"completed_synthetic_response_telemetry":{"gpu":{"gpu_board_power_w":{"maximum":300.22,"mean":299.99376470588237,"median":299.99,"minimum":299.79,"p95":300.078,"samples":85},"gpu_memory_used_mib":{"maximum":44922.0,"mean":44922.0,"median":44922.0,"minimum":44922.0,"p95":44922.0,"samples":85},"gpu_sm_clock_mhz":{"maximum":1612.0,"mean":1496.4235294117648,"median":1500.0,"minimum":1372.0,"p95":1567.0,"samples":85},"gpu_utilization_percent":{"maximum":100.0,"mean":99.95294117647059,"median":100.0,"minimum":99.0,"p95":100.0,"samples":85}},"gpu_sample_count":85,"host_mem_available_bytes":{"maximum":96183144448,"mean":96139663721.41176,"median":96137916416,"minimum":96109170688,"p95":96163622912.0,"samples":85},"host_sample_count":85,"meaning":"Samples inside the union of completed synthetic response intervals; excludes loading, warmup and practical chat.","model_process_pss_bytes":{"maximum":1221457920,"mean":1209659066.7294118,"median":1211758592,"minimum":1203547136,"p95":1213958144.0,"samples":85},"model_process_rss_bytes":{"maximum":1224421376,"mean":1212639448.8470588,"median":1214722048,"minimum":1206513664,"p95":1216921600.0,"samples":85},"process_sample_count":85,"sampled_process_counter_deltas":{"deltas":{"major_faults":0,"minor_faults":57,"read_syscall_bytes":0,"storage_read_bytes":0,"storage_write_bytes":61440},"meaning":"Deltas between sampled endpoints within each interval; excludes unsampled edges. Storage reads are not all memory reads; rchar includes cached reads.","observed_endpoint_span_seconds":80.87426589097595,"observed_intervals":6,"requested_intervals":6},"task_cgroup_current_bytes":{"maximum":79866089472,"mean":79843703820.04706,"median":79846080512,"minimum":79834087424,"p95":79854927052.8,"samples":85}},"completed_synthetic_rounds":6,"finished_at":"2026-09-27T07:01:16.337762+00:00","id":"q4-full","maximum_gpu_sample_gap_seconds":1.0085429309983738,"model":{"architecture":"qwen2","key":"qwen25_72b_q4_k_m","license":"Qwen License Agreement, release 2024-09-19; custom, not Apache-2.0","quantization":"Q4_K_M","repository":"Qwen/Qwen2.5-72B-Instruct-GGUF","revision":"7ca3bc388f97b264c4283bc9bf1055e2abc38441","weight_bytes":44010465472,"weight_sha256":["3531a90691b5f4edb4bdf488246d72ce0b5d40cacfab1489718a662cec386e43","a4cfb079461591e2c23ef73953aeba4b2520a9b93afaa841f31adf4e480647ab","c67c3641d018b7ca646e7c37b50152687d4b52ce6ab6bf170b006a756b52778d","92f5149d685ad5c5ab0fc30f5e0725d95a746d9a8e54d8397707b97dc435cd76","f5abb6ecd7f3ec4438a4e89ae704acd43405182a0477776d5e8aa35a71a5ecdc","ead67a4ac2648b0f7b71fea755f910ed208f3f8b881b55c208fc3fdca67359ee","f7b168a5fdf78695736a6441dc846b19251339750ea78bf993057b7bf499671d","4ab253ada85c14097597546ac13cfb1426cb175d1d8da520b7a2904d0c323e32","51661b80af773b488e807a03fc7ea31bb64630c50e3ee77fbb5e37bdd9ff9979","fcc180433606a81e2137ba1f088defe3ad51ec9353f5d129a7209614597e16f9","57b9437649f12062bdbed6c623c3f6adc6924a3d2f4e7aabdd3ab32883f5485d","60ae16428f9edb62b60e6c8657ba9344c6fae6ef44c497845dc6a6fd002e4d5b"]},"observed_thermal_slowdown":false,"phase":"offload","placement":{"meaning":"Runtime allocations; CPU expert placement is requested separately. Not measured resident RAM.","model_buffer_allocations":[{"device":"CPU_Mapped","reported_mib":668.25},{"device":"CUDA0","reported_mib":41297.69}],"offloaded_layers":81,"requested_cpu_expert_layers":0,"total_layers":81},"practical_check_scope":"Three authored protocol/final-answer checks, not a general model quality benchmark.","practical_checks":[],"record_sha256":"eb9dbc0ebbb311a86bc4b79cbb8aa62abe51c7ff5227b5f847a2d1241c2dd45c","settings":{"batch":2048,"cache_prompt":false,"cache_ram_bytes":0,"concurrency":1,"context_shift":false,"cpu_expert_layers":0,"enable_thinking":null,"fit":"off","flash_attention":"on","generation_tokens":256,"kv_cache":"f16","lazy_mode":"off","load_mode":"mmap","mlock":false,"practical_max_output_tokens":1024,"practical_tasks":0,"prompt_token_counts":[2048,8192],"reasoning_effort":null,"repetitions":3,"requested_gpu_layers":999,"separate_warmup_requests_per_context":1,"server_slots":1,"socket_timeout_policy":"remaining_case_deadline","threads":16,"total_context_tokens":8704,"ubatch":512},"status":"completed","stop_category":null,"synthetic_groups":[{"completed_rounds":3,"context_tokens":2048,"end_to_end_tokens_per_second":{"maximum":22.561633989669705,"mean":22.380782470297998,"median":22.378564718631438,"minimum":22.20214870259285,"p95":22.54332706256588,"samples":3},"generated_tokens":768,"process_major_faults_total":0,"process_percent_host_capacity":{"maximum":3.1223003168275643,"mean":3.1218854735829527,"median":3.122085945517769,"minimum":3.121270158403524,"p95":3.1222788796965846,"samples":3},"response_seconds":{"maximum":11.530415521003306,"mean":11.438876565992055,"median":11.439518272003625,"minimum":11.346695904969238,"p95":11.521325796103337,"samples":3},"server_decode_tokens_per_second":{"maximum":26.27659423188569,"mean":26.07091165761903,"median":26.070546079792432,"minimum":25.865594661178967,"p95":26.255989416676364,"samples":3},"status":"completed","synthetic_first_token_seconds":{"maximum":1.6716317150276154,"mean":1.6573423466761597,"median":1.658250460983254,"minimum":1.6421448640176095,"p95":1.6702935896231792,"samples":3}},{"completed_rounds":3,"context_tokens":8192,"end_to_end_tokens_per_second":{"maximum":15.017619664246991,"mean":14.931806684631315,"median":14.921844775740391,"minimum":14.855955613906564,"p95":15.008042175396332,"samples":3},"generated_tokens":768,"process_major_faults_total":0,"process_percent_host_capacity":{"maximum":3.1220090990651133,"mean":3.121598621598756,"median":3.121880970335804,"minimum":3.1209057953953505,"p95":3.121996286192182,"samples":3},"response_seconds":{"maximum":17.232146261958405,"mean":17.144948292989284,"median":17.156055692001246,"minimum":17.0466429250082,"p95":17.22453720496269,"samples":3},"server_decode_tokens_per_second":{"maximum":25.326173806081957,"mean":25.186612365847278,"median":25.167049995381106,"minimum":25.066613296078774,"p95":25.310261425011873,"samples":3},"status":"completed","synthetic_first_token_seconds":{"maximum":7.058988248987589,"mean":7.020786065006784,"median":7.025630396034103,"minimum":6.97773954999866,"p95":7.05565246369224,"samples":3}}],"synthetic_response_sample_coverage":{"meaning":"Union of nominal one-second sample supports clipped to completed synthetic request intervals; not continuous utilization.","requested_response_seconds":85.75943036103854,"sample_supported_fraction":0.9984659923848805,"sample_supported_seconds":85.6278747417964},"synthetic_rounds":[{"context_tokens":2048,"cpu_snapshot_interval":{"host_busy_percent":3.248183219555164,"interval_seconds":11.353551022999454,"process_cpu_seconds":11.34,"process_major_faults":0,"process_minor_faults":76,"process_percent_host_capacity":3.121270158403524,"process_percent_one_logical_cpu":99.88064506891277},"end_to_end_tokens_per_second":22.561633989669705,"finished_at":"2026-09-27T06:59:44.573816+00:00","generated_tokens":256,"request_wall_seconds":11.353395794983953,"response_seconds":11.346695904969238,"round":0,"server_decode_tokens_per_second":26.27659423188569,"started_at":"2026-09-27T06:59:33.220260+00:00","synthetic_first_token_seconds":1.6421448640176095},{"context_tokens":2048,"cpu_snapshot_interval":{"host_busy_percent":3.2291552835755657,"interval_seconds":11.439882898994256,"process_cpu_seconds":11.43,"process_major_faults":0,"process_minor_faults":76,"process_percent_host_capacity":3.1223003168275643,"process_percent_one_logical_cpu":99.91361013848206},"end_to_end_tokens_per_second":22.378564718631438,"finished_at":"2026-09-27T06:59:56.018657+00:00","generated_tokens":256,"request_wall_seconds":11.439762005000375,"response_seconds":11.439518272003625,"round":1,"server_decode_tokens_per_second":26.070546079792432,"started_at":"2026-09-27T06:59:44.578764+00:00","synthetic_first_token_seconds":1.658250460983254},{"context_tokens":2048,"cpu_snapshot_interval":{"host_busy_percent":3.233412836079792,"interval_seconds":11.530752397025935,"process_cpu_seconds":11.52,"process_major_faults":0,"process_minor_faults":74,"process_percent_host_capacity":3.122085945517769,"process_percent_one_logical_cpu":99.90675025656861},"end_to_end_tokens_per_second":22.20214870259285,"finished_at":"2026-09-27T07:00:07.554172+00:00","generated_tokens":256,"request_wall_seconds":11.530644086014945,"response_seconds":11.530415521003306,"round":2,"server_decode_tokens_per_second":25.865594661178967,"started_at":"2026-09-27T06:59:56.023409+00:00","synthetic_first_token_seconds":1.6716317150276154},{"context_tokens":8192,"cpu_snapshot_interval":{"host_busy_percent":3.228467715322847,"interval_seconds":17.047014445997775,"process_cpu_seconds":17.03,"process_major_faults":0,"process_minor_faults":175,"process_percent_host_capacity":3.121880970335804,"process_percent_one_logical_cpu":99.90019105074573},"end_to_end_tokens_per_second":15.017619664246991,"finished_at":"2026-09-27T07:00:41.755572+00:00","generated_tokens":256,"request_wall_seconds":17.046893169987015,"response_seconds":17.0466429250082,"round":0,"server_decode_tokens_per_second":25.326173806081957,"started_at":"2026-09-27T07:00:24.708557+00:00","synthetic_first_token_seconds":6.97773954999866},{"context_tokens":8192,"cpu_snapshot_interval":{"host_busy_percent":3.2387926480499845,"interval_seconds":17.156420209037606,"process_cpu_seconds":17.14,"process_major_faults":0,"process_minor_faults":167,"process_percent_host_capacity":3.1220090990651133,"process_percent_one_logical_cpu":99.90429117008362},"end_to_end_tokens_per_second":14.921844775740391,"finished_at":"2026-09-27T07:00:58.916653+00:00","generated_tokens":256,"request_wall_seconds":17.156293494044803,"response_seconds":17.156055692001246,"round":1,"server_decode_tokens_per_second":25.167049995381106,"started_at":"2026-09-27T07:00:41.760218+00:00","synthetic_first_token_seconds":7.025630396034103},{"context_tokens":8192,"cpu_snapshot_interval":{"host_busy_percent":3.248036850982028,"interval_seconds":17.232577182992827,"process_cpu_seconds":17.21,"process_major_faults":0,"process_minor_faults":166,"process_percent_host_capacity":3.1209057953953505,"process_percent_one_logical_cpu":99.86898545265122},"end_to_end_tokens_per_second":14.855955613906564,"finished_at":"2026-09-27T07:01:16.154134+00:00","generated_tokens":256,"request_wall_seconds":17.232441811007448,"response_seconds":17.232146261958405,"round":2,"server_decode_tokens_per_second":25.066613296078774,"started_at":"2026-09-27T07:00:58.921537+00:00","synthetic_first_token_seconds":7.058988248987589}],"thermal_states_fully_available":true,"whole_case_telemetry":{"gpu":{"gpu_board_power_w":{"maximum":300.22,"mean":262.8214814814815,"median":299.99,"minimum":14.8,"p95":300.073,"samples":135},"gpu_memory_used_mib":{"maximum":44922.0,"mean":39852.57777777778,"median":44922.0,"minimum":2.0,"p95":44922.0,"samples":135},"gpu_sm_clock_mhz":{"maximum":2295.0,"mean":1596.8,"median":1515.0,"minimum":180.0,"p95":2272.0,"samples":135},"gpu_utilization_percent":{"maximum":100.0,"mean":86.96296296296296,"median":100.0,"minimum":0.0,"p95":100.0,"samples":135}},"gpu_sample_count":135,"host_mem_available_bytes":{"maximum":96501841920,"mean":96160067300.43077,"median":96142043136.0,"minimum":96109170688,"p95":96272954777.6,"samples":130},"host_sample_count":130,"meaning":"Whole monitored case, including model loading, warmup, synthetic requests, practical chat and cleanup.","model_process_pss_bytes":{"maximum":44345792512,"mean":5018296621.643411,"median":1211410432,"minimum":1194694656,"p95":35546540031.999985,"samples":129},"model_process_rss_bytes":{"maximum":44348755968,"mean":5021293290.170543,"median":1214373888,"minimum":1197658112,"p95":35549503487.999985,"samples":129},"process_sample_count":129,"sampled_process_counter_deltas":{"deltas":{"major_faults":36,"minor_faults":675062,"read_syscall_bytes":13,"storage_read_bytes":33542475776,"storage_write_bytes":122880},"meaning":"Deltas between sampled endpoints within each interval; excludes unsampled edges. Storage reads are not all memory reads; rchar includes cached reads.","observed_endpoint_span_seconds":133.40514452999923,"observed_intervals":1,"requested_intervals":1},"task_cgroup_current_bytes":{"maximum":79887130624,"mean":79830709043.2,"median":79842607104.0,"minimum":79659089920,"p95":79850843955.2,"samples":130}}},{"case_wall_seconds_including_cooldown_load_warmup_cleanup":2397.005270417023,"completed_synthetic_response_telemetry":{"gpu":{"gpu_board_power_w":{"maximum":151.89,"mean":85.86188370684434,"median":81.37,"minimum":79.75,"p95":104.565,"samples":1651},"gpu_memory_used_mib":{"maximum":23072.0,"mean":23065.648697758934,"median":23072.0,"minimum":23058.0,"p95":23072.0,"samples":1651},"gpu_sm_clock_mhz":{"maximum":2332.0,"mean":2265.74318594791,"median":2272.0,"minimum":1845.0,"p95":2295.0,"samples":1651},"gpu_utilization_percent":{"maximum":100.0,"mean":11.63840096910963,"median":0.0,"minimum":0.0,"p95":88.0,"samples":1651}},"gpu_sample_count":1651,"host_mem_available_bytes":{"maximum":94660886528,"mean":94573956764.31444,"median":94572523520.0,"minimum":94494953472,"p95":94631029350.4,"samples":1094},"host_sample_count":1094,"meaning":"Samples inside the union of completed synthetic response intervals; excludes loading, warmup and practical chat.","model_process_pss_bytes":{"maximum":24367884288,"mean":24359668685.45521,"median":24364712960.0,"minimum":24350567424,"p95":24366330880.0,"samples":1094},"model_process_rss_bytes":{"maximum":24370868224,"mean":24362727311.678246,"median":24367697920.0,"minimum":24354967552,"p95":24369315840.0,"samples":1094},"process_sample_count":1094,"sampled_process_counter_deltas":{"deltas":{"major_faults":0,"minor_faults":850,"read_syscall_bytes":0,"storage_read_bytes":0,"storage_write_bytes":753664},"meaning":"Deltas between sampled endpoints within each interval; excludes unsampled edges. Storage reads are not all memory reads; rchar includes cached reads.","observed_endpoint_span_seconds":1641.875876798993,"observed_intervals":6,"requested_intervals":6},"task_cgroup_current_bytes":{"maximum":79762886656,"mean":79599925480.13162,"median":79564417024.0,"minimum":79511523328,"p95":79724648038.4,"samples":1094}},"completed_synthetic_rounds":6,"finished_at":"2026-09-27T07:41:13.346577+00:00","id":"q4-half","maximum_gpu_sample_gap_seconds":1.0485468429978937,"model":{"architecture":"qwen2","key":"qwen25_72b_q4_k_m","license":"Qwen License Agreement, release 2024-09-19; custom, not Apache-2.0","quantization":"Q4_K_M","repository":"Qwen/Qwen2.5-72B-Instruct-GGUF","revision":"7ca3bc388f97b264c4283bc9bf1055e2abc38441","weight_bytes":44010465472,"weight_sha256":["3531a90691b5f4edb4bdf488246d72ce0b5d40cacfab1489718a662cec386e43","a4cfb079461591e2c23ef73953aeba4b2520a9b93afaa841f31adf4e480647ab","c67c3641d018b7ca646e7c37b50152687d4b52ce6ab6bf170b006a756b52778d","92f5149d685ad5c5ab0fc30f5e0725d95a746d9a8e54d8397707b97dc435cd76","f5abb6ecd7f3ec4438a4e89ae704acd43405182a0477776d5e8aa35a71a5ecdc","ead67a4ac2648b0f7b71fea755f910ed208f3f8b881b55c208fc3fdca67359ee","f7b168a5fdf78695736a6441dc846b19251339750ea78bf993057b7bf499671d","4ab253ada85c14097597546ac13cfb1426cb175d1d8da520b7a2904d0c323e32","51661b80af773b488e807a03fc7ea31bb64630c50e3ee77fbb5e37bdd9ff9979","fcc180433606a81e2137ba1f088defe3ad51ec9353f5d129a7209614597e16f9","57b9437649f12062bdbed6c623c3f6adc6924a3d2f4e7aabdd3ab32883f5485d","60ae16428f9edb62b60e6c8657ba9344c6fae6ef44c497845dc6a6fd002e4d5b"]},"observed_thermal_slowdown":false,"phase":"offload","placement":{"meaning":"Runtime allocations; CPU expert placement is requested separately. Not measured resident RAM.","model_buffer_allocations":[{"device":"CPU_Mapped","reported_mib":3796.51},{"device":"CPU_Mapped","reported_mib":3698.21},{"device":"CPU_Mapped","reported_mib":3780.0},{"device":"CPU_Mapped","reported_mib":3758.87},{"device":"CPU_Mapped","reported_mib":3780.06},{"device":"CPU_Mapped","reported_mib":2489.76},{"device":"CUDA0","reported_mib":20662.52}],"offloaded_layers":40,"requested_cpu_expert_layers":0,"total_layers":81},"practical_check_scope":"Three authored protocol/final-answer checks, not a general model quality benchmark.","practical_checks":[],"record_sha256":"ac532f7a91e400ecf0bc245fa57df15571f022fec492724985b03f5841b4583c","settings":{"batch":2048,"cache_prompt":false,"cache_ram_bytes":0,"concurrency":1,"context_shift":false,"cpu_expert_layers":0,"enable_thinking":null,"fit":"off","flash_attention":"on","generation_tokens":256,"kv_cache":"f16","lazy_mode":"off","load_mode":"mmap","mlock":false,"practical_max_output_tokens":1024,"practical_tasks":0,"prompt_token_counts":[2048,8192],"reasoning_effort":null,"repetitions":3,"requested_gpu_layers":40,"separate_warmup_requests_per_context":1,"server_slots":1,"socket_timeout_policy":"remaining_case_deadline","threads":16,"total_context_tokens":8704,"ubatch":512},"status":"completed","stop_category":null,"synthetic_groups":[{"completed_rounds":3,"context_tokens":2048,"end_to_end_tokens_per_second":{"maximum":1.0259597546104118,"mean":1.0257673518018657,"median":1.0256743485760798,"minimum":1.025667952219106,"p95":1.0259312140069785,"samples":3},"generated_tokens":768,"process_major_faults_total":0,"process_percent_host_capacity":{"maximum":46.96900255302731,"mean":46.96399084284636,"median":46.96377666986252,"minimum":46.959193305649244,"p95":46.96847996471083,"samples":3},"response_seconds":{"maximum":249.59344731998863,"mean":249.56926544297798,"median":249.5918907939922,"minimum":249.52245821495308,"p95":249.593291667389,"samples":3},"server_decode_tokens_per_second":{"maximum":1.0753894431143844,"mean":1.0752842293155311,"median":1.0752987478264153,"minimum":1.0751644970057934,"p95":1.0753803735855876,"samples":3},"status":"completed","synthetic_first_token_seconds":{"maximum":12.405359767959453,"mean":12.395018015309082,"median":12.398921652988065,"minimum":12.380772624979727,"p95":12.404715956462315,"samples":3}},{"completed_rounds":3,"context_tokens":8192,"end_to_end_tokens_per_second":{"maximum":0.8516457474153253,"mean":0.8509680880909979,"median":0.8510538170372035,"minimum":0.8502046998204648,"p95":0.8515865543775131,"samples":3},"generated_tokens":768,"process_major_faults_total":0,"process_percent_host_capacity":{"maximum":41.477068411758815,"mean":41.47344784202933,"median":41.47489702951262,"minimum":41.468378084816564,"p95":41.476851273534194,"samples":3},"response_seconds":{"maximum":301.1039577340125,"mean":300.8339880830026,"median":300.80353894800646,"minimum":300.5944675669889,"p95":301.0739158554119,"samples":3},"server_decode_tokens_per_second":{"maximum":1.021093775230151,"mean":1.020123239332381,"median":1.0201182684315688,"minimum":1.0191576743354231,"p95":1.0209962245502928,"samples":3},"status":"completed","synthetic_first_token_seconds":{"maximum":50.896970200003125,"mean":50.86368395232906,"median":50.86190725595225,"minimum":50.83217440103181,"p95":50.89346390559804,"samples":3}}],"synthetic_response_sample_coverage":{"meaning":"Union of nominal one-second sample supports clipped to completed synthetic request intervals; not continuous utilization.","requested_response_seconds":1651.2117659900105,"sample_supported_fraction":0.9967973477605081,"sample_supported_seconds":1645.9235089297872},"synthetic_rounds":[{"context_tokens":2048,"cpu_snapshot_interval":{"host_busy_percent":48.4844462255172,"interval_seconds":249.59236786101246,"process_cpu_seconds":3750.61,"process_major_faults":0,"process_minor_faults":176,"process_percent_host_capacity":46.959193305649244,"process_percent_one_logical_cpu":1502.6941857807758},"end_to_end_tokens_per_second":1.0256743485760798,"finished_at":"2026-09-27T07:12:49.722631+00:00","generated_tokens":256,"request_wall_seconds":249.59218785603298,"response_seconds":249.5918907939922,"round":0,"server_decode_tokens_per_second":1.0752987478264153,"started_at":"2026-09-27T07:08:40.130266+00:00","synthetic_first_token_seconds":12.405359767959453},{"context_tokens":2048,"cpu_snapshot_interval":{"host_busy_percent":48.51933878831305,"interval_seconds":249.52294306800468,"process_cpu_seconds":3750.35,"process_major_faults":0,"process_minor_faults":154,"process_percent_host_capacity":46.96900255302731,"process_percent_one_logical_cpu":1503.008081696874},"end_to_end_tokens_per_second":1.0259597546104118,"finished_at":"2026-09-27T07:16:59.303702+00:00","generated_tokens":256,"request_wall_seconds":249.52277030603727,"response_seconds":249.52245821495308,"round":1,"server_decode_tokens_per_second":1.0753894431143844,"started_at":"2026-09-27T07:12:49.780732+00:00","synthetic_first_token_seconds":12.398921652988065},{"context_tokens":2048,"cpu_snapshot_interval":{"host_busy_percent":48.483254625124765,"interval_seconds":249.5939600939746,"process_cpu_seconds":3751.0,"process_major_faults":0,"process_minor_faults":167,"process_percent_host_capacity":46.96377666986252,"process_percent_one_logical_cpu":1502.8408534356006},"end_to_end_tokens_per_second":1.025667952219106,"finished_at":"2026-09-27T07:21:08.955486+00:00","generated_tokens":256,"request_wall_seconds":249.59377860696986,"response_seconds":249.59344731998863,"round":2,"server_decode_tokens_per_second":1.0751644970057934,"started_at":"2026-09-27T07:16:59.361514+00:00","synthetic_first_token_seconds":12.380772624979727},{"context_tokens":8192,"cpu_snapshot_interval":{"host_busy_percent":42.85339102163519,"interval_seconds":301.1045751629863,"process_cpu_seconds":3996.25,"process_major_faults":0,"process_minor_faults":360,"process_percent_host_capacity":41.47489702951262,"process_percent_one_logical_cpu":1327.1967049444038},"end_to_end_tokens_per_second":0.8502046998204648,"finished_at":"2026-09-27T07:31:11.470017+00:00","generated_tokens":256,"request_wall_seconds":301.1043590210029,"response_seconds":301.1039577340125,"round":0,"server_decode_tokens_per_second":1.0191576743354231,"started_at":"2026-09-27T07:26:10.365421+00:00","synthetic_first_token_seconds":50.896970200003125},{"context_tokens":8192,"cpu_snapshot_interval":{"host_busy_percent":42.83598994160311,"interval_seconds":300.804015996051,"process_cpu_seconds":3992.47,"process_major_faults":0,"process_minor_faults":465,"process_percent_host_capacity":41.477068411758815,"process_percent_one_logical_cpu":1327.266189176282},"end_to_end_tokens_per_second":0.8510538170372035,"finished_at":"2026-09-27T07:36:12.334538+00:00","generated_tokens":256,"request_wall_seconds":300.80384594900534,"response_seconds":300.80353894800646,"round":1,"server_decode_tokens_per_second":1.0201182684315688,"started_at":"2026-09-27T07:31:11.530500+00:00","synthetic_first_token_seconds":50.83217440103181},{"context_tokens":8192,"cpu_snapshot_interval":{"host_busy_percent":42.79206444441441,"interval_seconds":300.59500939497957,"process_cpu_seconds":3988.86,"process_major_faults":0,"process_minor_faults":252,"process_percent_host_capacity":41.468378084816564,"process_percent_one_logical_cpu":1326.98809871413},"end_to_end_tokens_per_second":0.8516457474153253,"finished_at":"2026-09-27T07:41:12.984901+00:00","generated_tokens":256,"request_wall_seconds":300.5948242509621,"response_seconds":300.5944675669889,"round":2,"server_decode_tokens_per_second":1.021093775230151,"started_at":"2026-09-27T07:36:12.389829+00:00","synthetic_first_token_seconds":50.86190725595225}],"thermal_states_fully_available":true,"whole_case_telemetry":{"gpu":{"gpu_board_power_w":{"maximum":151.89,"mean":85.64786425339366,"median":81.3,"minimum":16.08,"p95":104.40949999999998,"samples":2210},"gpu_memory_used_mib":{"maximum":23072.0,"mean":23042.219909502262,"median":23072.0,"minimum":2.0,"p95":23072.0,"samples":2210},"gpu_sm_clock_mhz":{"maximum":2332.0,"mean":2264.581447963801,"median":2272.0,"minimum":180.0,"p95":2295.0,"samples":2210},"gpu_utilization_percent":{"maximum":100.0,"mean":11.75158371040724,"median":0.0,"minimum":0.0,"p95":88.0,"samples":2210}},"gpu_sample_count":2210,"host_mem_available_bytes":{"maximum":96539258880,"mean":94582910735.71623,"median":94575390720.0,"minimum":94494953472,"p95":94634216448.0,"samples":1466},"host_sample_count":1466,"meaning":"Whole monitored case, including model loading, warmup, synthetic requests, practical chat and cleanup.","model_process_pss_bytes":{"maximum":44346136576,"mean":24386928776.393444,"median":24363987968.0,"minimum":24183633920,"p95":24366330880.0,"samples":1464},"model_process_rss_bytes":{"maximum":44349104128,"mean":24389981925.420765,"median":24366972928.0,"minimum":24186617856,"p95":24369315840.0,"samples":1464},"process_sample_count":1464,"sampled_process_counter_deltas":{"deltas":{"major_faults":49,"minor_faults":670202,"read_syscall_bytes":13,"storage_read_bytes":376832,"storage_write_bytes":1052672},"meaning":"Deltas between sampled endpoints within each interval; excludes unsampled edges. Storage reads are not all memory reads; rchar includes cached reads.","observed_endpoint_span_seconds":2207.407058215991,"observed_intervals":1,"requested_intervals":1},"task_cgroup_current_bytes":{"maximum":79792500736,"mean":79617826799.23602,"median":79618897920.0,"minimum":77636681728,"p95":79760701440.0,"samples":1466}}},{"case_wall_seconds_including_cooldown_load_warmup_cleanup":3356.0603392009507,"completed_synthetic_response_telemetry":{"gpu":{"gpu_board_power_w":{"maximum":128.79,"mean":81.51851284175642,"median":78.215,"minimum":75.71,"p95":103.5535,"samples":2414},"gpu_memory_used_mib":{"maximum":12522.0,"mean":12510.144159072079,"median":12522.0,"minimum":12496.0,"p95":12522.0,"samples":2414},"gpu_sm_clock_mhz":{"maximum":2332.0,"mean":2270.75932062966,"median":2272.0,"minimum":2100.0,"p95":2295.0,"samples":2414},"gpu_utilization_percent":{"maximum":100.0,"mean":10.679784589892295,"median":0.0,"minimum":0.0,"p95":88.0,"samples":2414}},"gpu_sample_count":2414,"host_mem_available_bytes":{"maximum":93916610560,"mean":93808274930.49751,"median":93812375552,"minimum":93682814976,"p95":93859961241.6,"samples":1403},"host_sample_count":1403,"meaning":"Samples inside the union of completed synthetic response intervals; excludes loading, warmup and practical chat.","model_process_pss_bytes":{"maximum":35467370496,"mean":35456058797.890236,"median":35465952256,"minimum":35436528640,"p95":35467369472.0,"samples":1403},"model_process_rss_bytes":{"maximum":35470389248,"mean":35459139763.546684,"median":35468972032,"minimum":35439550464,"p95":35470389248.0,"samples":1403},"process_sample_count":1403,"sampled_process_counter_deltas":{"deltas":{"major_faults":0,"minor_faults":456,"read_syscall_bytes":0,"storage_read_bytes":0,"storage_write_bytes":1052672},"meaning":"Deltas between sampled endpoints within each interval; excludes unsampled edges. Storage reads are not all memory reads; rchar includes cached reads.","observed_endpoint_span_seconds":2404.3422632078873,"observed_intervals":6,"requested_intervals":6},"task_cgroup_current_bytes":{"maximum":79639121920,"mean":79582972305.42552,"median":79602151424,"minimum":79464935424,"p95":79620349952.0,"samples":1403}},"completed_synthetic_rounds":6,"finished_at":"2026-09-27T08:37:09.413985+00:00","id":"q4-quarter","maximum_gpu_sample_gap_seconds":1.0552173189935274,"model":{"architecture":"qwen2","key":"qwen25_72b_q4_k_m","license":"Qwen License Agreement, release 2024-09-19; custom, not Apache-2.0","quantization":"Q4_K_M","repository":"Qwen/Qwen2.5-72B-Instruct-GGUF","revision":"7ca3bc388f97b264c4283bc9bf1055e2abc38441","weight_bytes":44010465472,"weight_sha256":["3531a90691b5f4edb4bdf488246d72ce0b5d40cacfab1489718a662cec386e43","a4cfb079461591e2c23ef73953aeba4b2520a9b93afaa841f31adf4e480647ab","c67c3641d018b7ca646e7c37b50152687d4b52ce6ab6bf170b006a756b52778d","92f5149d685ad5c5ab0fc30f5e0725d95a746d9a8e54d8397707b97dc435cd76","f5abb6ecd7f3ec4438a4e89ae704acd43405182a0477776d5e8aa35a71a5ecdc","ead67a4ac2648b0f7b71fea755f910ed208f3f8b881b55c208fc3fdca67359ee","f7b168a5fdf78695736a6441dc846b19251339750ea78bf993057b7bf499671d","4ab253ada85c14097597546ac13cfb1426cb175d1d8da520b7a2904d0c323e32","51661b80af773b488e807a03fc7ea31bb64630c50e3ee77fbb5e37bdd9ff9979","fcc180433606a81e2137ba1f088defe3ad51ec9353f5d129a7209614597e16f9","57b9437649f12062bdbed6c623c3f6adc6924a3d2f4e7aabdd3ab32883f5485d","60ae16428f9edb62b60e6c8657ba9344c6fae6ef44c497845dc6a6fd002e4d5b"]},"observed_thermal_slowdown":false,"phase":"offload","placement":{"meaning":"Runtime allocations; CPU expert placement is requested separately. Not measured resident RAM.","model_buffer_allocations":[{"device":"CPU_Mapped","reported_mib":3796.51},{"device":"CPU_Mapped","reported_mib":3698.21},{"device":"CPU_Mapped","reported_mib":3780.0},{"device":"CPU_Mapped","reported_mib":3758.87},{"device":"CPU_Mapped","reported_mib":3780.06},{"device":"CPU_Mapped","reported_mib":3758.87},{"device":"CPU_Mapped","reported_mib":3649.56},{"device":"CPU_Mapped","reported_mib":3799.34},{"device":"CPU_Mapped","reported_mib":1180.14},{"device":"CUDA0","reported_mib":10764.37}],"offloaded_layers":20,"requested_cpu_expert_layers":0,"total_layers":81},"practical_check_scope":"Three authored protocol/final-answer checks, not a general model quality benchmark.","practical_checks":[],"record_sha256":"b4a31a6f88efc0d0fbf882898f6f9cebe5c45ca91438d2a090349aa5bf061fea","settings":{"batch":2048,"cache_prompt":false,"cache_ram_bytes":0,"concurrency":1,"context_shift":false,"cpu_expert_layers":0,"enable_thinking":null,"fit":"off","flash_attention":"on","generation_tokens":256,"kv_cache":"f16","lazy_mode":"off","load_mode":"mmap","mlock":false,"practical_max_output_tokens":1024,"practical_tasks":0,"prompt_token_counts":[2048,8192],"reasoning_effort":null,"repetitions":3,"requested_gpu_layers":20,"separate_warmup_requests_per_context":1,"server_slots":1,"socket_timeout_policy":"remaining_case_deadline","threads":16,"total_context_tokens":8704,"ubatch":512},"status":"completed","stop_category":null,"synthetic_groups":[{"completed_rounds":3,"context_tokens":2048,"end_to_end_tokens_per_second":{"maximum":0.7013341052329727,"mean":0.701202602178391,"median":0.7012849783098507,"minimum":0.7009887229923496,"p95":0.7013291925406605,"samples":3},"generated_tokens":768,"process_major_faults_total":0,"process_percent_host_capacity":{"maximum":47.5096755201097,"mean":47.50553008444596,"median":47.50472599520471,"minimum":47.50218873802347,"p95":47.509180567619204,"samples":3},"response_seconds":{"maximum":365.1984569840133,"mean":365.0870822846773,"median":365.04418020899175,"minimum":365.01860966102686,"p95":365.1830293065112,"samples":3},"server_decode_tokens_per_second":{"maximum":0.7340297133673556,"mean":0.733935543108075,"median":0.7340103847040825,"minimum":0.7337665312527869,"p95":0.7340277805010282,"samples":3},"status":"completed","synthetic_first_token_seconds":{"maximum":17.637427559995558,"mean":17.62843499101776,"median":17.626818710006773,"minimum":17.62105870305095,"p95":17.63636667499668,"samples":3}},{"completed_rounds":3,"context_tokens":8192,"end_to_end_tokens_per_second":{"maximum":0.5825000382803726,"mean":0.5823115658426037,"median":0.5824616954051061,"minimum":0.5819729638423324,"p95":0.5824962039928459,"samples":3},"generated_tokens":768,"process_major_faults_total":0,"process_percent_host_capacity":{"maximum":42.10484388012346,"mean":42.10116718544757,"median":42.10106130418197,"minimum":42.09759637203728,"p95":42.10446562252931,"samples":3},"response_seconds":{"maximum":439.88297722599236,"mean":439.62726913398365,"median":439.5138805169845,"minimum":439.4849496589741,"p95":439.84606755509157,"samples":3},"server_decode_tokens_per_second":{"maximum":0.6942791371869166,"mean":0.6940278638454777,"median":0.694182201615894,"minimum":0.6936222527336223,"p95":0.6942694436298144,"samples":3},"status":"completed","synthetic_first_token_seconds":{"maximum":72.24734390201047,"mean":72.2064675160024,"median":72.22615488199517,"minimum":72.14590376400156,"p95":72.24522500000894,"samples":3}}],"synthetic_response_sample_coverage":{"meaning":"Union of nominal one-second sample supports clipped to completed synthetic request intervals; not continuous utilization.","requested_response_seconds":2414.1450248520123,"sample_supported_fraction":0.9967236809588818,"sample_supported_seconds":2406.2355155390687},"synthetic_rounds":[{"context_tokens":2048,"cpu_snapshot_interval":{"host_busy_percent":49.27854515950404,"interval_seconds":365.1989615820348,"process_cpu_seconds":5551.28,"process_major_faults":0,"process_minor_faults":83,"process_percent_host_capacity":47.50218873802347,"process_percent_one_logical_cpu":1520.070039616751},"end_to_end_tokens_per_second":0.7009887229923496,"finished_at":"2026-09-27T07:55:40.044919+00:00","generated_tokens":256,"request_wall_seconds":365.19879893003963,"response_seconds":365.1984569840133,"round":0,"server_decode_tokens_per_second":0.7337665312527869,"started_at":"2026-09-27T07:49:34.845950+00:00","synthetic_first_token_seconds":17.626818710006773},{"context_tokens":2048,"cpu_snapshot_interval":{"host_busy_percent":49.28621767727671,"interval_seconds":365.0190684350091,"process_cpu_seconds":5549.42,"process_major_faults":0,"process_minor_faults":79,"process_percent_host_capacity":47.5096755201097,"process_percent_one_logical_cpu":1520.3096166435105},"end_to_end_tokens_per_second":0.7013341052329727,"finished_at":"2026-09-27T08:01:45.150425+00:00","generated_tokens":256,"request_wall_seconds":365.01890212501166,"response_seconds":365.01860966102686,"round":1,"server_decode_tokens_per_second":0.7340297133673556,"started_at":"2026-09-27T07:55:40.131331+00:00","synthetic_first_token_seconds":17.62105870305095},{"context_tokens":2048,"cpu_snapshot_interval":{"host_busy_percent":49.27091610269337,"interval_seconds":365.0446010729647,"process_cpu_seconds":5549.23,"process_major_faults":0,"process_minor_faults":172,"process_percent_host_capacity":47.50472599520471,"process_percent_one_logical_cpu":1520.1512318465507},"end_to_end_tokens_per_second":0.7012849783098507,"finished_at":"2026-09-27T08:07:50.276692+00:00","generated_tokens":256,"request_wall_seconds":365.0444458829588,"response_seconds":365.04418020899175,"round":2,"server_decode_tokens_per_second":0.7340103847040825,"started_at":"2026-09-27T08:01:45.232068+00:00","synthetic_first_token_seconds":17.637427559995558},{"context_tokens":8192,"cpu_snapshot_interval":{"host_busy_percent":43.708825928668,"interval_seconds":439.4855091419886,"process_cpu_seconds":5921.43,"process_major_faults":0,"process_minor_faults":498,"process_percent_host_capacity":42.10484388012346,"process_percent_one_logical_cpu":1347.3550041639508},"end_to_end_tokens_per_second":0.5825000382803726,"finished_at":"2026-09-27T08:22:29.346032+00:00","generated_tokens":256,"request_wall_seconds":439.4853082670015,"response_seconds":439.4849496589741,"round":0,"server_decode_tokens_per_second":0.694182201615894,"started_at":"2026-09-27T08:15:09.860511+00:00","synthetic_first_token_seconds":72.14590376400156},{"context_tokens":8192,"cpu_snapshot_interval":{"host_busy_percent":43.705407271659276,"interval_seconds":439.5144044919871,"process_cpu_seconds":5920.8,"process_major_faults":0,"process_minor_faults":173,"process_percent_host_capacity":42.09759637203728,"process_percent_one_logical_cpu":1347.123083905193},"end_to_end_tokens_per_second":0.5824616954051061,"finished_at":"2026-09-27T08:29:48.940982+00:00","generated_tokens":256,"request_wall_seconds":439.51424308202695,"response_seconds":439.5138805169845,"round":1,"server_decode_tokens_per_second":0.6942791371869166,"started_at":"2026-09-27T08:22:29.426553+00:00","synthetic_first_token_seconds":72.22615488199517},{"context_tokens":8192,"cpu_snapshot_interval":{"host_busy_percent":43.73398136206829,"interval_seconds":439.8835071210051,"process_cpu_seconds":5926.26,"process_major_faults":0,"process_minor_faults":175,"process_percent_host_capacity":42.10106130418197,"process_percent_one_logical_cpu":1347.233961733823},"end_to_end_tokens_per_second":0.5819729638423324,"finished_at":"2026-09-27T08:37:08.916183+00:00","generated_tokens":256,"request_wall_seconds":439.8833265649737,"response_seconds":439.88297722599236,"round":2,"server_decode_tokens_per_second":0.6936222527336223,"started_at":"2026-09-27T08:29:49.032650+00:00","synthetic_first_token_seconds":72.24734390201047}],"thermal_states_fully_available":true,"whole_case_telemetry":{"gpu":{"gpu_board_power_w":{"maximum":128.79,"mean":81.35600557793616,"median":78.13,"minimum":16.31,"p95":103.6,"samples":3227},"gpu_memory_used_mib":{"maximum":12522.0,"mean":12501.080880074373,"median":12520.0,"minimum":2.0,"p95":12522.0,"samples":3227},"gpu_sm_clock_mhz":{"maximum":2332.0,"mean":2270.0529903935544,"median":2272.0,"minimum":180.0,"p95":2295.0,"samples":3227},"gpu_utilization_percent":{"maximum":100.0,"mean":10.76169817167648,"median":0.0,"minimum":0.0,"p95":88.0,"samples":3227}},"gpu_sample_count":3227,"host_mem_available_bytes":{"maximum":96488124416,"mean":93807796210.89978,"median":93804103680.0,"minimum":93682814976,"p95":93855240192.0,"samples":1876},"host_sample_count":1876,"meaning":"Whole monitored case, including model loading, warmup, synthetic requests, practical chat and cleanup.","model_process_pss_bytes":{"maximum":44345984000,"mean":35449552226.44054,"median":35463912448,"minimum":20068582400,"p95":35467369472.0,"samples":1875},"model_process_rss_bytes":{"maximum":44348952576,"mean":35452636954.624,"median":35466932224,"minimum":20071550976,"p95":35470389248.0,"samples":1875},"process_sample_count":1875,"sampled_process_counter_deltas":{"deltas":{"major_faults":0,"minor_faults":587594,"read_syscall_bytes":13,"storage_read_bytes":0,"storage_write_bytes":1441792},"meaning":"Deltas between sampled endpoints within each interval; excludes unsampled edges. Storage reads are not all memory reads; rchar includes cached reads.","observed_endpoint_span_seconds":3225.0150158809847,"observed_intervals":1,"requested_intervals":1},"task_cgroup_current_bytes":{"maximum":79677063168,"mean":79592343715.75267,"median":79603474432.0,"minimum":77591044096,"p95":79662526464.0,"samples":1876}}},{"case_wall_seconds_including_cooldown_load_warmup_cleanup":168.67818345100386,"completed_synthetic_response_telemetry":{"gpu":{"gpu_board_power_w":{"maximum":304.26,"mean":299.7254545454545,"median":300.32,"minimum":295.34,"p95":304.17499999999995,"samples":11},"gpu_memory_used_mib":{"maximum":60980.0,"mean":60980.0,"median":60980.0,"minimum":60980.0,"p95":60980.0,"samples":11},"gpu_sm_clock_mhz":{"maximum":2205.0,"mean":2067.7272727272725,"median":2167.0,"minimum":1837.0,"p95":2197.5,"samples":11},"gpu_utilization_percent":{"maximum":100.0,"mean":96.9090909090909,"median":96.0,"minimum":96.0,"p95":100.0,"samples":11}},"gpu_sample_count":11,"host_mem_available_bytes":{"maximum":95835840512,"mean":95822113326.54546,"median":95820488704,"minimum":95807123456,"p95":95833702400.0,"samples":11},"host_sample_count":11,"meaning":"Samples inside the union of completed synthetic response intervals; excludes loading, warmup and practical chat.","model_process_pss_bytes":{"maximum":787760128,"mean":782783301.8181819,"median":786387968,"minimum":775484416,"p95":787758080.0,"samples":11},"model_process_rss_bytes":{"maximum":790724608,"mean":785876805.8181819,"median":789352448,"minimum":778448896,"p95":790722560.0,"samples":11},"process_sample_count":11,"sampled_process_counter_deltas":{"deltas":{"major_faults":0,"minor_faults":21,"read_syscall_bytes":0,"storage_read_bytes":0,"storage_write_bytes":8192},"meaning":"Deltas between sampled endpoints within each interval; excludes unsampled edges. Storage reads are not all memory reads; rchar includes cached reads.","observed_endpoint_span_seconds":5.0855146999820136,"observed_intervals":4,"requested_intervals":6},"task_cgroup_current_bytes":{"maximum":86953459712,"mean":86940625082.18182,"median":86938574848,"minimum":86936174592,"p95":86948157440.0,"samples":11}},"completed_synthetic_rounds":6,"finished_at":"2026-09-27T08:39:58.100667+00:00","id":"gpt120","maximum_gpu_sample_gap_seconds":1.0085606679785997,"model":{"architecture":"gpt-oss","key":"gpt_oss_120b_mxfp4","license":"Apache-2.0","quantization":"MXFP4","repository":"ggml-org/gpt-oss-120b-GGUF","revision":"238abdd290bb874b90a5da1b4549881b7d05c091","weight_bytes":63387346208,"weight_sha256":["582bd40f6886200101f4c4ed9f25f3fe80cc14c86e9e2b37746cd8904a0c622d"]},"observed_thermal_slowdown":false,"phase":"capacity","placement":{"meaning":"Runtime allocations; CPU expert placement is requested separately. Not measured resident RAM.","model_buffer_allocations":[{"device":"CPU_Mapped","reported_mib":586.82},{"device":"CUDA0","reported_mib":59851.68}],"offloaded_layers":37,"requested_cpu_expert_layers":0,"total_layers":37},"practical_check_scope":"Three authored protocol/final-answer checks, not a general model quality benchmark.","practical_checks":[{"all_completion_tokens_per_response_second":158.87294147051108,"completion_tokens_including_reasoning":43,"exact_expected_answer":true,"expected_final_answer_valid":true,"finish_reason":"stop","protocol_completed":true,"response_seconds":0.27065653598401695,"strict_json_object_valid":true,"task":"arithmetic","time_to_first_generated_event_seconds":0.09872354700928554,"time_to_first_visible_answer_seconds":0.23107441700994968,"usable_final_answer":true},{"all_completion_tokens_per_response_second":157.774595640092,"completion_tokens_including_reasoning":38,"exact_expected_answer":true,"expected_final_answer_valid":true,"finish_reason":"stop","protocol_completed":true,"response_seconds":0.24084992799907923,"strict_json_object_valid":true,"task":"extraction","time_to_first_generated_event_seconds":0.09198999195359647,"time_to_first_visible_answer_seconds":0.16127040795981884,"usable_final_answer":true},{"all_completion_tokens_per_response_second":157.5665023569006,"completion_tokens_including_reasoning":41,"exact_expected_answer":true,"expected_final_answer_valid":true,"finish_reason":"stop","protocol_completed":true,"response_seconds":0.26020759099628776,"strict_json_object_valid":true,"task":"classification","time_to_first_generated_event_seconds":0.09682328201597556,"time_to_first_visible_answer_seconds":0.22113073198124766,"usable_final_answer":true}],"record_sha256":"ccc093e5f70b89e37e1922afd2f1b64fd4417d0a342bee809502be03568e70b7","settings":{"batch":2048,"cache_prompt":false,"cache_ram_bytes":0,"concurrency":1,"context_shift":false,"cpu_expert_layers":0,"enable_thinking":null,"fit":"off","flash_attention":"on","generation_tokens":256,"kv_cache":"f16","lazy_mode":"off","load_mode":"mmap","mlock":false,"practical_max_output_tokens":1024,"practical_tasks":3,"prompt_token_counts":[2048,8192],"reasoning_effort":"low","repetitions":3,"requested_gpu_layers":999,"separate_warmup_requests_per_context":1,"server_slots":1,"socket_timeout_policy":"remaining_case_deadline","threads":16,"total_context_tokens":8704,"ubatch":512},"status":"completed","stop_category":null,"synthetic_groups":[{"completed_rounds":3,"context_tokens":2048,"end_to_end_tokens_per_second":{"maximum":178.1160600042134,"mean":177.7701214459848,"median":178.03356057298308,"minimum":177.16074376075798,"p95":178.10781006109036,"samples":3},"generated_tokens":768,"process_major_faults_total":0,"process_percent_host_capacity":{"maximum":3.1519040574611377,"mean":3.124084236105666,"median":3.12865438560508,"minimum":3.09169426525078,"p95":3.149579090275532,"samples":3},"response_seconds":{"maximum":1.4450153830111958,"mean":1.4400705456694898,"median":1.4379311359953135,"minimum":1.4372651180019602,"p95":1.4443069583096075,"samples":3},"server_decode_tokens_per_second":{"maximum":221.76709237251654,"mean":221.7327060722943,"median":221.71695873890872,"minimum":221.71406710545762,"p95":221.76207900915577,"samples":3},"status":"completed","synthetic_first_token_seconds":{"maximum":0.29474928602576256,"mean":0.28990076133050025,"median":0.2876625449862331,"minimum":0.28729045297950506,"p95":0.29404061192180964,"samples":3}},{"completed_rounds":3,"context_tokens":8192,"end_to_end_tokens_per_second":{"maximum":110.91731050360823,"mean":110.85796452044715,"median":110.9151435010248,"minimum":110.74143955670841,"p95":110.91709380334989,"samples":3},"generated_tokens":768,"process_major_faults_total":0,"process_percent_host_capacity":{"maximum":3.1356795216590876,"mean":3.1209553492635873,"median":3.1136377348508733,"minimum":3.1135487912808015,"p95":3.133475342978266,"samples":3},"response_seconds":{"maximum":2.3116910979733802,"mean":2.309262510990569,"median":2.308070764003787,"minimum":2.308025670994539,"p95":2.311329064576421,"samples":3},"server_decode_tokens_per_second":{"maximum":215.36907502782907,"mean":215.25240757576316,"median":215.26544339616268,"minimum":215.12270430329772,"p95":215.35871186466244,"samples":3},"status":"completed","synthetic_first_token_seconds":{"maximum":1.1260871259728447,"mean":1.124355525651481,"median":1.1237770399893634,"minimum":1.123202410992235,"p95":1.1258561173744965,"samples":3}}],"synthetic_response_sample_coverage":{"meaning":"Union of nominal one-second sample supports clipped to completed synthetic request intervals; not continuous utilization.","requested_response_seconds":11.249449516064487,"sample_supported_fraction":0.9993302839463225,"sample_supported_seconds":11.241915579128545},"synthetic_rounds":[{"context_tokens":2048,"cpu_snapshot_interval":{"host_busy_percent":3.217391304347826,"interval_seconds":1.4376230739871971,"process_cpu_seconds":1.45,"process_major_faults":0,"process_minor_faults":75,"process_percent_host_capacity":3.1519040574611377,"process_percent_one_logical_cpu":100.8609298387564},"end_to_end_tokens_per_second":178.1160600042134,"finished_at":"2026-09-27T08:39:44.972226+00:00","generated_tokens":256,"request_wall_seconds":1.4374833730398677,"response_seconds":1.4372651180019602,"round":0,"server_decode_tokens_per_second":221.76709237251654,"started_at":"2026-09-27T08:39:43.534598+00:00","synthetic_first_token_seconds":0.28729045297950506},{"context_tokens":2048,"cpu_snapshot_interval":{"host_busy_percent":3.2439446366782008,"interval_seconds":1.4454048869665712,"process_cpu_seconds":1.43,"process_major_faults":0,"process_minor_faults":76,"process_percent_host_capacity":3.09169426525078,"process_percent_one_logical_cpu":98.93421648802496},"end_to_end_tokens_per_second":177.16074376075798,"finished_at":"2026-09-27T08:39:46.423065+00:00","generated_tokens":256,"request_wall_seconds":1.4452448390075006,"response_seconds":1.4450153830111958,"round":1,"server_decode_tokens_per_second":221.71695873890872,"started_at":"2026-09-27T08:39:44.977640+00:00","synthetic_first_token_seconds":0.29474928602576256},{"context_tokens":2048,"cpu_snapshot_interval":{"host_busy_percent":3.217391304347826,"interval_seconds":1.4383180260192603,"process_cpu_seconds":1.44,"process_major_faults":0,"process_minor_faults":77,"process_percent_host_capacity":3.12865438560508,"process_percent_one_logical_cpu":100.11694033936256},"end_to_end_tokens_per_second":178.03356057298308,"finished_at":"2026-09-27T08:39:47.866509+00:00","generated_tokens":256,"request_wall_seconds":1.4381742990226485,"response_seconds":1.4379311359953135,"round":2,"server_decode_tokens_per_second":221.71406710545762,"started_at":"2026-09-27T08:39:46.428176+00:00","synthetic_first_token_seconds":0.2876625449862331},{"context_tokens":8192,"cpu_snapshot_interval":{"host_busy_percent":3.262929867316545,"interval_seconds":2.3084590869839303,"process_cpu_seconds":2.3,"process_major_faults":0,"process_minor_faults":169,"process_percent_host_capacity":3.1135487912808015,"process_percent_one_logical_cpu":99.63356132098565},"end_to_end_tokens_per_second":110.9151435010248,"finished_at":"2026-09-27T08:39:52.504600+00:00","generated_tokens":256,"request_wall_seconds":2.30831027001841,"response_seconds":2.308070764003787,"round":0,"server_decode_tokens_per_second":215.36907502782907,"started_at":"2026-09-27T08:39:50.196132+00:00","synthetic_first_token_seconds":1.1237770399893634},{"context_tokens":8192,"cpu_snapshot_interval":{"host_busy_percent":3.22231248307609,"interval_seconds":2.3083931439905427,"process_cpu_seconds":2.3,"process_major_faults":0,"process_minor_faults":168,"process_percent_host_capacity":3.1136377348508733,"process_percent_one_logical_cpu":99.63640751522794},"end_to_end_tokens_per_second":110.91731050360823,"finished_at":"2026-09-27T08:39:54.817707+00:00","generated_tokens":256,"request_wall_seconds":2.308253513998352,"response_seconds":2.308025670994539,"round":1,"server_decode_tokens_per_second":215.26544339616268,"started_at":"2026-09-27T08:39:52.509308+00:00","synthetic_first_token_seconds":1.123202410992235},{"context_tokens":8192,"cpu_snapshot_interval":{"host_busy_percent":3.217085698837524,"interval_seconds":2.3120985259884037,"process_cpu_seconds":2.32,"process_major_faults":0,"process_minor_faults":167,"process_percent_host_capacity":3.1356795216590876,"process_percent_one_logical_cpu":100.3417446930908},"end_to_end_tokens_per_second":110.74143955670841,"finished_at":"2026-09-27T08:39:57.135055+00:00","generated_tokens":256,"request_wall_seconds":2.3119832209777087,"response_seconds":2.3116910979733802,"round":2,"server_decode_tokens_per_second":215.12270430329772,"started_at":"2026-09-27T08:39:54.822941+00:00","synthetic_first_token_seconds":1.1260871259728447}],"thermal_states_fully_available":true,"whole_case_telemetry":{"gpu":{"gpu_board_power_w":{"maximum":304.26,"mean":123.15410714285714,"median":72.24,"minimum":16.09,"p95":302.35,"samples":56},"gpu_memory_used_mib":{"maximum":60982.0,"mean":29569.571428571428,"median":560.0,"minimum":2.0,"p95":60980.0,"samples":56},"gpu_sm_clock_mhz":{"maximum":2310.0,"mean":1775.6607142857142,"median":2205.0,"minimum":180.0,"p95":2310.0,"samples":56},"gpu_utilization_percent":{"maximum":100.0,"mean":37.80357142857143,"median":0.0,"minimum":0.0,"p95":97.75,"samples":56}},"gpu_sample_count":56,"host_mem_available_bytes":{"maximum":96513536000,"mean":95995402403.84,"median":96000301056.0,"minimum":95807123456,"p95":96255614566.4,"samples":50},"host_sample_count":50,"meaning":"Whole monitored case, including model loading, warmup, synthetic requests, practical chat and cleanup.","model_process_pss_bytes":{"maximum":63191684096,"mean":24869091056.32653,"median":18047583232,"minimum":738870272,"p95":63191677952.0,"samples":49},"model_process_rss_bytes":{"maximum":63194648576,"mean":24872084793.469387,"median":18050547712,"minimum":741834752,"p95":63194644480.0,"samples":49},"process_sample_count":49,"sampled_process_counter_deltas":{"deltas":{"major_faults":527,"minor_faults":1327351,"read_syscall_bytes":13,"storage_read_bytes":66957901824,"storage_write_bytes":53248},"meaning":"Deltas between sampled endpoints within each interval; excludes unsampled edges. Storage reads are not all memory reads; rchar includes cached reads.","observed_endpoint_span_seconds":54.13240855000913,"observed_intervals":1,"requested_intervals":1},"task_cgroup_current_bytes":{"maximum":86972710912,"mean":85898072473.6,"median":86948876288.0,"minimum":76791627776,"p95":86967723622.4,"samples":50}}},{"case_wall_seconds_including_cooldown_load_warmup_cleanup":300.7333330209949,"completed_synthetic_response_telemetry":{"gpu":{"gpu_board_power_w":{"maximum":300.19,"mean":299.9938181818182,"median":299.99,"minimum":299.77,"p95":300.07,"samples":110},"gpu_memory_used_mib":{"maximum":76294.0,"mean":76294.0,"median":76294.0,"minimum":76294.0,"p95":76294.0,"samples":110},"gpu_sm_clock_mhz":{"maximum":1747.0,"mean":1649.0545454545454,"median":1680.0,"minimum":1477.0,"p95":1736.4,"samples":110},"gpu_utilization_percent":{"maximum":100.0,"mean":100.0,"median":100.0,"minimum":100.0,"p95":100.0,"samples":110}},"gpu_sample_count":110,"host_mem_available_bytes":{"maximum":96089788416,"mean":96066580631.7037,"median":96062722048.0,"minimum":96034291712,"p95":96086418432.0,"samples":108},"host_sample_count":108,"meaning":"Samples inside the union of completed synthetic response intervals; excludes loading, warmup and practical chat.","model_process_pss_bytes":{"maximum":1860451328,"mean":1840726802.9629629,"median":1853541376.0,"minimum":1814436864,"p95":1860451328.0,"samples":108},"model_process_rss_bytes":{"maximum":1863409664,"mean":1843724781.0370371,"median":1856520192.0,"minimum":1817395200,"p95":1863409664.0,"samples":108},"process_sample_count":108,"sampled_process_counter_deltas":{"deltas":{"major_faults":0,"minor_faults":48,"read_syscall_bytes":0,"storage_read_bytes":0,"storage_write_bytes":69632},"meaning":"Deltas between sampled endpoints within each interval; excludes unsampled edges. Storage reads are not all memory reads; rchar includes cached reads.","observed_endpoint_span_seconds":104.75946743803797,"observed_intervals":6,"requested_intervals":6},"task_cgroup_current_bytes":{"maximum":86958448640,"mean":86929725933.03703,"median":86926063616.0,"minimum":86919168000,"p95":86944211353.6,"samples":108}},"completed_synthetic_rounds":6,"finished_at":"2026-09-27T08:44:58.840689+00:00","id":"qwen72-q8","maximum_gpu_sample_gap_seconds":1.0098843040177599,"model":{"architecture":"qwen2","key":"qwen25_72b_q8_0","license":"Qwen","quantization":"Q8_0","repository":"Qwen/Qwen2.5-72B-Instruct-GGUF","revision":"7ca3bc388f97b264c4283bc9bf1055e2abc38441","weight_bytes":77530000800,"weight_sha256":["31a2b3c4e62dc185b54b9c33b1b2898f06fde167e3311cc5380e08692098e781","747e66c38a1f4f8d496533e2108c7592e10220afc65c480d13cbe14d0109d700","4346236ce045e136ad4cbcd7b86a88ae66dfadeed033175965e9f331a76e90d9","152f922d2dd4c18d49f6b3c00bb1cc20629ed042d494527e1c350da41043a446","24d6a097d3d3ce727b1f123957713cdaba0d5d64b9aaf559afb4703ea51ce577","4ef22906aa68f1b7c067c4469677d7490455a9ea1a2fcda67af29bb7d1ffc49e","b9dd79952b0a7d771633f32c253da0b771eb09efe173eeef50c250979c5ddd22","af925e8985c7d5e30d6ee98e275d93cdc2ca5777173d5601ebc9d6fbd4f4d01e","057a909a1bfd954aaf9b4b94b226b98a1a7b2a5b19861a708bf2879bb12c704e","c84a7441c2feab7213ee35b989334079276701a4bc39017e68ac7096966a08b6","48f91fc504df34b89edc7d709925b9b69900a3a3671b1b4ce8d7384ff3117255","84c3b6949803166b84bb8ffda810fbf2f56aec4e32b9e0f1614a43f84d4214a2","00fd616480122ef2fddf3888436ee62d04ec5dc081cad0f8d0db9e7f1b241faf","58a93e4a25a3bd34ae4327f4b40d1fd7c90b20ae34ed29a5fc0804333031d998","f3b643f7b02f52da7a68ab0eee711ba160904ed235bea64da6cca5b46bb5db48","721536952fea2b497e74681da2c65b1bd783ef2f3193ab41057fcaa31a64b495","3b21b0bd05319458227e5865d1a79e62e1d63168bf8060e8d0d6f1e0da51c324","1804cb39eefead3acd09ffafd064d969ec07161aceb62edbbe8975a0472391eb","a6c8e5f402cb84f8d29813f811cb93c3a694dfc8a6398e2081031ced3a49b5ad","047d7b4c5c363d17e86f19b3bd66e38663cb1484555aa897cd4892c1993881d3","b6a7a5be990bed3f18695ded8d8fca3f545176045a37784e0443e2bd6633c6bc"]},"observed_thermal_slowdown":false,"phase":"capacity","placement":{"meaning":"Runtime allocations; CPU expert placement is requested separately. Not measured resident RAM.","model_buffer_allocations":[{"device":"CPU_Mapped","reported_mib":1262.25},{"device":"CUDA0","reported_mib":72670.41}],"offloaded_layers":81,"requested_cpu_expert_layers":0,"total_layers":81},"practical_check_scope":"Three authored protocol/final-answer checks, not a general model quality benchmark.","practical_checks":[{"all_completion_tokens_per_response_second":17.322248027429783,"completion_tokens_including_reasoning":9,"exact_expected_answer":true,"expected_final_answer_valid":true,"finish_reason":"stop","protocol_completed":true,"response_seconds":0.5195630489615723,"strict_json_object_valid":true,"task":"arithmetic","time_to_first_generated_event_seconds":0.08058539894409478,"time_to_first_visible_answer_seconds":0.08058539894409478,"usable_final_answer":true},{"all_completion_tokens_per_response_second":17.735107171643207,"completion_tokens_including_reasoning":23,"exact_expected_answer":false,"expected_final_answer_valid":false,"finish_reason":"stop","protocol_completed":true,"response_seconds":1.2968627579975873,"strict_json_object_valid":false,"task":"extraction","time_to_first_generated_event_seconds":0.08757721597794443,"time_to_first_visible_answer_seconds":0.08757721597794443,"usable_final_answer":true},{"all_completion_tokens_per_response_second":17.421902684561562,"completion_tokens_including_reasoning":14,"exact_expected_answer":false,"expected_final_answer_valid":false,"finish_reason":"stop","protocol_completed":true,"response_seconds":0.8035861669923179,"strict_json_object_valid":false,"task":"classification","time_to_first_generated_event_seconds":0.088394224003423,"time_to_first_visible_answer_seconds":0.088394224003423,"usable_final_answer":true}],"record_sha256":"81835d12b4546153652824e253a63fe3ea0ae9d559d4d09f2b97a5c16b7026b7","settings":{"batch":2048,"cache_prompt":false,"cache_ram_bytes":0,"concurrency":1,"context_shift":false,"cpu_expert_layers":0,"enable_thinking":null,"fit":"off","flash_attention":"on","generation_tokens":256,"kv_cache":"f16","lazy_mode":"off","load_mode":"mmap","mlock":false,"practical_max_output_tokens":1024,"practical_tasks":3,"prompt_token_counts":[2048,8192],"reasoning_effort":null,"repetitions":3,"requested_gpu_layers":999,"separate_warmup_requests_per_context":1,"server_slots":1,"socket_timeout_policy":"remaining_case_deadline","threads":16,"total_context_tokens":8704,"ubatch":512},"status":"completed","stop_category":null,"synthetic_groups":[{"completed_rounds":3,"context_tokens":2048,"end_to_end_tokens_per_second":{"maximum":16.6722186413895,"mean":16.5289930393237,"median":16.49119470865624,"minimum":16.423565767925368,"p95":16.65411624811617,"samples":3},"generated_tokens":768,"process_major_faults_total":0,"process_percent_host_capacity":{"maximum":3.1234493242837957,"mean":3.1225222297770268,"median":3.122222020569965,"minimum":3.12189534447732,"p95":3.1233265939124126,"samples":3},"response_seconds":{"maximum":15.587358044984285,"mean":15.488559569329178,"median":15.5234356590081,"minimum":15.35488500399515,"p95":15.580965806386667,"samples":3},"server_decode_tokens_per_second":{"maximum":18.580382962823496,"mean":18.399532164867054,"median":18.384412440667013,"minimum":18.233801091110657,"p95":18.560785910607848,"samples":3},"status":"completed","synthetic_first_token_seconds":{"maximum":1.652893572987523,"mean":1.6285955609831337,"median":1.630639548995532,"minimum":1.6022535609663464,"p95":1.650668170588324,"samples":3}},{"completed_rounds":3,"context_tokens":8192,"end_to_end_tokens_per_second":{"maximum":12.10368645258993,"mean":12.051843439859258,"median":12.06526491767696,"minimum":11.986578949310886,"p95":12.099844299098633,"samples":3},"generated_tokens":768,"process_major_faults_total":0,"process_percent_host_capacity":{"maximum":3.122425625764185,"mean":3.1217200642319645,"median":3.1219041196784714,"minimum":3.1208304472532364,"p95":3.1223734751556136,"samples":3},"response_seconds":{"maximum":21.357219694007654,"mean":21.241911690992612,"median":21.217934437969234,"minimum":21.150580941000953,"p95":21.343291168403812,"samples":3},"server_decode_tokens_per_second":{"maximum":17.82214505660523,"mean":17.738073185418425,"median":17.75457166295167,"minimum":17.637502836698374,"p95":17.815387717239872,"samples":3},"status":"completed","synthetic_first_token_seconds":{"maximum":6.899148746044375,"mean":6.8663084533375995,"median":6.8574764329823665,"minimum":6.842300180986058,"p95":6.894981514738174,"samples":3}}],"synthetic_response_sample_coverage":{"meaning":"Union of nominal one-second sample supports clipped to completed synthetic request intervals; not continuous utilization.","requested_response_seconds":110.19285684998613,"sample_supported_fraction":0.9990995831491478,"sample_supported_seconds":110.09363734483486},"synthetic_rounds":[{"context_tokens":2048,"cpu_snapshot_interval":{"host_busy_percent":3.2646083080617916,"interval_seconds":15.355255289003253,"process_cpu_seconds":15.34,"process_major_faults":0,"process_minor_faults":73,"process_percent_host_capacity":3.12189534447732,"process_percent_one_logical_cpu":99.90065102327424},"end_to_end_tokens_per_second":16.6722186413895,"finished_at":"2026-09-27T08:43:00.162732+00:00","generated_tokens":256,"request_wall_seconds":15.355112514982466,"response_seconds":15.35488500399515,"round":0,"server_decode_tokens_per_second":18.580382962823496,"started_at":"2026-09-27T08:42:44.807479+00:00","synthetic_first_token_seconds":1.630639548995532},{"context_tokens":2048,"cpu_snapshot_interval":{"host_busy_percent":3.263276563192012,"interval_seconds":15.523799935006537,"process_cpu_seconds":15.51,"process_major_faults":0,"process_minor_faults":71,"process_percent_host_capacity":3.122222020569965,"process_percent_one_logical_cpu":99.91110465823888},"end_to_end_tokens_per_second":16.49119470865624,"finished_at":"2026-09-27T08:43:15.693852+00:00","generated_tokens":256,"request_wall_seconds":15.523666316992603,"response_seconds":15.5234356590081,"round":1,"server_decode_tokens_per_second":18.384412440667013,"started_at":"2026-09-27T08:43:00.170021+00:00","synthetic_first_token_seconds":1.652893572987523},{"context_tokens":2048,"cpu_snapshot_interval":{"host_busy_percent":3.2659689642728256,"interval_seconds":15.587734887027182,"process_cpu_seconds":15.58,"process_major_faults":0,"process_minor_faults":71,"process_percent_host_capacity":3.1234493242837957,"process_percent_one_logical_cpu":99.95037837708146},"end_to_end_tokens_per_second":16.423565767925368,"finished_at":"2026-09-27T08:43:31.288148+00:00","generated_tokens":256,"request_wall_seconds":15.587601607956458,"response_seconds":15.587358044984285,"round":2,"server_decode_tokens_per_second":18.233801091110657,"started_at":"2026-09-27T08:43:15.700412+00:00","synthetic_first_token_seconds":1.6022535609663464},{"context_tokens":8192,"cpu_snapshot_interval":{"host_busy_percent":3.259504425301792,"interval_seconds":21.15095386299072,"process_cpu_seconds":21.13,"process_major_faults":0,"process_minor_faults":175,"process_percent_host_capacity":3.1219041196784714,"process_percent_one_logical_cpu":99.90093182971108},"end_to_end_tokens_per_second":12.10368645258993,"finished_at":"2026-09-27T08:44:13.434428+00:00","generated_tokens":256,"request_wall_seconds":21.150823330972344,"response_seconds":21.150580941000953,"round":0,"server_decode_tokens_per_second":17.82214505660523,"started_at":"2026-09-27T08:43:52.283473+00:00","synthetic_first_token_seconds":6.842300180986058},{"context_tokens":8192,"cpu_snapshot_interval":{"host_busy_percent":3.259540747941614,"interval_seconds":21.218310676980764,"process_cpu_seconds":21.19,"process_major_faults":0,"process_minor_faults":166,"process_percent_host_capacity":3.1208304472532364,"process_percent_one_logical_cpu":99.86657431210357},"end_to_end_tokens_per_second":12.06526491767696,"finished_at":"2026-09-27T08:44:34.659226+00:00","generated_tokens":256,"request_wall_seconds":21.218187698046677,"response_seconds":21.217934437969234,"round":1,"server_decode_tokens_per_second":17.75457166295167,"started_at":"2026-09-27T08:44:13.440905+00:00","synthetic_first_token_seconds":6.8574764329823665},{"context_tokens":8192,"cpu_snapshot_interval":{"host_busy_percent":3.257195534159582,"interval_seconds":21.35759438102832,"process_cpu_seconds":21.34,"process_major_faults":0,"process_minor_faults":167,"process_percent_host_capacity":3.122425625764185,"process_percent_one_logical_cpu":99.91762002445392},"end_to_end_tokens_per_second":11.986578949310886,"finished_at":"2026-09-27T08:44:56.023499+00:00","generated_tokens":256,"request_wall_seconds":21.35746538103558,"response_seconds":21.357219694007654,"round":2,"server_decode_tokens_per_second":17.637502836698374,"started_at":"2026-09-27T08:44:34.665885+00:00","synthetic_first_token_seconds":6.899148746044375}],"thermal_states_fully_available":true,"whole_case_telemetry":{"gpu":{"gpu_board_power_w":{"maximum":301.6,"mean":244.41552083333335,"median":299.985,"minimum":15.92,"p95":300.07,"samples":192},"gpu_memory_used_mib":{"maximum":76294.0,"mean":63493.333333333336,"median":76294.0,"minimum":2.0,"p95":76294.0,"samples":192},"gpu_sm_clock_mhz":{"maximum":2310.0,"mean":1642.4895833333333,"median":1695.0,"minimum":180.0,"p95":2293.4999999999995,"samples":192},"gpu_utilization_percent":{"maximum":100.0,"mean":81.47395833333333,"median":100.0,"minimum":0.0,"p95":100.0,"samples":192}},"gpu_sample_count":192,"host_mem_available_bytes":{"maximum":96520880128,"mean":96079422118.89503,"median":96071413760,"minimum":96034291712,"p95":96161054720.0,"samples":181},"host_sample_count":181,"meaning":"Whole monitored case, including model loading, warmup, synthetic requests, practical chat and cleanup.","model_process_pss_bytes":{"maximum":77868674048,"mean":9987741787.022223,"median":1853541376.0,"minimum":1704188928,"p95":66537673523.19995,"samples":180},"model_process_rss_bytes":{"maximum":77871632384,"mean":9990731889.777779,"median":1856520192.0,"minimum":1707147264,"p95":66540631859.19995,"samples":180},"process_sample_count":180,"sampled_process_counter_deltas":{"deltas":{"major_faults":8,"minor_faults":1452716,"read_syscall_bytes":13,"storage_read_bytes":73981358080,"storage_write_bytes":131072},"meaning":"Deltas between sampled endpoints within each interval; excludes unsampled edges. Storage reads are not all memory reads; rchar includes cached reads.","observed_endpoint_span_seconds":190.3435920070042,"observed_intervals":1,"requested_intervals":1},"task_cgroup_current_bytes":{"maximum":86972829696,"mean":86935634349.96686,"median":86938062848,"minimum":86367371264,"p95":86966538240.0,"samples":181}}},{"case_wall_seconds_including_cooldown_load_warmup_cleanup":1095.0899538400117,"completed_synthetic_response_telemetry":{"gpu":{"gpu_board_power_w":{"maximum":134.7,"mean":97.61186311787073,"median":93.71,"minimum":87.61,"p95":127.33500000000001,"samples":526},"gpu_memory_used_mib":{"maximum":89796.0,"mean":89795.20152091255,"median":89796.0,"minimum":89794.0,"p95":89796.0,"samples":526},"gpu_sm_clock_mhz":{"maximum":2317.0,"mean":2283.790874524715,"median":2280.0,"minimum":1875.0,"p95":2317.0,"samples":526},"gpu_utilization_percent":{"maximum":100.0,"mean":34.351711026615966,"median":10.0,"minimum":2.0,"p95":85.0,"samples":526}},"gpu_sample_count":526,"host_mem_available_bytes":{"maximum":95776518144,"mean":95715279226.98932,"median":95695937536,"minimum":95666692096,"p95":95770460160.0,"samples":281},"host_sample_count":281,"meaning":"Samples inside the union of completed synthetic response intervals; excludes loading, warmup and practical chat.","model_process_pss_bytes":{"maximum":55334830080,"mean":55145110524.35587,"median":55333445632,"minimum":54833031168,"p95":55334830080.0,"samples":281},"model_process_rss_bytes":{"maximum":55337811968,"mean":55148146323.58719,"median":55336439808,"minimum":54837063680,"p95":55337811968.0,"samples":281},"process_sample_count":281,"sampled_process_counter_deltas":{"deltas":{"major_faults":0,"minor_faults":98,"read_syscall_bytes":0,"storage_read_bytes":0,"storage_write_bytes":229376},"meaning":"Deltas between sampled endpoints within each interval; excludes unsampled edges. Storage reads are not all memory reads; rchar includes cached reads.","observed_endpoint_span_seconds":519.6157901960541,"observed_intervals":6,"requested_intervals":6},"task_cgroup_current_bytes":{"maximum":86954258432,"mean":86921684736.91103,"median":86911365120,"minimum":86908473344,"p95":86944989184.0,"samples":281}},"completed_synthetic_rounds":6,"finished_at":"2026-09-27T09:03:13.936137+00:00","id":"qwen235-hybrid","maximum_gpu_sample_gap_seconds":1.5674057049909607,"model":{"architecture":"qwen3moe","key":"qwen3_235b_q5_k_m","license":"Apache-2.0","quantization":"Q5_K_M","repository":"Qwen/Qwen3-235B-A22B-GGUF","revision":"211e807ecd6d37446a747409108783e39a04e80f","weight_bytes":166814156992,"weight_sha256":["706382f26d0d6f72cd89c9ac1bfeb40d4b4afaee986dad2ea229ea24f9168fab","9ab54360c3809c580439b75fe7e16f3af24adcc701c352e39b06ad557aca0ca3","93c863b35e810b97605dc973c2881e608751a269e773b3772c42e5e25cb01d79","c0322fa3f66ca392a9d7f95bb2543b6f6213ac04c16e613a2ae3a82429452071","1dce96f79ba9c2d86d36bbd856101f7a2e4f3c4cff4ef4a2a970c0ba336edb15","15a2192eadc2cc2da6fc0df6d25cb6d39626f292716526cfee8baba9043fbe62"]},"observed_thermal_slowdown":false,"phase":"capacity","placement":{"meaning":"Runtime allocations; CPU expert placement is requested separately. Not measured resident RAM.","model_buffer_allocations":[{"device":"CPU_Mapped","reported_mib":27759.73},{"device":"CPU_Mapped","reported_mib":28323.86},{"device":"CPU_Mapped","reported_mib":18322.9},{"device":"CUDA0","reported_mib":86732.62}],"offloaded_layers":95,"requested_cpu_expert_layers":44,"total_layers":95},"practical_check_scope":"Three authored protocol/final-answer checks, not a general model quality benchmark.","practical_checks":[{"all_completion_tokens_per_response_second":1.7231570766526845,"completion_tokens_including_reasoning":13,"exact_expected_answer":false,"expected_final_answer_valid":false,"finish_reason":"stop","protocol_completed":true,"response_seconds":7.544291914033238,"strict_json_object_valid":false,"task":"arithmetic","time_to_first_generated_event_seconds":4.4448636720189825,"time_to_first_visible_answer_seconds":4.4448636720189825,"usable_final_answer":true},{"all_completion_tokens_per_response_second":1.8127850842961986,"completion_tokens_including_reasoning":19,"exact_expected_answer":true,"expected_final_answer_valid":true,"finish_reason":"stop","protocol_completed":true,"response_seconds":10.481110069027636,"strict_json_object_valid":true,"task":"extraction","time_to_first_generated_event_seconds":6.458732172031887,"time_to_first_visible_answer_seconds":6.458732172031887,"usable_final_answer":true},{"all_completion_tokens_per_response_second":1.563542784092261,"completion_tokens_including_reasoning":14,"exact_expected_answer":false,"expected_final_answer_valid":false,"finish_reason":"stop","protocol_completed":true,"response_seconds":8.954024247010238,"strict_json_object_valid":false,"task":"classification","time_to_first_generated_event_seconds":6.181021316035185,"time_to_first_visible_answer_seconds":6.181021316035185,"usable_final_answer":true}],"record_sha256":"b4596ba0b10a895ef31922bf705fe96f452ddcf91a94761ce31ee2d94bdc656c","settings":{"batch":2048,"cache_prompt":false,"cache_ram_bytes":0,"concurrency":1,"context_shift":false,"cpu_expert_layers":44,"enable_thinking":false,"fit":"off","flash_attention":"on","generation_tokens":256,"kv_cache":"f16","lazy_mode":"off","load_mode":"mmap","mlock":false,"practical_max_output_tokens":1024,"practical_tasks":3,"prompt_token_counts":[2048,8192],"reasoning_effort":null,"repetitions":3,"requested_gpu_layers":999,"separate_warmup_requests_per_context":1,"server_slots":1,"socket_timeout_policy":"remaining_case_deadline","threads":16,"total_context_tokens":8704,"ubatch":512},"status":"completed","stop_category":null,"synthetic_groups":[{"completed_rounds":3,"context_tokens":2048,"end_to_end_tokens_per_second":{"maximum":3.6287374041073703,"mean":3.622629696413158,"median":3.626357141568686,"minimum":3.6127945435634174,"p95":3.628499377853502,"samples":3},"generated_tokens":768,"process_major_faults_total":0,"process_percent_host_capacity":{"maximum":38.14443102122907,"mean":38.142436051427985,"median":38.14418506796768,"minimum":38.138692065087206,"p95":38.14440642590293,"samples":3},"response_seconds":{"maximum":70.85927442403045,"mean":70.6671629580087,"median":70.59426030202303,"minimum":70.54795414797263,"p95":70.8327730118297,"samples":3},"server_decode_tokens_per_second":{"maximum":4.720220848582423,"mean":4.71217984918084,"median":4.717140377010508,"minimum":4.6991783219495895,"p95":4.719912801425232,"samples":3},"status":"completed","synthetic_first_token_seconds":{"maximum":16.535908652993385,"mean":16.505473350000102,"median":16.49099237000337,"minimum":16.489519027003553,"p95":16.531417024694385,"samples":3}},{"completed_rounds":3,"context_tokens":8192,"end_to_end_tokens_per_second":{"maximum":2.426304310649536,"mean":2.424460530643768,"median":2.424830014883551,"minimum":2.4222472663982164,"p95":2.4261568810729375,"samples":3},"generated_tokens":768,"process_major_faults_total":0,"process_percent_host_capacity":{"maximum":26.599649492938177,"mean":26.576182672773356,"median":26.575179343665628,"minimum":26.553719181716264,"p95":26.59720247801092,"samples":3},"response_seconds":{"maximum":105.68698066100478,"mean":105.59055065635282,"median":105.57441075402312,"minimum":105.51026055403054,"p95":105.67572367030661,"samples":3},"server_decode_tokens_per_second":{"maximum":4.716323324493772,"mean":4.710968167669336,"median":4.714649186004725,"minimum":4.701931992509509,"p95":4.716155910644868,"samples":3},"status":"completed","synthetic_first_token_seconds":{"maximum":51.50652379600797,"mean":51.46113450233437,"median":51.45365717000095,"minimum":51.423222540994175,"p95":51.501237133407265,"samples":3}}],"synthetic_response_sample_coverage":{"meaning":"Union of nominal one-second sample supports clipped to completed synthetic request intervals; not continuous utilization.","requested_response_seconds":528.7750799870118,"sample_supported_fraction":0.9942198682370065,"sample_supported_seconds":525.7186903516995},"synthetic_rounds":[{"context_tokens":2048,"cpu_snapshot_interval":{"host_busy_percent":40.01638146680539,"interval_seconds":70.5948035120382,"process_cpu_seconds":861.69,"process_major_faults":0,"process_minor_faults":86,"process_percent_host_capacity":38.14418506796768,"process_percent_one_logical_cpu":1220.6139221749659},"end_to_end_tokens_per_second":3.626357141568686,"finished_at":"2026-09-27T08:53:21.042546+00:00","generated_tokens":256,"request_wall_seconds":70.59456794400467,"response_seconds":70.59426030202303,"round":0,"server_decode_tokens_per_second":4.717140377010508,"started_at":"2026-09-27T08:52:10.447728+00:00","synthetic_first_token_seconds":16.535908652993385},{"context_tokens":2048,"cpu_snapshot_interval":{"host_busy_percent":39.948658233208064,"interval_seconds":70.85978680598782,"process_cpu_seconds":864.93,"process_major_faults":0,"process_minor_faults":75,"process_percent_host_capacity":38.14443102122907,"process_percent_one_logical_cpu":1220.6217926793302},"end_to_end_tokens_per_second":3.6127945435634174,"finished_at":"2026-09-27T08:54:32.074858+00:00","generated_tokens":256,"request_wall_seconds":70.85964214702835,"response_seconds":70.85927442403045,"round":1,"server_decode_tokens_per_second":4.6991783219495895,"started_at":"2026-09-27T08:53:21.215055+00:00","synthetic_first_token_seconds":16.49099237000337},{"context_tokens":2048,"cpu_snapshot_interval":{"host_busy_percent":40.009658924757865,"interval_seconds":70.54843399999663,"process_cpu_seconds":861.0,"process_major_faults":0,"process_minor_faults":89,"process_percent_host_capacity":38.138692065087206,"process_percent_one_logical_cpu":1220.4381460827906},"end_to_end_tokens_per_second":3.6287374041073703,"finished_at":"2026-09-27T08:55:42.796420+00:00","generated_tokens":256,"request_wall_seconds":70.54821964097209,"response_seconds":70.54795414797263,"round":2,"server_decode_tokens_per_second":4.720220848582423,"started_at":"2026-09-27T08:54:32.247973+00:00","synthetic_first_token_seconds":16.489519027003553},{"context_tokens":8192,"cpu_snapshot_interval":{"host_busy_percent":28.195178017589956,"interval_seconds":105.51081193995196,"process_cpu_seconds":897.27,"process_major_faults":0,"process_minor_faults":168,"process_percent_host_capacity":26.575179343665628,"process_percent_one_logical_cpu":850.4057389973001},"end_to_end_tokens_per_second":2.426304310649536,"finished_at":"2026-09-27T08:59:14.195160+00:00","generated_tokens":256,"request_wall_seconds":105.51060808997136,"response_seconds":105.51026055403054,"round":0,"server_decode_tokens_per_second":4.714649186004725,"started_at":"2026-09-27T08:57:28.684337+00:00","synthetic_first_token_seconds":51.423222540994175},{"context_tokens":8192,"cpu_snapshot_interval":{"host_busy_percent":28.238400960511957,"interval_seconds":105.6874828650034,"process_cpu_seconds":899.6,"process_major_faults":0,"process_minor_faults":186,"process_percent_host_capacity":26.599649492938177,"process_percent_one_logical_cpu":851.1887837740217},"end_to_end_tokens_per_second":2.4222472663982164,"finished_at":"2026-09-27T09:01:00.054139+00:00","generated_tokens":256,"request_wall_seconds":105.68732000200544,"response_seconds":105.68698066100478,"round":1,"server_decode_tokens_per_second":4.701931992509509,"started_at":"2026-09-27T08:59:14.366636+00:00","synthetic_first_token_seconds":51.45365717000095},{"context_tokens":8192,"cpu_snapshot_interval":{"host_busy_percent":28.177641200537625,"interval_seconds":105.57490010402398,"process_cpu_seconds":897.09,"process_major_faults":0,"process_minor_faults":169,"process_percent_host_capacity":26.553719181716264,"process_percent_one_logical_cpu":849.7190138149205},"end_to_end_tokens_per_second":2.424830014883551,"finished_at":"2026-09-27T09:02:45.800741+00:00","generated_tokens":256,"request_wall_seconds":105.57472216302995,"response_seconds":105.57441075402312,"round":2,"server_decode_tokens_per_second":4.716323324493772,"started_at":"2026-09-27T09:01:00.225815+00:00","synthetic_first_token_seconds":51.50652379600797}],"thermal_states_fully_available":true,"whole_case_telemetry":{"gpu":{"gpu_board_power_w":{"maximum":134.7,"mean":90.0707934655776,"median":93.15,"minimum":15.45,"p95":126.55199999999999,"samples":857},"gpu_memory_used_mib":{"maximum":89796.0,"mean":82403.28588098017,"median":89794.0,"minimum":2.0,"p95":89796.0,"samples":857},"gpu_sm_clock_mhz":{"maximum":2325.0,"mean":2155.6266044340723,"median":2280.0,"minimum":172.0,"p95":2317.0,"samples":857},"gpu_utilization_percent":{"maximum":100.0,"mean":29.7024504084014,"median":10.0,"minimum":0.0,"p95":84.0,"samples":857}},"gpu_sample_count":857,"host_mem_available_bytes":{"maximum":96489607168,"mean":95777516114.58064,"median":95719641088.0,"minimum":95666692096,"p95":96130357248.0,"samples":496},"host_sample_count":496,"meaning":"Whole monitored case, including model loading, warmup, synthetic requests, practical chat and cleanup.","model_process_pss_bytes":{"maximum":86505224192,"mean":56533773838.510124,"median":55333082112.0,"minimum":1423906816,"p95":86110927257.6,"samples":494},"model_process_rss_bytes":{"maximum":86508187648,"mean":56536806462.18623,"median":55336427520.0,"minimum":1426870272,"p95":86113890713.6,"samples":494},"process_sample_count":494,"sampled_process_counter_deltas":{"deltas":{"major_faults":150207,"minor_faults":5316091,"read_syscall_bytes":13,"storage_read_bytes":286047768576,"storage_write_bytes":352256},"meaning":"Deltas between sampled endpoints within each interval; excludes unsampled edges. Storage reads are not all memory reads; rchar includes cached reads.","observed_endpoint_span_seconds":854.6193209009944,"observed_intervals":1,"requested_intervals":1},"task_cgroup_current_bytes":{"maximum":86973087744,"mean":86917490621.93549,"median":86936125440.0,"minimum":80896401408,"p95":86969487360.0,"samples":496}}}],"completed_case_count":6,"finished_at":"2026-09-27T09:05:28.971549+00:00","hardware":{"cpu":["AMD EPYC 4565P 16-Core Processor"],"driver":"595.91.07","gpu":"NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition","gpu_power_limit_w":300,"gpu_reported_memory_mib":97887.0,"kernel":"6.12.0-211.56.1.el10_2.0.1.x86_64","linux_visible_ram_bytes":100271353856,"logical_cpus":32,"physical_system_ram_gb":96,"planned_system_ram_gb_not_measured":192},"historical_results_pooled":false,"kind":"capacity_offload_campaign","limitations":["Synthetic /completion first-token latency is not chat final-answer latency, including for GPT-OSS.","Server decode throughput follows llama.cpp b11146: 255 decode steps for 256 output tokens; the first output token comes from prefill. End-to-end throughput counts all 256 output tokens.","Practical chat separately measures first generated content and first visible final-answer content; generated usage can include reasoning.","GPU board power is not whole-server wall-socket power. Sampled telemetry can miss events between samples.","Partial CPU offload and MoE expert offload are distinct; active parameters do not describe total model storage.","Integrity reads precede loading. No cold-cache claim; mmap file reads can occur even without swap.","A successful bounded run does not prove long-duration stability or broad model quality.","A missing case or harness stop is not proof of GPU failure or that the model cannot fit.","Physical measured system RAM is 96 GB; the offered 192 GB configuration was not measured."],"models_manifest_sha256":"f8e715d97450a7120e9d32aa9eb4d50996ec769c0fad9ba9c1cb5ab67bede09b","plan_sha256":"4c4a17de3e1f76e9cf66873b3268855214359059c5ccacaaf48e6af97fee3e4c","planned_case_count":6,"record_sha256":"b55ccee2f73023cf554fcad4745cadba3ebf50274edddb1346842f6163753699","runner_sha256":"c01ffbddca9cfc65f5dab27529c93166816a73c0cdbefb53d1b9f95adc844166","schema":1,"started_at":"2026-09-27T06:59:01.501050+00:00","status":"completed","stop_category":null,"tool":{"commit":"7fe450e19305b828c199d602c23a8337aaa1f03b","name":"llama.cpp","release":"b11146","version":"0.5.0-dev"}}