LOCALAIRECIPES / BENCHMARK RUN REPORT
spot-check llm-inference-bench 0.8.0 exllamav3 0f685aa + row32 four-slot profile layout: tp ×TP2 on 2x CMP170HX 64GiB, 4 active sequences GLM-5.3-Flash
attached to /r/glm-5-3-flash-low-effort-on-2x-cmp170hx-exl3-tp2-four-slots variant: GLM-5.3-Flash
Uploaded report
report.md (markdown, shown as-is)
Re-upload of run 4140365f3b3be8f4a92d60ea6fc957d3fbcd4872 with the headline fields the site reads (one agent, short prompt, 78 input tokens); every measurement is unchanged. # GLM-5.3-Flash, low effort, on 2x CMP170HX Four native slots; EXL3 3.05bpw with the Weightless GLP44 steering vector at alpha 2, GPU-resident Q8 KV. 1, 2 and 4 agents at short and nominal 32K/128K context, one 30-second run per cell. Exact completion-token usage, including reasoning; actual input lengths are retained. A separate four-request 512-token completion cohort. Inference speed only, not accuracy or intelligence. Differences between efforts are not separated from run-to-run variation. Full methods, tables and caveats: https://github.com/alexvnesta/glm53-flash-cmp170hx-tp/blob/fa0adf1899015f692ddce30391ee7e740a0910c9/docs/COMMUNITY_BENCHMARK_2026-10-17.md
Run file
raw JSON (verbatim; gpu uuids + private hosts masked)
{
"tier": "spot-check",
"tool": "llm-inference-bench",
"model": "GLM-5.3-Flash",
"driver": "610.57.04",
"engine": "exllamav3",
"median": 41.72178474486878485549823381006717681884765625,
"metrics": {
"ttft_ms": 325.6752750021405518054962158203125,
"decode_tok_s": 41.72178474486878485549823381006717681884765625
},
"results": [
{
"ttft_avg": 0.3256752750021405518054962158203125,
"ttft_p50": 0.3256752750021405518054962158203125,
"ttft_p90": 0.3256752750021405518054962158203125,
"ttft_p99": 0.3256752750021405518054962158203125,
"wall_time": 35.5309422709979116916656494140625,
"num_errors": 0,
"concurrency": 1,
"underfilled": false,
"ready_reason": "metrics_unavailable, active_streams=1/1, stable=3.0s",
"total_tokens": 1250,
"aggregate_tps": 41.72178474486878485549823381006717681884765625,
"loop_detected": false,
"num_completed": 1,
"request_count": 1,
"avg_queue_reqs": 0,
"benchmark_mode": "duration",
"context_tokens": 0,
"failure_reason": "",
"max_queue_reqs": 0,
"queue_fraction": 0,
"timeout_reason": "",
"accept_norm_tps": 0,
"request_samples": [
{
"ttft": 0.3256752750021405518054962158203125,
"latency": 0,
"completed": false,
"input_tokens": 78,
"output_tokens": 1504,
"output_tps_per_user": 42.76595847924937743300688453018665313720703125,
"time_to_second_token": 0.057417034986428916454315185546875,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.02338308401260814906041929361890652216970920562744140625,
"chunk_inter_token_latency_avg": 0.023476803788209786583163207751567824743688106536865234375
}
],
"warmup_duration": 5.5099999999999997868371792719699442386627197265625,
"aggregate_source": "openai_continuous_usage",
"avg_running_reqs": 0,
"capacity_limited": false,
"hardware_summary": {
"samples": 14,
"gpu_count": 2,
"temp_avg_c": 50.96000000000000085265128291212022304534912109375,
"temp_max_c": 54,
"vram_total_mb": 131072,
"cpu_temp_max_c": 65.25,
"cpu_util_avg_pct": 10.1699999999999999289457264239899814128875732421875,
"duration_seconds": 28.693000000000001392663762089796364307403564453125,
"gpu_util_avg_pct": 87.6400000000000005684341886080801486968994140625,
"gpu_util_max_pct": 94,
"mem_util_avg_pct": 14.6400000000000005684341886080801486968994140625,
"mem_util_max_pct": 16,
"pcie_rx_avg_mb_s": 509.5,
"pcie_rx_max_mb_s": 603,
"pcie_tx_avg_mb_s": 455.79000000000002046363078989088535308837890625,
"pcie_tx_max_mb_s": 545,
"vram_used_avg_mb": 126720,
"vram_used_max_mb": 126720,
"power_total_avg_w": 335.26999999999998181010596454143524169921875,
"power_total_max_w": 363.720000000000027284841053187847137451171875,
"vram_used_avg_pct": 96.68000000000000682121026329696178436279296875,
"vram_used_max_pct": 96.68000000000000682121026329696178436279296875,
"power_limit_total_w": 360
},
"loop_diagnostics": [],
"max_running_reqs": 0,
"warmup_timed_out": false,
"input_seq_len_avg": 78,
"output_seq_len_avg": 1504,
"output_seq_len_p50": 1504,
"output_seq_len_p90": 1504,
"output_seq_len_p99": 1504,
"server_spec_drafts": 0,
"server_steps_per_s": 0,
"server_utilization": 0,
"accept_norm_ref_len": 0,
"measurement_seconds": 29.960367000000001524995241197757422924041748046875,
"per_request_avg_tps": 41.72178474486878485549823381006717681884765625,
"request_latency_avg": 0,
"request_latency_p50": 0,
"request_latency_p90": 0,
"request_latency_p99": 0,
"server_engine_steps": 0,
"client_output_tokens": 1250,
"request_count_target": 0,
"server_output_tokens": 0,
"warmup_request_count": 0,
"effective_concurrency": 0,
"server_gen_throughput": 0,
"loop_detection_enabled": true,
"server_spec_pos_accept": [],
"completed_request_count": 0,
"inter_token_latency_avg": 0.02338308401260814906041929361890652216970920562744140625,
"inter_token_latency_p50": 0.02338308401260814906041929361890652216970920562744140625,
"inter_token_latency_p90": 0.02338308401260814906041929361890652216970920562744140625,
"inter_token_latency_p99": 0.02338308401260814906041929361890652216970920562744140625,
"output_tps_per_user_avg": 42.76595847924937743300688453018665313720703125,
"output_tps_per_user_p50": 42.76595847924937743300688453018665313720703125,
"output_tps_per_user_p90": 42.76595847924937743300688453018665313720703125,
"output_tps_per_user_p99": 42.76595847924937743300688453018665313720703125,
"server_spec_accept_rate": 0,
"loop_diagnostics_omitted": 0,
"measurement_wall_seconds": 30.0005910000000000081854523159563541412353515625,
"server_spec_draft_tokens": 0,
"time_to_second_token_avg": 0.057417034986428916454315185546875,
"time_to_second_token_p50": 0.057417034986428916454315185546875,
"time_to_second_token_p90": 0.057417034986428916454315185546875,
"time_to_second_token_p99": 0.057417034986428916454315185546875,
"server_spec_accept_length": 0,
"e2e_output_tps_per_user_avg": 0,
"e2e_output_tps_per_user_p50": 0,
"e2e_output_tps_per_user_p90": 0,
"e2e_output_tps_per_user_p99": 0,
"server_accept_len_effective": 0,
"server_spec_accepted_tokens": 0,
"chunk_inter_token_latency_avg": 0.023476803788209786583163207751567824743688106536865234375,
"chunk_inter_token_latency_p50": 0.023476803788209786583163207751567824743688106536865234375,
"chunk_inter_token_latency_p90": 0.023476803788209786583163207751567824743688106536865234375,
"chunk_inter_token_latency_p99": 0.023476803788209786583163207751567824743688106536865234375
},
{
"ttft_avg": 0.5812644319958053529262542724609375,
"ttft_p50": 0.5812644319958053529262542724609375,
"ttft_p90": 0.5816280703875236213207244873046875,
"ttft_p99": 0.58170988902566023170948028564453125,
"wall_time": 35.553524311049841344356536865234375,
"num_errors": 0,
"concurrency": 2,
"underfilled": false,
"ready_reason": "metrics_unavailable, active_streams=2/2, stable=3.0s",
"total_tokens": 1882,
"aggregate_tps": 62.83373783456255523560685105621814727783203125,
"loop_detected": false,
"num_completed": 2,
"request_count": 2,
"avg_queue_reqs": 0,
"benchmark_mode": "duration",
"context_tokens": 0,
"failure_reason": "",
"max_queue_reqs": 0,
"queue_fraction": 0,
"timeout_reason": "",
"accept_norm_tps": 0,
"request_samples": [
{
"ttft": 0.58080988400615751743316650390625,
"latency": 0,
"completed": false,
"input_tokens": 78,
"output_tokens": 1105,
"output_tps_per_user": 31.64965423408434475049944012425839900970458984375,
"time_to_second_token": 0.0807216879911720752716064453125,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.031595921794402216653541159985252306796610355377197265625,
"chunk_inter_token_latency_avg": 0.03162456723573893457857053590487339533865451812744140625
},
{
"ttft": 0.581718979985453188419342041015625,
"latency": 0,
"completed": false,
"input_tokens": 78,
"output_tokens": 1105,
"output_tps_per_user": 31.650830124577399971030899905599653720855712890625,
"time_to_second_token": 0.08088596002198755741119384765625,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.03159474794386145368729756910397554747760295867919921875,
"chunk_inter_token_latency_avg": 0.031623392320963776558340185829365509562194347381591796875
}
],
"warmup_duration": 5.51100000000000012079226507921703159809112548828125,
"aggregate_source": "openai_continuous_usage",
"avg_running_reqs": 0,
"capacity_limited": false,
"hardware_summary": {
"samples": 14,
"gpu_count": 2,
"temp_avg_c": 52.25,
"temp_max_c": 54,
"vram_total_mb": 131072,
"cpu_temp_max_c": 66.1200000000000045474735088646411895751953125,
"cpu_util_avg_pct": 26.199999999999999289457264239899814128875732421875,
"duration_seconds": 28.76899999999999835154085303656756877899169921875,
"gpu_util_avg_pct": 76.5400000000000062527760746888816356658935546875,
"gpu_util_max_pct": 89,
"mem_util_avg_pct": 10.28999999999999914734871708787977695465087890625,
"mem_util_max_pct": 12,
"pcie_rx_avg_mb_s": 431.6399999999999863575794734060764312744140625,
"pcie_rx_max_mb_s": 567,
"pcie_tx_avg_mb_s": 346,
"pcie_tx_max_mb_s": 479,
"vram_used_avg_mb": 126720,
"vram_used_max_mb": 126720,
"power_total_avg_w": 311.94999999999998863131622783839702606201171875,
"power_total_max_w": 356.93000000000000682121026329696178436279296875,
"vram_used_avg_pct": 96.68000000000000682121026329696178436279296875,
"vram_used_max_pct": 96.68000000000000682121026329696178436279296875,
"power_limit_total_w": 360
},
"loop_diagnostics": [],
"max_running_reqs": 0,
"warmup_timed_out": false,
"input_seq_len_avg": 78,
"output_seq_len_avg": 1105,
"output_seq_len_p50": 1105,
"output_seq_len_p90": 1105,
"output_seq_len_p99": 1105,
"server_spec_drafts": 0,
"server_steps_per_s": 0,
"server_utilization": 0,
"accept_norm_ref_len": 0,
"measurement_seconds": 29.952062000000001518174030934460461139678955078125,
"per_request_avg_tps": 31.416868917281277617803425528109073638916015625,
"request_latency_avg": 0,
"request_latency_p50": 0,
"request_latency_p90": 0,
"request_latency_p99": 0,
"server_engine_steps": 0,
"client_output_tokens": 1882,
"request_count_target": 0,
"server_output_tokens": 0,
"warmup_request_count": 0,
"effective_concurrency": 0,
"server_gen_throughput": 0,
"loop_detection_enabled": true,
"server_spec_pos_accept": [],
"completed_request_count": 0,
"inter_token_latency_avg": 0.03159533486913183863986631649822811596095561981201171875,
"inter_token_latency_p50": 0.03159533486913183863986631649822811596095561981201171875,
"inter_token_latency_p90": 0.0315958044093481438263637528507388196885585784912109375,
"inter_token_latency_p99": 0.03159591005589680590137646731818676926195621490478515625,
"output_tps_per_user_avg": 31.6502421793308741371220094151794910430908203125,
"output_tps_per_user_p50": 31.6502421793308741371220094151794910430908203125,
"output_tps_per_user_p90": 31.65071253552809338316365028731524944305419921875,
"output_tps_per_user_p99": 31.65081836567247108860101434402167797088623046875,
"server_spec_accept_rate": 0,
"loop_diagnostics_omitted": 0,
"measurement_wall_seconds": 30.00025600000000025602275854907929897308349609375,
"server_spec_draft_tokens": 0,
"time_to_second_token_avg": 0.080803824006579816341400146484375,
"time_to_second_token_p50": 0.080803824006579816341400146484375,
"time_to_second_token_p90": 0.08086953281890600642167754585898364894092082977294921875,
"time_to_second_token_p99": 0.08088431730167940092446343669507768936455249786376953125,
"server_spec_accept_length": 0,
"e2e_output_tps_per_user_avg": 0,
"e2e_output_tps_per_user_p50": 0,
"e2e_output_tps_per_user_p90": 0,
"e2e_output_tps_per_user_p99": 0,
"server_accept_len_effective": 0,
"server_spec_accepted_tokens": 0,
"chunk_inter_token_latency_avg": 0.031623979778351352099008408913505263626575469970703125,
"chunk_inter_token_latency_p50": 0.031623979778351352099008408913505263626575469970703125,
"chunk_inter_token_latency_p90": 0.03162444974426141530710054894370841793715953826904296875,
"chunk_inter_token_latency_p99": 0.0316245554865911826514235372087568975985050201416015625
},
{
"ttft_avg": 1.0343336299993097782135009765625,
"ttft_p50": 1.0346325300051830708980560302734375,
"ttft_p90": 1.0354357703879941698943412120570428669452667236328125,
"ttft_p99": 1.03572041842096940200690369238145649433135986328125,
"wall_time": 35.527733177994377911090850830078125,
"num_errors": 0,
"concurrency": 4,
"underfilled": false,
"ready_reason": "metrics_unavailable, active_streams=4/4, stable=3.0s",
"total_tokens": 2472,
"aggregate_tps": 82.702490526015054683739435859024524688720703125,
"loop_detected": false,
"num_completed": 4,
"request_count": 4,
"avg_queue_reqs": 0,
"benchmark_mode": "duration",
"context_tokens": 0,
"failure_reason": "",
"max_queue_reqs": 0,
"queue_fraction": 0,
"timeout_reason": "",
"accept_norm_tps": 0,
"request_samples": [
{
"ttft": 1.0357520459801889955997467041015625,
"latency": 0,
"completed": false,
"input_tokens": 78,
"output_tokens": 743,
"output_tps_per_user": 21.585614256403214739066243055276572704315185546875,
"time_to_second_token": 0.1072184600052423775196075439453125,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.046327150486503175519370500978766358457505702972412109375,
"chunk_inter_token_latency_avg": 0.0463896702577400177336386377646704204380512237548828125
},
{
"ttft": 1.0323174140066839754581451416015625,
"latency": 0,
"completed": false,
"input_tokens": 78,
"output_tokens": 743,
"output_tps_per_user": 21.58657449337491840424263500608503818511962890625,
"time_to_second_token": 0.1071929520112462341785430908203125,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.046325089712909640893290230678758234716951847076416015625,
"chunk_inter_token_latency_avg": 0.04638760670307551237101506558246910572052001953125
},
{
"ttft": 1.034697794006206095218658447265625,
"latency": 0,
"completed": false,
"input_tokens": 78,
"output_tokens": 743,
"output_tps_per_user": 21.58572080930594694336832617409527301788330078125,
"time_to_second_token": 0.10724474000744521617889404296875,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.046326921803273031497116818400172633118927478790283203125,
"chunk_inter_token_latency_avg": 0.0463894412658955312966924111606203950941562652587890625
},
{
"ttft": 1.03456726600416004657745361328125,
"latency": 0,
"completed": false,
"input_tokens": 78,
"output_tokens": 743,
"output_tps_per_user": 21.585639567205472388877751654945313930511474609375,
"time_to_second_token": 0.10722699598409235477447509765625,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.04632709616439974198431883678495069034397602081298828125,
"chunk_inter_token_latency_avg": 0.046389615862327406026199838606771663762629032135009765625
}
],
"warmup_duration": 5.519999999999999573674358543939888477325439453125,
"aggregate_source": "openai_continuous_usage",
"avg_running_reqs": 0,
"capacity_limited": false,
"hardware_summary": {
"samples": 14,
"gpu_count": 2,
"temp_avg_c": 53.6099999999999994315658113919198513031005859375,
"temp_max_c": 55,
"vram_total_mb": 131072,
"cpu_temp_max_c": 72.25,
"cpu_util_avg_pct": 20.96000000000000085265128291212022304534912109375,
"duration_seconds": 28.803000000000000824229573481716215610504150390625,
"gpu_util_avg_pct": 78.81999999999999317878973670303821563720703125,
"gpu_util_max_pct": 89,
"mem_util_avg_pct": 8.8599999999999994315658113919198513031005859375,
"mem_util_max_pct": 10,
"pcie_rx_avg_mb_s": 585.7100000000000363797880709171295166015625,
"pcie_rx_max_mb_s": 869,
"pcie_tx_avg_mb_s": 497.93000000000000682121026329696178436279296875,
"pcie_tx_max_mb_s": 718,
"vram_used_avg_mb": 126720,
"vram_used_max_mb": 126720,
"power_total_avg_w": 298.529999999999972715158946812152862548828125,
"power_total_max_w": 372.8799999999999954525264911353588104248046875,
"vram_used_avg_pct": 96.68000000000000682121026329696178436279296875,
"vram_used_max_pct": 96.68000000000000682121026329696178436279296875,
"power_limit_total_w": 360
},
"loop_diagnostics": [],
"max_running_reqs": 0,
"warmup_timed_out": false,
"input_seq_len_avg": 78,
"output_seq_len_avg": 743,
"output_seq_len_p50": 743,
"output_seq_len_p90": 743,
"output_seq_len_p99": 743,
"server_spec_drafts": 0,
"server_steps_per_s": 0,
"server_utilization": 0,
"accept_norm_ref_len": 0,
"measurement_seconds": 29.89027300000000053614712669514119625091552734375,
"per_request_avg_tps": 20.67562263150376367093485896475613117218017578125,
"request_latency_avg": 0,
"request_latency_p50": 0,
"request_latency_p90": 0,
"request_latency_p99": 0,
"server_engine_steps": 0,
"client_output_tokens": 2472,
"request_count_target": 0,
"server_output_tokens": 0,
"warmup_request_count": 0,
"effective_concurrency": 0,
"server_gen_throughput": 0,
"loop_detection_enabled": true,
"server_spec_pos_accept": [],
"completed_request_count": 0,
"inter_token_latency_avg": 0.046326564541771399208247572687469073571264743804931640625,
"inter_token_latency_p50": 0.04632700898383639021016477954617585055530071258544921875,
"inter_token_latency_p90": 0.046327134189872144764965611329898820258677005767822265625,
"inter_token_latency_p99": 0.04632714885684006966837245045098825357854366302490234375,
"output_tps_per_user_avg": 21.58588728157238989524557837285101413726806640625,
"output_tps_per_user_p50": 21.58568018825570788976619951426982879638671875,
"output_tps_per_user_p90": 21.5863183881542255448948708362877368927001953125,
"output_tps_per_user_p99": 21.586548882852849118307858589105308055877685546875,
"server_spec_accept_rate": 0,
"loop_diagnostics_omitted": 0,
"measurement_wall_seconds": 30.00057199999999824058249942027032375335693359375,
"server_spec_draft_tokens": 0,
"time_to_second_token_avg": 0.10722078700200654566287994384765625,
"time_to_second_token_p50": 0.10722272799466736614704132080078125,
"time_to_second_token_p90": 0.107239416800439357757568359375,
"time_to_second_token_p99": 0.10724420768674462756120391304648364894092082977294921875,
"server_spec_accept_length": 0,
"e2e_output_tps_per_user_avg": 0,
"e2e_output_tps_per_user_p50": 0,
"e2e_output_tps_per_user_p90": 0,
"e2e_output_tps_per_user_p99": 0,
"server_accept_len_effective": 0,
"server_spec_accepted_tokens": 0,
"chunk_inter_token_latency_avg": 0.046389083522259615122163012301825801841914653778076171875,
"chunk_inter_token_latency_p50": 0.0463895285641114651919991729300818406045436859130859375,
"chunk_inter_token_latency_p90": 0.0463896539391162321397388268451322801411151885986328125,
"chunk_inter_token_latency_p99": 0.04638966862587763639869109510982525534927845001220703125
},
{
"ttft_avg": 0.369262880994938313961029052734375,
"ttft_p50": 0.369262880994938313961029052734375,
"ttft_p90": 0.369262880994938313961029052734375,
"ttft_p99": 0.369262880994938313961029052734375,
"wall_time": 56.6125428469968028366565704345703125,
"num_errors": 0,
"concurrency": 1,
"underfilled": false,
"ready_reason": "metrics_unavailable, active_streams=1/1, stable=3.0s",
"total_tokens": 1428,
"aggregate_tps": 47.6047477119508783971468801610171794891357421875,
"loop_detected": false,
"num_completed": 1,
"request_count": 1,
"avg_queue_reqs": 0,
"benchmark_mode": "duration",
"context_tokens": 32768,
"failure_reason": "",
"max_queue_reqs": 0,
"queue_fraction": 0,
"timeout_reason": "",
"accept_norm_tps": 0,
"request_samples": [
{
"ttft": 0.369262880994938313961029052734375,
"latency": 0,
"completed": false,
"input_tokens": 32319,
"output_tokens": 1554,
"output_tps_per_user": 46.86884168356476720873615704476833343505859375,
"time_to_second_token": 0.0593620089930482208728790283203125,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.0213361364198310088369225923088379204273223876953125,
"chunk_inter_token_latency_avg": 0.0214188880801535579123662245137893478386104106903076171875
}
],
"warmup_duration": 26.55499999999999971578290569595992565155029296875,
"aggregate_source": "openai_continuous_usage",
"avg_running_reqs": 0,
"capacity_limited": false,
"hardware_summary": {
"samples": 14,
"gpu_count": 2,
"temp_avg_c": 55.71000000000000085265128291212022304534912109375,
"temp_max_c": 57,
"vram_total_mb": 131072,
"cpu_temp_max_c": 67.3799999999999954525264911353588104248046875,
"cpu_util_avg_pct": 9.9399999999999995026200849679298698902130126953125,
"duration_seconds": 28.7349999999999994315658113919198513031005859375,
"gpu_util_avg_pct": 87.56999999999999317878973670303821563720703125,
"gpu_util_max_pct": 92,
"mem_util_avg_pct": 14.1099999999999994315658113919198513031005859375,
"mem_util_max_pct": 15,
"pcie_rx_avg_mb_s": 500,
"pcie_rx_max_mb_s": 570,
"pcie_tx_avg_mb_s": 435.06999999999999317878973670303821563720703125,
"pcie_tx_max_mb_s": 477,
"vram_used_avg_mb": 127356,
"vram_used_max_mb": 127356,
"power_total_avg_w": 339.54000000000002046363078989088535308837890625,
"power_total_max_w": 357.529999999999972715158946812152862548828125,
"vram_used_avg_pct": 97.159999999999996589394868351519107818603515625,
"vram_used_max_pct": 97.159999999999996589394868351519107818603515625,
"power_limit_total_w": 360
},
"loop_diagnostics": [],
"max_running_reqs": 0,
"warmup_timed_out": false,
"input_seq_len_avg": 32319,
"output_seq_len_avg": 1554,
"output_seq_len_p50": 1554,
"output_seq_len_p90": 1554,
"output_seq_len_p99": 1554,
"server_spec_drafts": 0,
"server_steps_per_s": 0,
"server_utilization": 0,
"accept_norm_ref_len": 0,
"measurement_seconds": 29.99700800000000100453689810819923877716064453125,
"per_request_avg_tps": 47.6047477119508783971468801610171794891357421875,
"request_latency_avg": 0,
"request_latency_p50": 0,
"request_latency_p90": 0,
"request_latency_p99": 0,
"server_engine_steps": 0,
"client_output_tokens": 1428,
"request_count_target": 0,
"server_output_tokens": 0,
"warmup_request_count": 0,
"effective_concurrency": 0,
"server_gen_throughput": 0,
"loop_detection_enabled": true,
"server_spec_pos_accept": [],
"completed_request_count": 0,
"inter_token_latency_avg": 0.0213361364198310088369225923088379204273223876953125,
"inter_token_latency_p50": 0.0213361364198310088369225923088379204273223876953125,
"inter_token_latency_p90": 0.0213361364198310088369225923088379204273223876953125,
"inter_token_latency_p99": 0.0213361364198310088369225923088379204273223876953125,
"output_tps_per_user_avg": 46.86884168356476720873615704476833343505859375,
"output_tps_per_user_p50": 46.86884168356476720873615704476833343505859375,
"output_tps_per_user_p90": 46.86884168356476720873615704476833343505859375,
"output_tps_per_user_p99": 46.86884168356476720873615704476833343505859375,
"server_spec_accept_rate": 0,
"loop_diagnostics_omitted": 0,
"measurement_wall_seconds": 30.00015700000000151703716255724430084228515625,
"server_spec_draft_tokens": 0,
"time_to_second_token_avg": 0.0593620089930482208728790283203125,
"time_to_second_token_p50": 0.0593620089930482208728790283203125,
"time_to_second_token_p90": 0.0593620089930482208728790283203125,
"time_to_second_token_p99": 0.0593620089930482208728790283203125,
"server_spec_accept_length": 0,
"e2e_output_tps_per_user_avg": 0,
"e2e_output_tps_per_user_p50": 0,
"e2e_output_tps_per_user_p90": 0,
"e2e_output_tps_per_user_p99": 0,
"server_accept_len_effective": 0,
"server_spec_accepted_tokens": 0,
"chunk_inter_token_latency_avg": 0.0214188880801535579123662245137893478386104106903076171875,
"chunk_inter_token_latency_p50": 0.0214188880801535579123662245137893478386104106903076171875,
"chunk_inter_token_latency_p90": 0.0214188880801535579123662245137893478386104106903076171875,
"chunk_inter_token_latency_p99": 0.0214188880801535579123662245137893478386104106903076171875
},
{
"ttft_avg": 0.7525122590013779699802398681640625,
"ttft_p50": 0.7525122590013779699802398681640625,
"ttft_p90": 0.7525946878013201057910919189453125,
"ttft_p99": 0.75261323428130710855299412287422455847263336181640625,
"wall_time": 35.097736009978689253330230712890625,
"num_errors": 0,
"concurrency": 2,
"underfilled": false,
"ready_reason": "metrics_unavailable, active_streams=2/2, stable=3.0s",
"total_tokens": 1340,
"aggregate_tps": 44.72876356445967616082270978949964046478271484375,
"loop_detected": false,
"num_completed": 2,
"request_count": 2,
"avg_queue_reqs": 0,
"benchmark_mode": "duration",
"context_tokens": 32768,
"failure_reason": "",
"max_queue_reqs": 0,
"queue_fraction": 0,
"timeout_reason": "",
"accept_norm_tps": 0,
"request_samples": [
{
"ttft": 0.7524092230014503002166748046875,
"latency": 0,
"completed": false,
"input_tokens": 32319,
"output_tokens": 757,
"output_tps_per_user": 22.42087483693462246492345002479851245880126953125,
"time_to_second_token": 0.12584797502495348453521728515625,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.0446012926468269677027222996912314556539058685302734375,
"chunk_inter_token_latency_avg": 0.044778987039842212125595466432059765793383121490478515625
},
{
"ttft": 0.752615295001305639743804931640625,
"latency": 0,
"completed": false,
"input_tokens": 32319,
"output_tokens": 757,
"output_tps_per_user": 22.4217802748879506680168560706079006195068359375,
"time_to_second_token": 0.124590807012282311916351318359375,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.04459949155420030175722700960250222124159336090087890625,
"chunk_inter_token_latency_avg": 0.0447771787715477154190324426963343285024166107177734375
}
],
"warmup_duration": 5.019999999999999573674358543939888477325439453125,
"aggregate_source": "openai_continuous_usage",
"avg_running_reqs": 0,
"capacity_limited": false,
"hardware_summary": {
"samples": 13,
"gpu_count": 2,
"temp_avg_c": 52.38000000000000255795384873636066913604736328125,
"temp_max_c": 54,
"vram_total_mb": 131072,
"cpu_temp_max_c": 64,
"cpu_util_avg_pct": 19.530000000000001136868377216160297393798828125,
"duration_seconds": 26.54299999999999926103555480949580669403076171875,
"gpu_util_avg_pct": 80.7699999999999960209606797434389591217041015625,
"gpu_util_max_pct": 92,
"mem_util_avg_pct": 8.269999999999999573674358543939888477325439453125,
"mem_util_max_pct": 10,
"pcie_rx_avg_mb_s": 353,
"pcie_rx_max_mb_s": 563,
"pcie_tx_avg_mb_s": 249.539999999999992041921359486877918243408203125,
"pcie_tx_max_mb_s": 321,
"vram_used_avg_mb": 128040,
"vram_used_max_mb": 128040,
"power_total_avg_w": 281.55000000000001136868377216160297393798828125,
"power_total_max_w": 302.08999999999997498889570124447345733642578125,
"vram_used_avg_pct": 97.68999999999999772626324556767940521240234375,
"vram_used_max_pct": 97.68999999999999772626324556767940521240234375,
"power_limit_total_w": 360
},
"loop_diagnostics": [],
"max_running_reqs": 0,
"warmup_timed_out": false,
"input_seq_len_avg": 32319,
"output_seq_len_avg": 757,
"output_seq_len_p50": 757,
"output_seq_len_p90": 757,
"output_seq_len_p99": 757,
"server_spec_drafts": 0,
"server_steps_per_s": 0,
"server_utilization": 0,
"accept_norm_ref_len": 0,
"measurement_seconds": 29.95835100000000039699443732388317584991455078125,
"per_request_avg_tps": 22.364381782229838080411354894749820232391357421875,
"request_latency_avg": 0,
"request_latency_p50": 0,
"request_latency_p90": 0,
"request_latency_p99": 0,
"server_engine_steps": 0,
"client_output_tokens": 1340,
"request_count_target": 0,
"server_output_tokens": 0,
"warmup_request_count": 0,
"effective_concurrency": 0,
"server_gen_throughput": 0,
"loop_detection_enabled": true,
"server_spec_pos_accept": [],
"completed_request_count": 0,
"inter_token_latency_avg": 0.044600392100513634729974654646866838447749614715576171875,
"inter_token_latency_p50": 0.044600392100513634729974654646866838447749614715576171875,
"inter_token_latency_p90": 0.044601112537564303883730332245249883271753787994384765625,
"inter_token_latency_p99": 0.04460127463590070340249127411880181171000003814697265625,
"output_tps_per_user_avg": 22.421327555911286566470153047703206539154052734375,
"output_tps_per_user_p50": 22.421327555911286566470153047703206539154052734375,
"output_tps_per_user_p90": 22.421689731092616426622043945826590061187744140625,
"output_tps_per_user_p99": 22.421771220508418309691478498280048370361328125,
"server_spec_accept_rate": 0,
"loop_diagnostics_omitted": 0,
"measurement_wall_seconds": 30.00049299999999874444256420247256755828857421875,
"server_spec_draft_tokens": 0,
"time_to_second_token_avg": 0.1252193910186178982257843017578125,
"time_to_second_token_p50": 0.1252193910186178982257843017578125,
"time_to_second_token_p90": 0.125722258223686378375560934728127904236316680908203125,
"time_to_second_token_p99": 0.1258354033448267628170214038618723861873149871826171875,
"server_spec_accept_length": 0,
"e2e_output_tps_per_user_avg": 0,
"e2e_output_tps_per_user_p50": 0,
"e2e_output_tps_per_user_p90": 0,
"e2e_output_tps_per_user_p99": 0,
"server_accept_len_effective": 0,
"server_spec_accepted_tokens": 0,
"chunk_inter_token_latency_avg": 0.0447780829056949603028670026105828583240509033203125,
"chunk_inter_token_latency_p50": 0.0447780829056949603028670026105828583240509033203125,
"chunk_inter_token_latency_p90": 0.04477880621301276453660733523065573535859584808349609375,
"chunk_inter_token_latency_p99": 0.044778968957159263897249701358305173926055431365966796875
},
{
"ttft_avg": 1.3875278134946711361408233642578125,
"ttft_p50": 1.38750035897828638553619384765625,
"ttft_p90": 1.3882693307998124065960610096226446330547332763671875,
"ttft_p99": 1.3885471375967608498314120879513211548328399658203125,
"wall_time": 35.0940270979772321879863739013671875,
"num_errors": 0,
"concurrency": 4,
"underfilled": false,
"ready_reason": "metrics_unavailable, active_streams=4/4, stable=3.0s",
"total_tokens": 2040,
"aggregate_tps": 68.1811388429051277171311085112392902374267578125,
"loop_detected": false,
"num_completed": 4,
"request_count": 4,
"avg_queue_reqs": 0,
"benchmark_mode": "duration",
"context_tokens": 32768,
"failure_reason": "",
"max_queue_reqs": 0,
"queue_fraction": 0,
"timeout_reason": "",
"accept_norm_tps": 0,
"request_samples": [
{
"ttft": 1.3874516270007006824016571044921875,
"latency": 0,
"completed": false,
"input_tokens": 32319,
"output_tokens": 565,
"output_tps_per_user": 17.069398791096364931263451580889523029327392578125,
"time_to_second_token": 0.1782914969953708350658416748046875,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.05858437149652943876088073693608748726546764373779296875,
"chunk_inter_token_latency_avg": 0.058688428994747077194205076011712662875652313232421875
},
{
"ttft": 1.388578005018644034862518310546875,
"latency": 0,
"completed": false,
"input_tokens": 32319,
"output_tokens": 565,
"output_tps_per_user": 17.07069758101510359438179875724017620086669921875,
"time_to_second_token": 0.17933490802533924579620361328125,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.058579914221673849705762648909512790851294994354248046875,
"chunk_inter_token_latency_avg": 0.058683963802884632487977256687372573651373386383056640625
},
{
"ttft": 1.3875490909558720886707305908203125,
"latency": 0,
"completed": false,
"input_tokens": 32319,
"output_tokens": 565,
"output_tps_per_user": 17.070715973374273488616381655447185039520263671875,
"time_to_second_token": 0.179324298049323260784149169921875,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.058579851106405324279169377632570103742182254791259765625,
"chunk_inter_token_latency_avg": 0.058683900575510837549497722420710488222539424896240234375
},
{
"ttft": 1.386532531003467738628387451171875,
"latency": 0,
"completed": false,
"input_tokens": 32319,
"output_tokens": 565,
"output_tps_per_user": 17.070728366537021969406850985251367092132568359375,
"time_to_second_token": 0.1793170240125618875026702880859375,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.0585798085780718513770892741376883350312709808349609375,
"chunk_inter_token_latency_avg": 0.058683857971638588024720917246668250299990177154541015625
}
],
"warmup_duration": 5.01100000000000012079226507921703159809112548828125,
"aggregate_source": "openai_continuous_usage",
"avg_running_reqs": 0,
"capacity_limited": false,
"hardware_summary": {
"samples": 13,
"gpu_count": 2,
"temp_avg_c": 52.14999999999999857891452847979962825775146484375,
"temp_max_c": 53,
"vram_total_mb": 131072,
"cpu_temp_max_c": 66.5,
"cpu_util_avg_pct": 16.190000000000001278976924368180334568023681640625,
"duration_seconds": 26.51599999999999823785401531495153903961181640625,
"gpu_util_avg_pct": 82.1200000000000045474735088646411895751953125,
"gpu_util_max_pct": 90,
"mem_util_avg_pct": 10.4199999999999999289457264239899814128875732421875,
"mem_util_max_pct": 12,
"pcie_rx_avg_mb_s": 517.23000000000001818989403545856475830078125,
"pcie_rx_max_mb_s": 775,
"pcie_tx_avg_mb_s": 348.76999999999998181010596454143524169921875,
"pcie_tx_max_mb_s": 643,
"vram_used_avg_mb": 128040,
"vram_used_max_mb": 128040,
"power_total_avg_w": 284.04000000000002046363078989088535308837890625,
"power_total_max_w": 309.68000000000000682121026329696178436279296875,
"vram_used_avg_pct": 97.68999999999999772626324556767940521240234375,
"vram_used_max_pct": 97.68999999999999772626324556767940521240234375,
"power_limit_total_w": 360
},
"loop_diagnostics": [],
"max_running_reqs": 0,
"warmup_timed_out": false,
"input_seq_len_avg": 32319,
"output_seq_len_avg": 565,
"output_seq_len_p50": 565,
"output_seq_len_p90": 565,
"output_seq_len_p99": 565,
"server_spec_drafts": 0,
"server_steps_per_s": 0,
"server_utilization": 0,
"accept_norm_ref_len": 0,
"measurement_seconds": 29.920297999999998950215740478597581386566162109375,
"per_request_avg_tps": 17.045284710726281929282777127809822559356689453125,
"request_latency_avg": 0,
"request_latency_p50": 0,
"request_latency_p90": 0,
"request_latency_p99": 0,
"server_engine_steps": 0,
"client_output_tokens": 2040,
"request_count_target": 0,
"server_output_tokens": 0,
"warmup_request_count": 0,
"effective_concurrency": 0,
"server_gen_throughput": 0,
"loop_detection_enabled": true,
"server_spec_pos_accept": [],
"completed_request_count": 0,
"inter_token_latency_avg": 0.05858098635067011950017246135757886804640293121337890625,
"inter_token_latency_p50": 0.05857988266403958699246601327104144729673862457275390625,
"inter_token_latency_p90": 0.05858303431407275996267713935594656504690647125244140625,
"inter_token_latency_p99": 0.05858423777828376810550281561518204398453235626220703125,
"output_tps_per_user_avg": 17.070385178005690107738701044581830501556396484375,
"output_tps_per_user_p50": 17.0707067771946867651422508060932159423828125,
"output_tps_per_user_p90": 17.070724648588196714626974426209926605224609375,
"output_tps_per_user_p99": 17.070727994742139088657495449297130107879638671875,
"server_spec_accept_rate": 0,
"loop_diagnostics_omitted": 0,
"measurement_wall_seconds": 30.0005009999999998626663000322878360748291015625,
"server_spec_draft_tokens": 0,
"time_to_second_token_avg": 0.1790669317706488072872161865234375,
"time_to_second_token_p50": 0.17932066103094257414340972900390625,
"time_to_second_token_p90": 0.1793317250325344447414721571476547978818416595458984375,
"time_to_second_token_p99": 0.1793345897260587573640577829792164266109466552734375,
"server_spec_accept_length": 0,
"e2e_output_tps_per_user_avg": 0,
"e2e_output_tps_per_user_p50": 0,
"e2e_output_tps_per_user_p90": 0,
"e2e_output_tps_per_user_p99": 0,
"server_accept_len_effective": 0,
"server_spec_accepted_tokens": 0,
"chunk_inter_token_latency_avg": 0.058685037836195282079376767114808899350464344024658203125,
"chunk_inter_token_latency_p50": 0.0586839321891977350187374895540415309369564056396484375,
"chunk_inter_token_latency_p90": 0.058687089437188345864004901386579149402678012847900390625,
"chunk_inter_token_latency_p99": 0.05868829503899120059173810659558512270450592041015625
},
{
"ttft_avg": 0.7310018099960871040821075439453125,
"ttft_p50": 0.7310018099960871040821075439453125,
"ttft_p90": 0.7310018099960871040821075439453125,
"ttft_p99": 0.7310018099960871040821075439453125,
"wall_time": 123.714279089006595313549041748046875,
"num_errors": 0,
"concurrency": 1,
"underfilled": false,
"ready_reason": "metrics_unavailable, active_streams=1/1, stable=3.0s",
"total_tokens": 1299,
"aggregate_tps": 43.36507712997916996755520813167095184326171875,
"loop_detected": false,
"num_completed": 1,
"request_count": 1,
"avg_queue_reqs": 0,
"benchmark_mode": "duration",
"context_tokens": 131072,
"failure_reason": "",
"max_queue_reqs": 0,
"queue_fraction": 0,
"timeout_reason": "",
"accept_norm_tps": 0,
"request_samples": [
{
"ttft": 0.7310018099960871040821075439453125,
"latency": 0,
"completed": false,
"input_tokens": 128886,
"output_tokens": 1467,
"output_tps_per_user": 43.4549218970450823462670086883008480072021484375,
"time_to_second_token": 0.0597036319668404757976531982421875,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.0230123529474804931760356652148402645252645015716552734375,
"chunk_inter_token_latency_avg": 0.0231545020048087892006183352577863843180239200592041015625
}
],
"warmup_duration": 93.69499999999999317878973670303821563720703125,
"aggregate_source": "openai_continuous_usage",
"avg_running_reqs": 0,
"capacity_limited": false,
"hardware_summary": {
"samples": 14,
"gpu_count": 2,
"temp_avg_c": 56.1099999999999994315658113919198513031005859375,
"temp_max_c": 58,
"vram_total_mb": 131072,
"cpu_temp_max_c": 64.5,
"cpu_util_avg_pct": 9.519999999999999573674358543939888477325439453125,
"duration_seconds": 28.702000000000001733724275254644453525543212890625,
"gpu_util_avg_pct": 87.9599999999999937472239253111183643341064453125,
"gpu_util_max_pct": 93,
"mem_util_avg_pct": 14.25,
"mem_util_max_pct": 16,
"pcie_rx_avg_mb_s": 465.79000000000002046363078989088535308837890625,
"pcie_rx_max_mb_s": 583,
"pcie_tx_avg_mb_s": 447.43000000000000682121026329696178436279296875,
"pcie_tx_max_mb_s": 503,
"vram_used_avg_mb": 128066,
"vram_used_max_mb": 128066,
"power_total_avg_w": 344.29000000000002046363078989088535308837890625,
"power_total_max_w": 366.57999999999998408384271897375583648681640625,
"vram_used_avg_pct": 97.7099999999999937472239253111183643341064453125,
"vram_used_max_pct": 97.7099999999999937472239253111183643341064453125,
"power_limit_total_w": 360
},
"loop_diagnostics": [],
"max_running_reqs": 0,
"warmup_timed_out": false,
"input_seq_len_avg": 128886,
"output_seq_len_avg": 1467,
"output_seq_len_p50": 1467,
"output_seq_len_p90": 1467,
"output_seq_len_p99": 1467,
"server_spec_drafts": 0,
"server_steps_per_s": 0,
"server_utilization": 0,
"accept_norm_ref_len": 0,
"measurement_seconds": 29.95497999999999905185177340172231197357177734375,
"per_request_avg_tps": 43.36507712997916996755520813167095184326171875,
"request_latency_avg": 0,
"request_latency_p50": 0,
"request_latency_p90": 0,
"request_latency_p99": 0,
"server_engine_steps": 0,
"client_output_tokens": 1299,
"request_count_target": 0,
"server_output_tokens": 0,
"warmup_request_count": 0,
"effective_concurrency": 0,
"server_gen_throughput": 0,
"loop_detection_enabled": true,
"server_spec_pos_accept": [],
"completed_request_count": 0,
"inter_token_latency_avg": 0.0230123529474804931760356652148402645252645015716552734375,
"inter_token_latency_p50": 0.0230123529474804931760356652148402645252645015716552734375,
"inter_token_latency_p90": 0.0230123529474804931760356652148402645252645015716552734375,
"inter_token_latency_p99": 0.0230123529474804931760356652148402645252645015716552734375,
"output_tps_per_user_avg": 43.4549218970450823462670086883008480072021484375,
"output_tps_per_user_p50": 43.4549218970450823462670086883008480072021484375,
"output_tps_per_user_p90": 43.4549218970450823462670086883008480072021484375,
"output_tps_per_user_p99": 43.4549218970450823462670086883008480072021484375,
"server_spec_accept_rate": 0,
"loop_diagnostics_omitted": 0,
"measurement_wall_seconds": 30.0001299999999986312104738317430019378662109375,
"server_spec_draft_tokens": 0,
"time_to_second_token_avg": 0.0597036319668404757976531982421875,
"time_to_second_token_p50": 0.0597036319668404757976531982421875,
"time_to_second_token_p90": 0.0597036319668404757976531982421875,
"time_to_second_token_p99": 0.0597036319668404757976531982421875,
"server_spec_accept_length": 0,
"e2e_output_tps_per_user_avg": 0,
"e2e_output_tps_per_user_p50": 0,
"e2e_output_tps_per_user_p90": 0,
"e2e_output_tps_per_user_p99": 0,
"server_accept_len_effective": 0,
"server_spec_accepted_tokens": 0,
"chunk_inter_token_latency_avg": 0.0231545020048087892006183352577863843180239200592041015625,
"chunk_inter_token_latency_p50": 0.0231545020048087892006183352577863843180239200592041015625,
"chunk_inter_token_latency_p90": 0.0231545020048087892006183352577863843180239200592041015625,
"chunk_inter_token_latency_p99": 0.0231545020048087892006183352577863843180239200592041015625
},
{
"ttft_avg": 1.4347215775051154196262359619140625,
"ttft_p50": 1.4347215775051154196262359619140625,
"ttft_p90": 1.4356504851137288358842170055140741169452667236328125,
"ttft_p99": 1.4358594893256668711956081097014248371124267578125,
"wall_time": 35.52073261700570583343505859375,
"num_errors": 0,
"concurrency": 2,
"underfilled": false,
"ready_reason": "metrics_unavailable, active_streams=2/2, stable=3.0s",
"total_tokens": 1296,
"aggregate_tps": 43.38180322174043368477214244194328784942626953125,
"loop_detected": false,
"num_completed": 2,
"request_count": 2,
"avg_queue_reqs": 0,
"benchmark_mode": "duration",
"context_tokens": 131072,
"failure_reason": "",
"max_queue_reqs": 0,
"queue_fraction": 0,
"timeout_reason": "",
"accept_norm_tps": 0,
"request_samples": [
{
"ttft": 1.43588271201588213443756103515625,
"latency": 0,
"completed": false,
"input_tokens": 128886,
"output_tokens": 725,
"output_tps_per_user": 21.97301867192306445986105245538055896759033203125,
"time_to_second_token": 0.119680150994099676609039306640625,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.04551036045301283528008440271150902844965457916259765625,
"chunk_inter_token_latency_avg": 0.045699723950043402709297879482619464397430419921875
},
{
"ttft": 1.433560442994348704814910888671875,
"latency": 0,
"completed": false,
"input_tokens": 128886,
"output_tokens": 725,
"output_tps_per_user": 21.97314426881050763995517627336084842681884765625,
"time_to_second_token": 0.120091582997702062129974365234375,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.045510100319117140210689598234239383600652217864990234375,
"chunk_inter_token_latency_avg": 0.045699462733759792232302032743973541073501110076904296875
}
],
"warmup_duration": 5.51499999999999968025576890795491635799407958984375,
"aggregate_source": "openai_continuous_usage",
"avg_running_reqs": 0,
"capacity_limited": false,
"hardware_summary": {
"samples": 14,
"gpu_count": 2,
"temp_avg_c": 52.53999999999999914734871708787977695465087890625,
"temp_max_c": 54,
"vram_total_mb": 131072,
"cpu_temp_max_c": 68.1200000000000045474735088646411895751953125,
"cpu_util_avg_pct": 19.489999999999998436805981327779591083526611328125,
"duration_seconds": 28.743999999999999772626324556767940521240234375,
"gpu_util_avg_pct": 82.8900000000000005684341886080801486968994140625,
"gpu_util_max_pct": 91,
"mem_util_avg_pct": 8.32000000000000028421709430404007434844970703125,
"mem_util_max_pct": 9,
"pcie_rx_avg_mb_s": 303.5,
"pcie_rx_max_mb_s": 434,
"pcie_tx_avg_mb_s": 267.5,
"pcie_tx_max_mb_s": 377,
"vram_used_avg_mb": 128066,
"vram_used_max_mb": 128066,
"power_total_avg_w": 287.1000000000000227373675443232059478759765625,
"power_total_max_w": 309.93999999999999772626324556767940521240234375,
"vram_used_avg_pct": 97.7099999999999937472239253111183643341064453125,
"vram_used_max_pct": 97.7099999999999937472239253111183643341064453125,
"power_limit_total_w": 360
},
"loop_diagnostics": [],
"max_running_reqs": 0,
"warmup_timed_out": false,
"input_seq_len_avg": 128886,
"output_seq_len_avg": 725,
"output_seq_len_p50": 725,
"output_seq_len_p90": 725,
"output_seq_len_p99": 725,
"server_spec_drafts": 0,
"server_steps_per_s": 0,
"server_utilization": 0,
"accept_norm_ref_len": 0,
"measurement_seconds": 29.87427699999999930469130049459636211395263671875,
"per_request_avg_tps": 21.690901610870216842386071220971643924713134765625,
"request_latency_avg": 0,
"request_latency_p50": 0,
"request_latency_p90": 0,
"request_latency_p99": 0,
"server_engine_steps": 0,
"client_output_tokens": 1296,
"request_count_target": 0,
"server_output_tokens": 0,
"warmup_request_count": 0,
"effective_concurrency": 0,
"server_gen_throughput": 0,
"loop_detection_enabled": true,
"server_spec_pos_accept": [],
"completed_request_count": 0,
"inter_token_latency_avg": 0.04551023038606498427594004851926001720130443572998046875,
"inter_token_latency_p50": 0.04551023038606498427594004851926001720130443572998046875,
"inter_token_latency_p90": 0.045510334439623267854813093435950577259063720703125,
"inter_token_latency_p99": 0.045510357851673878537557271783953183330595493316650390625,
"output_tps_per_user_avg": 21.97308147036678604990811436437070369720458984375,
"output_tps_per_user_p50": 21.97308147036678604990811436437070369720458984375,
"output_tps_per_user_p90": 21.973131709121762611403028131462633609771728515625,
"output_tps_per_user_p99": 21.973143012841632071285857819020748138427734375,
"server_spec_accept_rate": 0,
"loop_diagnostics_omitted": 0,
"measurement_wall_seconds": 30.000575999999998799694367335177958011627197265625,
"server_spec_draft_tokens": 0,
"time_to_second_token_avg": 0.1198858669959008693695068359375,
"time_to_second_token_p50": 0.1198858669959008693695068359375,
"time_to_second_token_p90": 0.12005043979734182635343842093789135105907917022705078125,
"time_to_second_token_p99": 0.1200874686776660371645419900232809595763683319091796875,
"server_spec_accept_length": 0,
"e2e_output_tps_per_user_avg": 0,
"e2e_output_tps_per_user_p50": 0,
"e2e_output_tps_per_user_p90": 0,
"e2e_output_tps_per_user_p99": 0,
"server_accept_len_effective": 0,
"server_spec_accepted_tokens": 0,
"chunk_inter_token_latency_avg": 0.045699593341901600940246908066910691559314727783203125,
"chunk_inter_token_latency_p50": 0.045699593341901600940246908066910691559314727783203125,
"chunk_inter_token_latency_p90": 0.045699697828415042355487685199477709829807281494140625,
"chunk_inter_token_latency_p99": 0.04569972133788056944947442161719663999974727630615234375
},
{
"ttft_avg": 2.564350451764767058193683624267578125,
"ttft_p50": 2.4457031080382876098155975341796875,
"ttft_p90": 2.778426669229520751258633026736788451671600341796875,
"ttft_p99": 2.90665558924491040926341156591661274433135986328125,
"wall_time": 37.6416040249750949442386627197265625,
"num_errors": 0,
"concurrency": 4,
"underfilled": false,
"ready_reason": "metrics_unavailable, active_streams=4/4, stable=3.0s",
"total_tokens": 1679,
"aggregate_tps": 56.09202846683027843255331390537321567535400390625,
"loop_detected": false,
"num_completed": 4,
"request_count": 4,
"avg_queue_reqs": 0,
"benchmark_mode": "duration",
"context_tokens": 131072,
"failure_reason": "",
"max_queue_reqs": 0,
"queue_fraction": 0,
"timeout_reason": "",
"accept_norm_tps": 0,
"request_samples": [
{
"ttft": 2.44509234395809471607208251953125,
"latency": 0,
"completed": false,
"input_tokens": 128886,
"output_tokens": 490,
"output_tps_per_user": 14.3793669860936912385795949376188218593597412109375,
"time_to_second_token": 0.4838951409910805523395538330078125,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.06954409056859746363077334763147518970072269439697265625,
"chunk_inter_token_latency_avg": 0.06968659895091015854529103989989380352199077606201171875
},
{
"ttft": 2.445981321041472256183624267578125,
"latency": 0,
"completed": false,
"input_tokens": 128886,
"output_tokens": 490,
"output_tps_per_user": 14.3813973548474667296659390558488667011260986328125,
"time_to_second_token": 0.4817859869799576699733734130859375,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.06953427231902016558695578396509517915546894073486328125,
"chunk_inter_token_latency_avg": 0.069676760581968977259492703524301759898662567138671875
},
{
"ttft": 2.44542489503510296344757080078125,
"latency": 0,
"completed": false,
"input_tokens": 128886,
"output_tokens": 490,
"output_tps_per_user": 14.38266686191607135469894274137914180755615234375,
"time_to_second_token": 0.47874122601933777332305908203125,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.06952813477505374872844612355038407258689403533935546875,
"chunk_inter_token_latency_avg": 0.069670610461068205321311097577563486993312835693359375
},
{
"ttft": 2.9209032470243982970714569091796875,
"latency": 0,
"completed": false,
"input_tokens": 128886,
"output_tokens": 479,
"output_tps_per_user": 14.26083580206037737525548436678946018218994140625,
"time_to_second_token": 0.178458156995475292205810546875,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.0701221172363208911093579445150680840015411376953125,
"chunk_inter_token_latency_avg": 0.07026912377140752397064460410547326318919658660888671875
}
],
"warmup_duration": 7.5220000000000002415845301584340631961822509765625,
"aggregate_source": "openai_continuous_usage",
"avg_running_reqs": 0,
"capacity_limited": false,
"hardware_summary": {
"samples": 14,
"gpu_count": 2,
"temp_avg_c": 53.03999999999999914734871708787977695465087890625,
"temp_max_c": 55,
"vram_total_mb": 131072,
"cpu_temp_max_c": 65.8799999999999954525264911353588104248046875,
"cpu_util_avg_pct": 15.1699999999999999289457264239899814128875732421875,
"duration_seconds": 28.83800000000000096633812063373625278472900390625,
"gpu_util_avg_pct": 84.4599999999999937472239253111183643341064453125,
"gpu_util_max_pct": 90,
"mem_util_avg_pct": 11.67999999999999971578290569595992565155029296875,
"mem_util_max_pct": 13,
"pcie_rx_avg_mb_s": 414.93000000000000682121026329696178436279296875,
"pcie_rx_max_mb_s": 793,
"pcie_tx_avg_mb_s": 358.6399999999999863575794734060764312744140625,
"pcie_tx_max_mb_s": 595,
"vram_used_avg_mb": 128066,
"vram_used_max_mb": 128066,
"power_total_avg_w": 313.92000000000001591615728102624416351318359375,
"power_total_max_w": 344.779999999999972715158946812152862548828125,
"vram_used_avg_pct": 97.7099999999999937472239253111183643341064453125,
"vram_used_max_pct": 97.7099999999999937472239253111183643341064453125,
"power_limit_total_w": 360
},
"loop_diagnostics": [],
"max_running_reqs": 0,
"warmup_timed_out": false,
"input_seq_len_avg": 128886,
"output_seq_len_avg": 487.25,
"output_seq_len_p50": 490,
"output_seq_len_p90": 490,
"output_seq_len_p99": 490,
"server_spec_drafts": 0,
"server_steps_per_s": 0,
"server_utilization": 0,
"accept_norm_ref_len": 0,
"measurement_seconds": 29.9329520000000002255546860396862030029296875,
"per_request_avg_tps": 14.0230071167075696081383284763433039188385009765625,
"request_latency_avg": 0,
"request_latency_p50": 0,
"request_latency_p90": 0,
"request_latency_p99": 0,
"server_engine_steps": 0,
"client_output_tokens": 1679,
"request_count_target": 0,
"server_output_tokens": 0,
"warmup_request_count": 0,
"effective_concurrency": 0,
"server_gen_throughput": 0,
"loop_detection_enabled": true,
"server_spec_pos_accept": [],
"completed_request_count": 0,
"inter_token_latency_avg": 0.0696821537247480637944363479618914425373077392578125,
"inter_token_latency_p50": 0.06953918144380881460886456579828518442809581756591796875,
"inter_token_latency_p90": 0.06994870923600386702911890779432724229991436004638671875,
"inter_token_latency_p99": 0.07010477643628919286467038318733102641999721527099609375,
"output_tps_per_user_avg": 14.3510667512294016745499902754090726375579833984375,
"output_tps_per_user_p50": 14.380382170470578984122766996733844280242919921875,
"output_tps_per_user_p90": 14.382286009795489434281989815644919872283935546875,
"output_tps_per_user_p99": 14.382628776704013517928615328855812549591064453125,
"server_spec_accept_rate": 0,
"loop_diagnostics_omitted": 0,
"measurement_wall_seconds": 30.001152000000001152102413470856845378875732421875,
"server_spec_draft_tokens": 0,
"time_to_second_token_avg": 0.40572012774646282196044921875,
"time_to_second_token_p50": 0.48026360649964772164821624755859375,
"time_to_second_token_p90": 0.48326239478774368762969970703125,
"time_to_second_token_p99": 0.483831866370746876970798666661721654236316680908203125,
"server_spec_accept_length": 0,
"e2e_output_tps_per_user_avg": 0,
"e2e_output_tps_per_user_p50": 0,
"e2e_output_tps_per_user_p90": 0,
"e2e_output_tps_per_user_p99": 0,
"server_accept_len_effective": 0,
"server_spec_accepted_tokens": 0,
"chunk_inter_token_latency_avg": 0.069825773441338723213078765184036456048488616943359375,
"chunk_inter_token_latency_p50": 0.06968167976643957484128577561932615935802459716796875,
"chunk_inter_token_latency_p90": 0.0700943663252583115674809732809080742299556732177734375,
"chunk_inter_token_latency_p99": 0.07025164802679260966922214493024512194097042083740234375
}
],
"metadata": {
"model": "GLM-5.3-Flash-low",
"engine": "openai_proxy",
"version": "0.8.0",
"ignore_eos": true,
"max_tokens": 2048,
"decode_mode": "duration",
"temperature": 0,
"prefill_mode": "skipped",
"skip_prefill": true,
"context_lengths": [
0,
32768,
131072
],
"duration_per_test": 30,
"concurrency_levels": [
1,
2,
4
],
"primary_decode_layer": "sustained_decode",
"input_token_targeting": "upstream estimate fallback: /tokenize returns token_length, not count",
"packaged_from_multiple_original_runs": true
},
"gpu_count": 2,
"gpu_model": "NVIDIA CMP 170HX",
"report_md": "Re-upload of run 4140365f3b3be8f4a92d60ea6fc957d3fbcd4872 with the headline fields the site reads (one agent, short prompt, 78 input tokens); every measurement is unchanged.\n\n# GLM-5.3-Flash, low effort, on 2x CMP170HX\n\nFour native slots; EXL3 3.05bpw with the Weightless GLP44 steering vector at alpha 2, GPU-resident Q8 KV. 1, 2 and 4 agents at short and nominal 32K/128K context, one 30-second run per cell. Exact completion-token usage, including reasoning; actual input lengths are retained. A separate four-request 512-token completion cohort.\n\nInference speed only, not accuracy or intelligence. Differences between efforts are not separated from run-to-run variation.\n\nFull methods, tables and caveats: https://github.com/alexvnesta/glm53-flash-cmp170hx-tp/blob/fa0adf1899015f692ddce30391ee7e740a0910c9/docs/COMMUNITY_BENCHMARK_2026-10-17.md\n",
"deployment": {
"offload": "none",
"layout_kind": "tp",
"layout_detail": "TP2 on 2x CMP170HX 64GiB, 4 active sequences",
"offload_detail": "Active weights/KV on GPU; inactive RAM/disk retention"
},
"concurrency": 1,
"context_len": 78,
"recipe_slug": "glm-5-3-flash-low-effort-on-2x-cmp170hx-exl3-tp2-four-slots",
"source_runs": [
{
"gates": {
"observed_efforts": [
"low"
],
"cpu_psi_avg10_max": 1.6100000000000000976996261670137755572795867919921875,
"public_router_posts": 0,
"peak_active_sequences": 4
},
"label": "low-ctx0",
"sha256": "02f7c28ff3d92ebb25d14edb4f7e0bbbd090f95c15a17bca863131e5aa557c3d",
"command": [
"/data2/llm-models/glm53-stream-usage-20261017/bench-venv/bin/python",
"/data2/llm-models/glm53-stream-usage-20261017/observe.py",
"/data2/llm-models/glm53-stream-usage-20261017/llm_decode_bench-v0.8.0.py",
"--",
"--host",
"http://***:8014",
"--model",
"GLM-5.3-Flash-low",
"--concurrency",
"1,2,4",
"--contexts",
"0",
"--prefill-contexts",
"0",
"--skip-prefill",
"--kv-budget",
"655360",
"--max-tokens",
"2048",
"--temperature",
"0",
"--duration",
"30",
"--token-targeting",
"exact",
"--decode-warmup-seconds",
"3",
"--display-mode",
"plain",
"--no-resume",
"--output",
"benchmark_results.json"
],
"started": "2026-10-17T12:31:26.340149+00:00",
"finished": "2026-10-17T12:33:34.792931+00:00",
"raw_file": "runs/low-ctx0-attempt2/benchmark_results.json"
},
{
"gates": {
"observed_efforts": [
"low"
],
"cpu_psi_avg10_max": 0.2099999999999999922284388276239042170345783233642578125,
"public_router_posts": 0,
"peak_active_sequences": 4
},
"label": "low-ctx32768",
"sha256": "a55fb47f1120b5f32ee70ac18939d6f0d052311de877d88fa9f2fd478082545a",
"command": [
"/data2/llm-models/glm53-stream-usage-20261017/bench-venv/bin/python",
"/data2/llm-models/glm53-stream-usage-20261017/observe.py",
"/data2/llm-models/glm53-stream-usage-20261017/llm_decode_bench-v0.8.0.py",
"--",
"--host",
"http://***:8014",
"--model",
"GLM-5.3-Flash-low",
"--concurrency",
"1,2,4",
"--contexts",
"32768",
"--prefill-contexts",
"32768",
"--skip-prefill",
"--kv-budget",
"655360",
"--max-tokens",
"2048",
"--temperature",
"0",
"--duration",
"30",
"--token-targeting",
"exact",
"--decode-warmup-seconds",
"3",
"--display-mode",
"plain",
"--no-resume",
"--output",
"benchmark_results.json"
],
"started": "2026-10-17T12:38:05.366662+00:00",
"finished": "2026-10-17T12:41:02.441380+00:00",
"raw_file": "runs/low-ctx32768-attempt1/benchmark_results.json"
},
{
"gates": {
"observed_efforts": [
"low"
],
"cpu_psi_avg10_max": 0.330000000000000015543122344752191565930843353271484375,
"public_router_posts": 0,
"peak_active_sequences": 4
},
"label": "low-ctx131072",
"sha256": "412c827e4f9a3912920332f23b319bbb92371febc3e49a82af6f1ec4fbe66741",
"command": [
"/data2/llm-models/glm53-stream-usage-20261017/bench-venv/bin/python",
"/data2/llm-models/glm53-stream-usage-20261017/observe.py",
"/data2/llm-models/glm53-stream-usage-20261017/llm_decode_bench-v0.8.0.py",
"--",
"--host",
"http://***:8014",
"--model",
"GLM-5.3-Flash-low",
"--concurrency",
"1,2,4",
"--contexts",
"131072",
"--prefill-contexts",
"131072",
"--skip-prefill",
"--kv-budget",
"655360",
"--max-tokens",
"2048",
"--temperature",
"0",
"--duration",
"30",
"--token-targeting",
"exact",
"--decode-warmup-seconds",
"3",
"--display-mode",
"plain",
"--no-resume",
"--output",
"benchmark_results.json"
],
"started": "2026-10-17T12:47:11.176143+00:00",
"finished": "2026-10-17T12:52:26.357106+00:00",
"raw_file": "runs/low-ctx131072-attempt1/benchmark_results.json"
},
{
"gates": {
"observed_efforts": [
"low"
],
"cpu_psi_avg10_max": 0.54000000000000003552713678800500929355621337890625,
"public_router_posts": 0,
"peak_active_sequences": 4
},
"label": "low-finite-ctx0",
"sha256": "36cf2e63ea4b3d8b45b25812e251de77ef1b50cc86bdf0d7b5be979d388983ff",
"command": [
"/data2/llm-models/glm53-stream-usage-20261017/bench-venv/bin/python",
"/data2/llm-models/glm53-stream-usage-20261017/observe.py",
"/data2/llm-models/glm53-stream-usage-20261017/llm_decode_bench-v0.8.0.py",
"--",
"--host",
"http://***:8014",
"--model",
"GLM-5.3-Flash-low",
"--concurrency",
"1,2,4",
"--contexts",
"0",
"--prefill-contexts",
"0",
"--skip-prefill",
"--kv-budget",
"655360",
"--max-tokens",
"512",
"--temperature",
"0",
"--duration",
"30",
"--token-targeting",
"exact",
"--decode-warmup-seconds",
"3",
"--display-mode",
"plain",
"--no-resume",
"--output",
"benchmark_results.json",
"--request-count",
"4",
"--warmup-request-count",
"1"
],
"started": "2026-10-17T13:11:31.105611+00:00",
"finished": "2026-10-17T13:14:20.871259+00:00",
"raw_file": "runs/low-finite-ctx0-attempt3/benchmark_results.json"
}
],
"tool_commit": "497350a81a77d7d11b4e418b2a26c30c3065949a",
"agent_audits": [
{
"source": "upstream shared_usage_token_count increments, common upstream window",
"concurrency": 1,
"benchmark_mode": "duration",
"context_tokens": 0,
"aggregate_source": "openai_continuous_usage",
"completion_tokens": [
1250
],
"tokens_per_second": [
41.72178474486878485549823381006717681884765625
],
"requested_duration": 30,
"measurement_seconds": 29.960367411025796258172704256139695644378662109375,
"request_count_target": 0,
"aggregate_tokens_per_second": 41.72178474486878485549823381006717681884765625,
"measurement_start_monotonic": 508570.595817135996185243129730224609375,
"reported_measurement_seconds_rounded": 29.960367000000001524995241197757422924041748046875
},
{
"source": "upstream shared_usage_token_count increments, common upstream window",
"concurrency": 2,
"benchmark_mode": "duration",
"context_tokens": 0,
"aggregate_source": "openai_continuous_usage",
"completion_tokens": [
941,
941
],
"tokens_per_second": [
31.416868917281277617803425528109073638916015625,
31.416868917281277617803425528109073638916015625
],
"requested_duration": 30,
"measurement_seconds": 29.952061819960363209247589111328125,
"request_count_target": 0,
"aggregate_tokens_per_second": 62.83373783456255523560685105621814727783203125,
"measurement_start_monotonic": 508608.1366135050193406641483306884765625,
"reported_measurement_seconds_rounded": 29.952062000000001518174030934460461139678955078125
},
{
"source": "upstream shared_usage_token_count increments, common upstream window",
"concurrency": 4,
"benchmark_mode": "duration",
"context_tokens": 0,
"aggregate_source": "openai_continuous_usage",
"completion_tokens": [
618,
618,
618,
618
],
"tokens_per_second": [
20.67562263150376367093485896475613117218017578125,
20.67562263150376367093485896475613117218017578125,
20.67562263150376367093485896475613117218017578125,
20.67562263150376367093485896475613117218017578125
],
"requested_duration": 30,
"measurement_seconds": 29.890272762975655496120452880859375,
"request_count_target": 0,
"aggregate_tokens_per_second": 82.702490526015054683739435859024524688720703125,
"measurement_start_monotonic": 508645.709708884009160101413726806640625,
"reported_measurement_seconds_rounded": 29.89027300000000053614712669514119625091552734375
},
{
"source": "upstream shared_usage_token_count increments, common upstream window",
"concurrency": 1,
"benchmark_mode": "duration",
"context_tokens": 32768,
"aggregate_source": "openai_continuous_usage",
"completion_tokens": [
1428
],
"tokens_per_second": [
47.6047477119508783971468801610171794891357421875
],
"requested_duration": 30,
"measurement_seconds": 29.997008042992092669010162353515625,
"request_count_target": 0,
"aggregate_tokens_per_second": 47.6047477119508783971468801610171794891357421875,
"measurement_start_monotonic": 509019.14180836197920143604278564453125,
"reported_measurement_seconds_rounded": 29.99700800000000100453689810819923877716064453125
},
{
"source": "upstream shared_usage_token_count increments, common upstream window",
"concurrency": 2,
"benchmark_mode": "duration",
"context_tokens": 32768,
"aggregate_source": "openai_continuous_usage",
"completion_tokens": [
670,
670
],
"tokens_per_second": [
22.364381782229838080411354894749820232391357421875,
22.364381782229838080411354894749820232391357421875
],
"requested_duration": 30,
"measurement_seconds": 29.9583510299562476575374603271484375,
"request_count_target": 0,
"aggregate_tokens_per_second": 44.72876356445967616082270978949964046478271484375,
"measurement_start_monotonic": 509056.2291248540277592837810516357421875,
"reported_measurement_seconds_rounded": 29.95835100000000039699443732388317584991455078125
},
{
"source": "upstream shared_usage_token_count increments, common upstream window",
"concurrency": 4,
"benchmark_mode": "duration",
"context_tokens": 32768,
"aggregate_source": "openai_continuous_usage",
"completion_tokens": [
510,
510,
510,
510
],
"tokens_per_second": [
17.045284710726281929282777127809822559356689453125,
17.045284710726281929282777127809822559356689453125,
17.045284710726281929282777127809822559356689453125,
17.045284710726281929282777127809822559356689453125
],
"requested_duration": 30,
"measurement_seconds": 29.920298115001063621321009122766554355621337890625,
"request_count_target": 0,
"aggregate_tokens_per_second": 68.1811388429051277171311085112392902374267578125,
"measurement_start_monotonic": 509093.327755630016326904296875,
"reported_measurement_seconds_rounded": 29.920297999999998950215740478597581386566162109375
},
{
"source": "upstream shared_usage_token_count increments, common upstream window",
"concurrency": 1,
"benchmark_mode": "duration",
"context_tokens": 131072,
"aggregate_source": "openai_continuous_usage",
"completion_tokens": [
1299
],
"tokens_per_second": [
43.36507712997916996755520813167095184326171875
],
"requested_duration": 30,
"measurement_seconds": 29.95497958199121057987213134765625,
"request_count_target": 0,
"aggregate_tokens_per_second": 43.36507712997916996755520813167095184326171875,
"measurement_start_monotonic": 509699.71860957401804625988006591796875,
"reported_measurement_seconds_rounded": 29.95497999999999905185177340172231197357177734375
},
{
"source": "upstream shared_usage_token_count increments, common upstream window",
"concurrency": 2,
"benchmark_mode": "duration",
"context_tokens": 131072,
"aggregate_source": "openai_continuous_usage",
"completion_tokens": [
648,
648
],
"tokens_per_second": [
21.690901610870216842386071220971643924713134765625,
21.690901610870216842386071220971643924713134765625
],
"requested_duration": 30,
"measurement_seconds": 29.8742768569500185549259185791015625,
"request_count_target": 0,
"aggregate_tokens_per_second": 43.38180322174043368477214244194328784942626953125,
"measurement_start_monotonic": 509737.261891062022186815738677978515625,
"reported_measurement_seconds_rounded": 29.87427699999999930469130049459636211395263671875
},
{
"source": "upstream shared_usage_token_count increments, common upstream window",
"concurrency": 4,
"benchmark_mode": "duration",
"context_tokens": 131072,
"aggregate_source": "openai_continuous_usage",
"completion_tokens": [
421,
421,
421,
416
],
"tokens_per_second": [
14.0647671140771581832495940034277737140655517578125,
14.0647671140771581832495940034277737140655517578125,
14.0647671140771581832495940034277737140655517578125,
13.8977271245988074355182106955908238887786865234375
],
"requested_duration": 30,
"measurement_seconds": 29.932952077011577785015106201171875,
"request_count_target": 0,
"aggregate_tokens_per_second": 56.09202846683027843255331390537321567535400390625,
"measurement_start_monotonic": 509776.800233462010510265827178955078125,
"reported_measurement_seconds_rounded": 29.9329520000000002255546860396862030029296875
}
],
"slot_ceiling": 4,
"tool_version": "0.8.0",
"burst_results": [
{
"ttft_avg": 0.28998448149650357663631439208984375,
"ttft_p50": 0.2867771349847316741943359375,
"ttft_p90": 0.296768423391040425673992331212502904236316680908203125,
"ttft_p99": 0.300271744134370244250220594040001742541790008544921875,
"wall_time": 48.0086075330036692321300506591796875,
"num_errors": 0,
"concurrency": 1,
"underfilled": false,
"ready_reason": "request_count=4, warmup_request_count=1",
"total_tokens": 2048,
"aggregate_tps": 42.659016898003045525911147706210613250732421875,
"loop_detected": false,
"num_completed": 4,
"request_count": 4,
"avg_queue_reqs": 0,
"benchmark_mode": "request-count",
"context_tokens": 0,
"failure_reason": "",
"max_queue_reqs": 0,
"queue_fraction": 0,
"timeout_reason": "",
"accept_norm_tps": 0,
"request_samples": [
{
"ttft": 0.2857226540218107402324676513671875,
"latency": 12.0110251500154845416545867919921875,
"completed": true,
"input_tokens": 78,
"output_tokens": 512,
"output_tps_per_user": 43.58096519681257774436744512058794498443603515625,
"time_to_second_token": 0.05840984196402132511138916015625,
"e2e_output_tps_per_user": 42.62750211619862739098607562482357025146484375,
"inter_token_latency_avg": 0.0229457974481285208179226486890911473892629146575927734375,
"chunk_inter_token_latency_avg": 0.0229907892078307328487252192417145124636590480804443359375
},
{
"ttft": 0.2858685299870558083057403564453125,
"latency": 11.9661789259989745914936065673828125,
"completed": true,
"input_tokens": 78,
"output_tokens": 512,
"output_tps_per_user": 43.74883737460212529413183801807463169097900390625,
"time_to_second_token": 0.056930845021270215511322021484375,
"e2e_output_tps_per_user": 42.7872592551307349140188307501375675201416015625,
"inter_token_latency_avg": 0.022857750285737610251857176990597508847713470458984375,
"chunk_inter_token_latency_avg": 0.022902569403944937620476451911599724553525447845458984375
},
{
"ttft": 0.2876857399824075400829315185546875,
"latency": 11.98849011096172034740447998046875,
"completed": true,
"input_tokens": 78,
"output_tokens": 512,
"output_tps_per_user": 43.6722112256997974100158899091184139251708984375,
"time_to_second_token": 0.0593637839774601161479949951171875,
"e2e_output_tps_per_user": 42.707630006872250305605120956897735595703125,
"inter_token_latency_avg": 0.022897855911896893010837317206096486188471317291259765625,
"chunk_inter_token_latency_avg": 0.0229427536685868870003357500308993621729314327239990234375
},
{
"ttft": 0.3006610019947402179241180419921875,
"latency": 11.9821277430164627730846405029296875,
"completed": true,
"input_tokens": 78,
"output_tokens": 512,
"output_tps_per_user": 43.74450668985983980974197038449347019195556640625,
"time_to_second_token": 0.0574142729747109115123748779296875,
"e2e_output_tps_per_user": 42.730307252683786600755411200225353240966796875,
"inter_token_latency_avg": 0.02286001319182333346713420496598700992763042449951171875,
"chunk_inter_token_latency_avg": 0.0229048367471014167884302281663622125051915645599365234375
}
],
"warmup_duration": 0,
"aggregate_source": "openai_completed_usage",
"avg_running_reqs": 0,
"capacity_limited": false,
"hardware_summary": {
"samples": 22,
"gpu_count": 2,
"temp_avg_c": 55.5499999999999971578290569595992565155029296875,
"temp_max_c": 57,
"vram_total_mb": 131072,
"cpu_temp_max_c": 67.75,
"cpu_util_avg_pct": 9.800000000000000710542735760100185871124267578125,
"duration_seconds": 46.42399999999999948840923025272786617279052734375,
"gpu_util_avg_pct": 87.43000000000000682121026329696178436279296875,
"gpu_util_max_pct": 93,
"mem_util_avg_pct": 14.449999999999999289457264239899814128875732421875,
"mem_util_max_pct": 16,
"pcie_rx_avg_mb_s": 497.94999999999998863131622783839702606201171875,
"pcie_rx_max_mb_s": 650,
"pcie_tx_avg_mb_s": 467.55000000000001136868377216160297393798828125,
"pcie_tx_max_mb_s": 557,
"vram_used_avg_mb": 128774,
"vram_used_max_mb": 128774,
"power_total_avg_w": 336.8999999999999772626324556767940521240234375,
"power_total_max_w": 369.1399999999999863575794734060764312744140625,
"vram_used_avg_pct": 98.25,
"vram_used_max_pct": 98.25,
"power_limit_total_w": 360
},
"loop_diagnostics": [],
"max_running_reqs": 0,
"warmup_timed_out": false,
"input_seq_len_avg": 78,
"output_seq_len_avg": 512,
"output_seq_len_p50": 512,
"output_seq_len_p90": 512,
"output_seq_len_p99": 512,
"server_spec_drafts": 0,
"server_steps_per_s": 0,
"server_utilization": 0,
"accept_norm_ref_len": 0,
"measurement_seconds": 48.00860800000000239151631831191480159759521484375,
"per_request_avg_tps": 42.659016898003045525911147706210613250732421875,
"request_latency_avg": 11.986955482498160563409328460693359375,
"request_latency_p50": 11.98530892698909156024456024169921875,
"request_latency_p90": 12.0042646382993556386509226285852491855621337890625,
"request_latency_p99": 12.010349098843871473718536435626447200775146484375,
"server_engine_steps": 0,
"client_output_tokens": 2048,
"request_count_target": 4,
"server_output_tokens": 0,
"warmup_request_count": 1,
"effective_concurrency": 0,
"server_gen_throughput": 0,
"loop_detection_enabled": true,
"server_spec_pos_accept": [],
"completed_request_count": 4,
"inter_token_latency_avg": 0.0228903542093965885195760989745394908823072910308837890625,
"inter_token_latency_p50": 0.0228789345518601132389857610860417480580508708953857421875,
"inter_token_latency_p90": 0.0229314149872590321288523540488313301466405391693115234375,
"inter_token_latency_p99": 0.0229443592020415719490156192250651656650006771087646484375,
"output_tps_per_user_avg": 43.68663012174358328820744645781815052032470703125,
"output_tps_per_user_p50": 43.7083589577798221625926089473068714141845703125,
"output_tps_per_user_p90": 43.74753816917944249098582076840102672576904296875,
"output_tps_per_user_p99": 43.7487074540598541716462932527065277099609375,
"server_spec_accept_rate": 0,
"loop_diagnostics_omitted": 0,
"measurement_wall_seconds": 0,
"server_spec_draft_tokens": 0,
"time_to_second_token_avg": 0.058029685984365642070770263671875,
"time_to_second_token_p50": 0.05791205746936611831188201904296875,
"time_to_second_token_p90": 0.05907760137342847606145568306601489894092082977294921875,
"time_to_second_token_p99": 0.05933516571705695075156228313062456436455249786376953125,
"server_spec_accept_length": 0,
"e2e_output_tps_per_user_avg": 42.7131746577213533555550384335219860076904296875,
"e2e_output_tps_per_user_p50": 42.7189686297780184531802660785615444183349609375,
"e2e_output_tps_per_user_p90": 42.77017365439665042003980488516390323638916015625,
"e2e_output_tps_per_user_p99": 42.7855506950573243329927208833396434783935546875,
"server_accept_len_effective": 0,
"server_spec_accepted_tokens": 0,
"chunk_inter_token_latency_avg": 0.0229352372568659944318536503260475001297891139984130859375,
"chunk_inter_token_latency_p50": 0.0229237952078441518943829890986307873390614986419677734375,
"chunk_inter_token_latency_p90": 0.0229763785460575797880977688691928051412105560302734375,
"chunk_inter_token_latency_p99": 0.02298934814165341788960716939982376061379909515380859375
},
{
"ttft_avg": 0.502745960999163798987865447998046875,
"ttft_p50": 0.54368438650271855294704437255859375,
"ttft_p90": 0.60251888839411549270153045654296875,
"ttft_p99": 0.625151887346873991191387176513671875,
"wall_time": 36.7314242179854772984981536865234375,
"num_errors": 0,
"concurrency": 2,
"underfilled": false,
"ready_reason": "request_count=4, warmup_request_count=1",
"total_tokens": 2048,
"aggregate_tps": 55.75607381423561292876911466009914875030517578125,
"loop_detected": false,
"num_completed": 4,
"request_count": 4,
"avg_queue_reqs": 0,
"benchmark_mode": "request-count",
"context_tokens": 0,
"failure_reason": "",
"max_queue_reqs": 0,
"queue_fraction": 0,
"timeout_reason": "",
"accept_norm_tps": 0,
"request_samples": [
{
"ttft": 0.5438407429610379040241241455078125,
"latency": 16.7854316169978119432926177978515625,
"completed": true,
"input_tokens": 78,
"output_tokens": 512,
"output_tps_per_user": 31.46243517418397317442213534377515316009521484375,
"time_to_second_token": 0.0887062770198099315166473388671875,
"e2e_output_tps_per_user": 30.502641319125917362953259726054966449737548828125,
"inter_token_latency_avg": 0.03178393517424026037954121193251921795308589935302734375,
"chunk_inter_token_latency_avg": 0.0318462566157583781834006231292732991278171539306640625
},
{
"ttft": 0.29594840598292648792266845703125,
"latency": 19.926191659993492066860198974609375,
"completed": true,
"input_tokens": 78,
"output_tokens": 512,
"output_tps_per_user": 26.03126173159366629761279909871518611907958984375,
"time_to_second_token": 0.3308161569875665009021759033203125,
"e2e_output_tps_per_user": 25.694824617589130610895153949968516826629638671875,
"inter_token_latency_avg": 0.0384153488336801685054666677388013340532779693603515625,
"chunk_inter_token_latency_avg": 0.0384906730470795377385684332693926990032196044921875
},
{
"ttft": 0.543528030044399201869964599609375,
"latency": 16.785209476016461849212646484375,
"completed": true,
"input_tokens": 78,
"output_tokens": 512,
"output_tps_per_user": 31.46225972352930710940199787728488445281982421875,
"time_to_second_token": 0.09044882399030029773712158203125,
"e2e_output_tps_per_user": 30.50304500111070638013188727200031280517578125,
"inter_token_latency_avg": 0.031784112418732017590894400882461923174560070037841796875,
"chunk_inter_token_latency_avg": 0.031846434207788355819079839648111374117434024810791015625
},
{
"ttft": 0.62766666500829160213470458984375,
"latency": 18.77810777700506150722503662109375,
"completed": true,
"input_tokens": 78,
"output_tokens": 512,
"output_tps_per_user": 28.153585736395566385681377141736447811126708984375,
"time_to_second_token": 0.096384732984006404876708984375,
"e2e_output_tps_per_user": 27.265793022392557531929924152791500091552734375,
"inter_token_latency_avg": 0.03551945423091344700150528979065711610019207000732421875,
"chunk_inter_token_latency_avg": 0.035589100219601509100808556240735924802720546722412109375
}
],
"warmup_duration": 0,
"aggregate_source": "openai_completed_usage",
"avg_running_reqs": 0,
"capacity_limited": false,
"hardware_summary": {
"samples": 17,
"gpu_count": 2,
"temp_avg_c": 54.78999999999999914734871708787977695465087890625,
"temp_max_c": 57,
"vram_total_mb": 131072,
"cpu_temp_max_c": 68.3799999999999954525264911353588104248046875,
"cpu_util_avg_pct": 22.75,
"duration_seconds": 35.39999999999999857891452847979962825775146484375,
"gpu_util_avg_pct": 79.409999999999996589394868351519107818603515625,
"gpu_util_max_pct": 90,
"mem_util_avg_pct": 11.6199999999999992184029906638897955417633056640625,
"mem_util_max_pct": 15,
"pcie_rx_avg_mb_s": 382.529999999999972715158946812152862548828125,
"pcie_rx_max_mb_s": 587,
"pcie_tx_avg_mb_s": 344.240000000000009094947017729282379150390625,
"pcie_tx_max_mb_s": 514,
"vram_used_avg_mb": 128774,
"vram_used_max_mb": 128774,
"power_total_avg_w": 307.91000000000002501110429875552654266357421875,
"power_total_max_w": 357.1299999999999954525264911353588104248046875,
"vram_used_avg_pct": 98.25,
"vram_used_max_pct": 98.25,
"power_limit_total_w": 360
},
"loop_diagnostics": [],
"max_running_reqs": 0,
"warmup_timed_out": false,
"input_seq_len_avg": 78,
"output_seq_len_avg": 512,
"output_seq_len_p50": 512,
"output_seq_len_p90": 512,
"output_seq_len_p99": 512,
"server_spec_drafts": 0,
"server_steps_per_s": 0,
"server_utilization": 0,
"accept_norm_ref_len": 0,
"measurement_seconds": 36.73142399999999696547092753462493419647216796875,
"per_request_avg_tps": 27.878036907117806464384557330049574375152587890625,
"request_latency_avg": 18.068735132503206841647624969482421875,
"request_latency_p50": 17.78176969700143672525882720947265625,
"request_latency_p90": 19.58176649509696432005512178875505924224853515625,
"request_latency_p99": 19.89174914350383716055148397572338581085205078125,
"server_engine_steps": 0,
"client_output_tokens": 2048,
"request_count_target": 4,
"server_output_tokens": 0,
"warmup_request_count": 1,
"effective_concurrency": 0,
"server_gen_throughput": 0,
"loop_detection_enabled": true,
"server_spec_pos_accept": [],
"completed_request_count": 4,
"inter_token_latency_avg": 0.03437571266439147510407536856291699223220348358154296875,
"inter_token_latency_p50": 0.03365178332482272882675289338294533081352710723876953125,
"inter_token_latency_p90": 0.03754658045285015621761459669869509525597095489501953125,
"inter_token_latency_p99": 0.0383284719955971631133451182904536835849285125732421875,
"output_tps_per_user_avg": 29.277385591425627353601157665252685546875,
"output_tps_per_user_p50": 29.80792272996243497118484810926020145416259765625,
"output_tps_per_user_p90": 31.462382538987572644373358343727886676788330078125,
"output_tps_per_user_p99": 31.46242991066433347668862552382051944732666015625,
"server_spec_accept_rate": 0,
"loop_diagnostics_omitted": 0,
"measurement_wall_seconds": 0,
"server_spec_draft_tokens": 0,
"time_to_second_token_avg": 0.1515889977454207837581634521484375,
"time_to_second_token_p50": 0.093416778487153351306915283203125,
"time_to_second_token_p90": 0.260486729786498505401226566391414962708950042724609375,
"time_to_second_token_p99": 0.323783214267459651392044861495378427207469940185546875,
"server_spec_accept_length": 0,
"e2e_output_tps_per_user_avg": 28.49157599005457797147755627520382404327392578125,
"e2e_output_tps_per_user_p50": 28.88421717075923567108475253917276859283447265625,
"e2e_output_tps_per_user_p90": 30.5029238965152700302496668882668018341064453125,
"e2e_output_tps_per_user_p99": 30.50303289065116274514366523362696170806884765625,
"server_accept_len_effective": 0,
"server_spec_accepted_tokens": 0,
"chunk_inter_token_latency_avg": 0.034443116022556945210464363071878324262797832489013671875,
"chunk_inter_token_latency_p50": 0.0337177672136949324599441979444236494600772857666015625,
"chunk_inter_token_latency_p90": 0.037620201198836127065572298988627153448760509490966796875,
"chunk_inter_token_latency_p99": 0.03840362586225519458960064866914763115346431732177734375
},
{
"ttft_avg": 0.88390277299913577735424041748046875,
"ttft_p50": 0.79969039800926111638545989990234375,
"ttft_p90": 1.041038761404343038208253346965648233890533447265625,
"ttft_p99": 1.133242849332745993962134889443404972553253173828125,
"wall_time": 26.3931710949982516467571258544921875,
"num_errors": 0,
"concurrency": 4,
"underfilled": false,
"ready_reason": "request_count=4, warmup_request_count=1",
"total_tokens": 2048,
"aggregate_tps": 77.59582933890482081551454029977321624755859375,
"loop_detected": false,
"num_completed": 4,
"request_count": 4,
"avg_queue_reqs": 0,
"benchmark_mode": "request-count",
"context_tokens": 0,
"failure_reason": "",
"max_queue_reqs": 0,
"queue_fraction": 0,
"timeout_reason": "",
"accept_norm_tps": 0,
"request_samples": [
{
"ttft": 0.7927425479865632951259613037109375,
"latency": 26.1162523240200243890285491943359375,
"completed": true,
"input_tokens": 78,
"output_tokens": 512,
"output_tps_per_user": 20.17887743521310284222636255435645580291748046875,
"time_to_second_token": 0.34762889402918517589569091796875,
"e2e_output_tps_per_user": 19.60465053131286339294092613272368907928466796875,
"inter_token_latency_avg": 0.04955677059889131153003205554341548122465610504150390625,
"chunk_inter_token_latency_avg": 0.049653940737320510778740612067849724553525447845458984375
},
{
"ttft": 0.7973896699841134250164031982421875,
"latency": 26.115989694953896105289459228515625,
"completed": true,
"input_tokens": 78,
"output_tokens": 512,
"output_tps_per_user": 20.1827904977384235962745151482522487640380859375,
"time_to_second_token": 0.345964192994870245456695556640625,
"e2e_output_tps_per_user": 19.604847680688436639684368856251239776611328125,
"inter_token_latency_avg": 0.049547162475479029908331796150378067977726459503173828125,
"chunk_inter_token_latency_avg": 0.04964431377445055471042678618687205016613006591796875
},
{
"ttft": 0.8019911260344088077545166015625,
"latency": 26.115878926008008420467376708984375,
"completed": true,
"input_tokens": 78,
"output_tokens": 512,
"output_tps_per_user": 20.1865475598944925650357617996633052825927734375,
"time_to_second_token": 0.3412027339800260961055755615234375,
"e2e_output_tps_per_user": 19.60493083348287512990282266400754451751708984375,
"inter_token_latency_avg": 0.04953794090014403128652276109278318472206592559814453125,
"chunk_inter_token_latency_avg": 0.0496350741175952958617045851497096009552478790283203125
},
{
"ttft": 1.14348774799145758152008056640625,
"latency": 26.187336909002624452114105224609375,
"completed": true,
"input_tokens": 78,
"output_tokens": 512,
"output_tps_per_user": 20.404211697438920936065187561325728893280029296875,
"time_to_second_token": 0.1287860120064578950405120849609375,
"e2e_output_tps_per_user": 19.551434411949912828276865184307098388671875,
"inter_token_latency_avg": 0.049009489551880953961404685514935408718883991241455078125,
"chunk_inter_token_latency_avg": 0.049105586590217971398875107524872873909771442413330078125
}
],
"warmup_duration": 0,
"aggregate_source": "openai_completed_usage",
"avg_running_reqs": 0,
"capacity_limited": false,
"hardware_summary": {
"samples": 12,
"gpu_count": 2,
"temp_avg_c": 55,
"temp_max_c": 57,
"vram_total_mb": 131072,
"cpu_temp_max_c": 70.25,
"cpu_util_avg_pct": 18.8900000000000005684341886080801486968994140625,
"duration_seconds": 24.34700000000000130739863379858434200286865234375,
"gpu_util_avg_pct": 82.6700000000000017053025658242404460906982421875,
"gpu_util_max_pct": 93,
"mem_util_avg_pct": 10.75,
"mem_util_max_pct": 13,
"pcie_rx_avg_mb_s": 512.5,
"pcie_rx_max_mb_s": 885,
"pcie_tx_avg_mb_s": 483.42000000000001591615728102624416351318359375,
"pcie_tx_max_mb_s": 770,
"vram_used_avg_mb": 128774,
"vram_used_max_mb": 128774,
"power_total_avg_w": 329.30000000000001136868377216160297393798828125,
"power_total_max_w": 369.3999999999999772626324556767940521240234375,
"vram_used_avg_pct": 98.25,
"vram_used_max_pct": 98.25,
"power_limit_total_w": 360
},
"loop_diagnostics": [],
"max_running_reqs": 0,
"warmup_timed_out": false,
"input_seq_len_avg": 78,
"output_seq_len_avg": 512,
"output_seq_len_p50": 512,
"output_seq_len_p90": 512,
"output_seq_len_p99": 512,
"server_spec_drafts": 0,
"server_steps_per_s": 0,
"server_utilization": 0,
"accept_norm_ref_len": 0,
"measurement_seconds": 26.393170999999998826979208388365805149078369140625,
"per_request_avg_tps": 19.3989573347262052038786350749433040618896484375,
"request_latency_avg": 26.133864463496138341724872589111328125,
"request_latency_p50": 26.11612100948696024715900421142578125,
"request_latency_p90": 26.166011533507845143731174175627529621124267578125,
"request_latency_p99": 26.18520437145314616600444423966109752655029296875,
"server_engine_steps": 0,
"client_output_tokens": 2048,
"request_count_target": 4,
"server_output_tokens": 0,
"warmup_request_count": 1,
"effective_concurrency": 0,
"server_gen_throughput": 0,
"loop_detection_enabled": true,
"server_spec_pos_accept": [],
"completed_request_count": 4,
"inter_token_latency_avg": 0.0494128408815988351410197765289922244846820831298828125,
"inter_token_latency_p50": 0.04954255168781153406687423057519481517374515533447265625,
"inter_token_latency_p90": 0.049553888161867627737411368116227095015347003936767578125,
"inter_token_latency_p99": 0.049556482355188939681323034847082453779876232147216796875,
"output_tps_per_user_avg": 20.2381067975712340967220370657742023468017578125,
"output_tps_per_user_p50": 20.1846690288164580806551384739577770233154296875,
"output_tps_per_user_p90": 20.33891245617559206948499195277690887451171875,
"output_tps_per_user_p99": 20.397681773312587694135800120420753955841064453125,
"server_spec_accept_rate": 0,
"loop_diagnostics_omitted": 0,
"measurement_wall_seconds": 0,
"server_spec_draft_tokens": 0,
"time_to_second_token_avg": 0.2908954582526348531246185302734375,
"time_to_second_token_p50": 0.34358346348744817078113555908203125,
"time_to_second_token_p90": 0.347129483718890685661762063318747095763683319091796875,
"time_to_second_token_p99": 0.347578952998155721321182909377967007458209991455078125,
"server_spec_accept_length": 0,
"e2e_output_tps_per_user_avg": 19.591465864358521997701245709322392940521240234375,
"e2e_output_tps_per_user_p50": 19.604749106000650016312647494487464427947998046875,
"e2e_output_tps_per_user_p90": 19.604905887644545003922758041881024837493896484375,
"e2e_output_tps_per_user_p99": 19.604928338899043183118919841945171356201171875,
"server_accept_len_effective": 0,
"server_spec_accepted_tokens": 0,
"chunk_inter_token_latency_avg": 0.04950972880489608318743677273232606239616870880126953125,
"chunk_inter_token_latency_p50": 0.04963969394602292528606568566829082556068897247314453125,
"chunk_inter_token_latency_p90": 0.04965105264845952326435707391283358447253704071044921875,
"chunk_inter_token_latency_p99": 0.049653651928434412721191648643070948310196399688720703125
}
],
"engine_version": "0f685aa + row32 four-slot profile",
"model_revision": "332ab457b709b7ba30dd9a448be5de03b80a7ac9",
"profile_digest": "7c6da98c10332410ccf838449ab87fe86c6907c351bdd72958685e034f593031",
"run_conditions": {
"note": "Dedicated quiet intervals on the shared service; yield to public router traffic; CPU PSI avg10 <2%. Effort via model alias. Q8 GPU KV; Weightless GLP-44 alpha2. One run per cell. No cold-prefill claim.",
"via_form": true,
"shared_server": true,
"repeats_per_cell": 1,
"duration_per_cell": 30
},
"reasoning_effort": "low",
"burst_agent_audits": [
{
"source": "upstream shared_usage_token_count increments, common upstream window",
"concurrency": 1,
"benchmark_mode": "request-count",
"context_tokens": 0,
"aggregate_source": "openai_completed_usage",
"completion_tokens": [
2048
],
"tokens_per_second": [
42.659016898003045525911147706210613250732421875
],
"requested_duration": 30,
"measurement_seconds": 48.00860753300367633755740826018154621124267578125,
"request_count_target": 4,
"aggregate_tokens_per_second": 42.659016898003045525911147706210613250732421875,
"measurement_start_monotonic": 510981.9057604349800385534763336181640625,
"reported_measurement_seconds_rounded": 48.00860800000000239151631831191480159759521484375
},
{
"source": "upstream shared_usage_token_count increments, common upstream window",
"concurrency": 2,
"benchmark_mode": "request-count",
"context_tokens": 0,
"aggregate_source": "openai_completed_usage",
"completion_tokens": [
1024,
1024
],
"tokens_per_second": [
27.878036907117806464384557330049574375152587890625,
27.878036907117806464384557330049574375152587890625
],
"requested_duration": 30,
"measurement_seconds": 36.7314242179854772984981536865234375,
"request_count_target": 4,
"aggregate_tokens_per_second": 55.75607381423561292876911466009914875030517578125,
"measurement_start_monotonic": 511043.98687705001793801784515380859375,
"reported_measurement_seconds_rounded": 36.73142399999999696547092753462493419647216796875
},
{
"source": "upstream shared_usage_token_count increments, common upstream window",
"concurrency": 4,
"benchmark_mode": "request-count",
"context_tokens": 0,
"aggregate_source": "openai_completed_usage",
"completion_tokens": [
512,
512,
512,
512
],
"tokens_per_second": [
19.3989573347262052038786350749433040618896484375,
19.3989573347262052038786350749433040618896484375,
19.3989573347262052038786350749433040618896484375,
19.3989573347262052038786350749433040618896484375
],
"requested_duration": 30,
"measurement_seconds": 26.3931710949982516467571258544921875,
"request_count_target": 4,
"aggregate_tokens_per_second": 77.59582933890482081551454029977321624755859375,
"measurement_start_monotonic": 511095.0394925180007703602313995361328125,
"reported_measurement_seconds_rounded": 26.393170999999998826979208388365805149078369140625
}
],
"engine_source_commit": "fa0adf1899015f692ddce30391ee7e740a0910c9"
}Run file preserved verbatim in the corpus (unknown keys never dropped) · derived rows regenerate from it on import.