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-max-effort-on-2x-cmp170hx-exl3-tp2-four-slots variant: GLM-5.3-Flash
Uploaded report
report.md (markdown, shown as-is)
# GLM-5.3-Flash, max 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",
"results": [
{
"ttft_avg": 0.331802289001643657684326171875,
"ttft_p50": 0.331802289001643657684326171875,
"ttft_p90": 0.331802289001643657684326171875,
"ttft_p99": 0.331802289001643657684326171875,
"wall_time": 35.5181007059873081743717193603515625,
"num_errors": 0,
"concurrency": 1,
"underfilled": false,
"ready_reason": "metrics_unavailable, active_streams=1/1, stable=3.0s",
"total_tokens": 1242,
"aggregate_tps": 41.4733142771386127378718811087310314178466796875,
"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": 0.331802289001643657684326171875,
"latency": 0,
"completed": false,
"input_tokens": 78,
"output_tokens": 1528,
"output_tps_per_user": 43.47085988170508841221817419864237308502197265625,
"time_to_second_token": 0.0605343150091357529163360595703125,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.0230039157891342889439290075870303553529083728790283203125,
"chunk_inter_token_latency_avg": 0.0230039157891342889439290075870303553529083728790283203125
}
],
"warmup_duration": 5.51199999999999956656893118633888661861419677734375,
"aggregate_source": "openai_continuous_usage",
"avg_running_reqs": 0,
"capacity_limited": false,
"hardware_summary": {
"samples": 14,
"gpu_count": 2,
"temp_avg_c": 54.28999999999999914734871708787977695465087890625,
"temp_max_c": 56,
"vram_total_mb": 131072,
"cpu_temp_max_c": 66.6200000000000045474735088646411895751953125,
"cpu_util_avg_pct": 10.6899999999999995026200849679298698902130126953125,
"duration_seconds": 28.70700000000000073896444519050419330596923828125,
"gpu_util_avg_pct": 87.43000000000000682121026329696178436279296875,
"gpu_util_max_pct": 92,
"mem_util_avg_pct": 14.1099999999999994315658113919198513031005859375,
"mem_util_max_pct": 15,
"pcie_rx_avg_mb_s": 525.2100000000000363797880709171295166015625,
"pcie_rx_max_mb_s": 628,
"pcie_tx_avg_mb_s": 458.70999999999997953636921010911464691162109375,
"pcie_tx_max_mb_s": 513,
"vram_used_avg_mb": 126720,
"vram_used_max_mb": 126720,
"power_total_avg_w": 340.5,
"power_total_max_w": 364.04000000000002046363078989088535308837890625,
"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": 1528,
"output_seq_len_p50": 1528,
"output_seq_len_p90": 1528,
"output_seq_len_p99": 1528,
"server_spec_drafts": 0,
"server_steps_per_s": 0,
"server_utilization": 0,
"accept_norm_ref_len": 0,
"measurement_seconds": 29.946967999999998255589161999523639678955078125,
"per_request_avg_tps": 41.4733142771386127378718811087310314178466796875,
"request_latency_avg": 0,
"request_latency_p50": 0,
"request_latency_p90": 0,
"request_latency_p99": 0,
"server_engine_steps": 0,
"client_output_tokens": 1242,
"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": 0.0230039157891342889439290075870303553529083728790283203125,
"inter_token_latency_p50": 0.0230039157891342889439290075870303553529083728790283203125,
"inter_token_latency_p90": 0.0230039157891342889439290075870303553529083728790283203125,
"inter_token_latency_p99": 0.0230039157891342889439290075870303553529083728790283203125,
"output_tps_per_user_avg": 43.47085988170508841221817419864237308502197265625,
"output_tps_per_user_p50": 43.47085988170508841221817419864237308502197265625,
"output_tps_per_user_p90": 43.47085988170508841221817419864237308502197265625,
"output_tps_per_user_p99": 43.47085988170508841221817419864237308502197265625,
"server_spec_accept_rate": 0,
"loop_diagnostics_omitted": 0,
"measurement_wall_seconds": 30.000153999999998433168002520687878131866455078125,
"server_spec_draft_tokens": 0,
"time_to_second_token_avg": 0.0605343150091357529163360595703125,
"time_to_second_token_p50": 0.0605343150091357529163360595703125,
"time_to_second_token_p90": 0.0605343150091357529163360595703125,
"time_to_second_token_p99": 0.0605343150091357529163360595703125,
"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": 0.0230039157891342889439290075870303553529083728790283203125,
"chunk_inter_token_latency_p50": 0.0230039157891342889439290075870303553529083728790283203125,
"chunk_inter_token_latency_p90": 0.0230039157891342889439290075870303553529083728790283203125,
"chunk_inter_token_latency_p99": 0.0230039157891342889439290075870303553529083728790283203125
},
{
"ttft_avg": 0.5956682324758730828762054443359375,
"ttft_p50": 0.5956682324758730828762054443359375,
"ttft_p90": 0.5961239576688968089257514293421991169452667236328125,
"ttft_p99": 0.5962264958373271195313236603396944701671600341796875,
"wall_time": 35.520828926004469394683837890625,
"num_errors": 0,
"concurrency": 2,
"underfilled": false,
"ready_reason": "metrics_unavailable, active_streams=2/2, stable=3.0s",
"total_tokens": 1956,
"aggregate_tps": 65.3581817056967366852404666133224964141845703125,
"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": 0.5950985759845934808254241943359375,
"latency": 0,
"completed": false,
"input_tokens": 78,
"output_tokens": 1150,
"output_tps_per_user": 32.974414245152814828543341718614101409912109375,
"time_to_second_token": 0.0892514240113086998462677001953125,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.0303265432576106629125600733232204220257699489593505859375,
"chunk_inter_token_latency_avg": 0.0303265432576106629125600733232204220257699489593505859375
},
{
"ttft": 0.5962378889671526849269866943359375,
"latency": 0,
"completed": false,
"input_tokens": 78,
"output_tokens": 1150,
"output_tps_per_user": 32.97588785625006124746505520306527614593505859375,
"time_to_second_token": 0.0891597699956037104129791259765625,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.0303251880391892381094720576584222726523876190185546875,
"chunk_inter_token_latency_avg": 0.0303251880391892381094720576584222726523876190185546875
}
],
"warmup_duration": 5.5129999999999999005240169935859739780426025390625,
"aggregate_source": "openai_continuous_usage",
"avg_running_reqs": 0,
"capacity_limited": false,
"hardware_summary": {
"samples": 14,
"gpu_count": 2,
"temp_avg_c": 53.82000000000000028421709430404007434844970703125,
"temp_max_c": 56,
"vram_total_mb": 131072,
"cpu_temp_max_c": 68.6200000000000045474735088646411895751953125,
"cpu_util_avg_pct": 25.019999999999999573674358543939888477325439453125,
"duration_seconds": 28.754000000000001335820343228988349437713623046875,
"gpu_util_avg_pct": 77.7099999999999937472239253111183643341064453125,
"gpu_util_max_pct": 88,
"mem_util_avg_pct": 10.21000000000000085265128291212022304534912109375,
"mem_util_max_pct": 12,
"pcie_rx_avg_mb_s": 401.93000000000000682121026329696178436279296875,
"pcie_rx_max_mb_s": 659,
"pcie_tx_avg_mb_s": 383.79000000000002046363078989088535308837890625,
"pcie_tx_max_mb_s": 513,
"vram_used_avg_mb": 126720,
"vram_used_max_mb": 126720,
"power_total_avg_w": 309.3500000000000227373675443232059478759765625,
"power_total_max_w": 365.67000000000001591615728102624416351318359375,
"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": 1150,
"output_seq_len_p50": 1150,
"output_seq_len_p90": 1150,
"output_seq_len_p99": 1150,
"server_spec_drafts": 0,
"server_steps_per_s": 0,
"server_utilization": 0,
"accept_norm_ref_len": 0,
"measurement_seconds": 29.927392999999998579596649506129324436187744140625,
"per_request_avg_tps": 32.67909085284836834262023330666124820709228515625,
"request_latency_avg": 0,
"request_latency_p50": 0,
"request_latency_p90": 0,
"request_latency_p99": 0,
"server_engine_steps": 0,
"client_output_tokens": 1956,
"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": 0.030325865648399952245739541467628441751003265380859375,
"inter_token_latency_p50": 0.030325865648399952245739541467628441751003265380859375,
"inter_token_latency_p90": 0.030326407735768522166974747733547701500356197357177734375,
"inter_token_latency_p99": 0.0303265297054264477971674551781688933260738849639892578125,
"output_tps_per_user_avg": 32.975151050701441590717877261340618133544921875,
"output_tps_per_user_p50": 32.975151050701441590717877261340618133544921875,
"output_tps_per_user_p90": 32.97574049514033589503014809451997280120849609375,
"output_tps_per_user_p99": 32.97587312013909155439250753261148929595947265625,
"server_spec_accept_rate": 0,
"loop_diagnostics_omitted": 0,
"measurement_wall_seconds": 30.00058299999999888996171648614108562469482421875,
"server_spec_draft_tokens": 0,
"time_to_second_token_avg": 0.0892055970034562051296234130859375,
"time_to_second_token_p50": 0.0892055970034562051296234130859375,
"time_to_second_token_p90": 0.0892422586097381953518237196476547978818416595458984375,
"time_to_second_token_p99": 0.089250507471151652172380863703438080847263336181640625,
"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": 0.030325865648399952245739541467628441751003265380859375,
"chunk_inter_token_latency_p50": 0.030325865648399952245739541467628441751003265380859375,
"chunk_inter_token_latency_p90": 0.030326407735768522166974747733547701500356197357177734375,
"chunk_inter_token_latency_p99": 0.0303265297054264477971674551781688933260738849639892578125
},
{
"ttft_avg": 1.0523720797500573098659515380859375,
"ttft_p50": 1.05272672299179248511791229248046875,
"ttft_p90": 1.053439006611006423241860829875804483890533447265625,
"ttft_p99": 1.05369458268338345163783742464147508144378662109375,
"wall_time": 35.626077958964742720127105712890625,
"num_errors": 0,
"concurrency": 4,
"underfilled": false,
"ready_reason": "metrics_unavailable, active_streams=4/4, stable=3.0s",
"total_tokens": 3036,
"aggregate_tps": 101.219270695103062962516560219228267669677734375,
"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": 1.0537229800247587263584136962890625,
"latency": 0,
"completed": false,
"input_tokens": 78,
"output_tokens": 874,
"output_tps_per_user": 25.337952056593533001205287291668355464935302734375,
"time_to_second_token": 0.106415795977227389812469482421875,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.039466488758304218986694422710570506751537322998046875,
"chunk_inter_token_latency_avg": 0.03960258009885009311545900345663540065288543701171875
},
{
"ttft": 1.05031189299188554286956787109375,
"latency": 0,
"completed": false,
"input_tokens": 78,
"output_tokens": 874,
"output_tps_per_user": 25.33600649545190464095867355354130268096923828125,
"time_to_second_token": 0.10646651196293532848358154296875,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.0394695194043114538917649269933463074266910552978515625,
"chunk_inter_token_latency_avg": 0.03960562119536080449844206441412097774446010589599609375
},
{
"ttft": 1.052776401978917419910430908203125,
"latency": 0,
"completed": false,
"input_tokens": 78,
"output_tokens": 874,
"output_tps_per_user": 25.33795318394350459811903419904410839080810546875,
"time_to_second_token": 0.10638931603170931339263916015625,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.03946648700233977569151733177932328544557094573974609375,
"chunk_inter_token_latency_avg": 0.03960257833683060446627877126957173459231853485107421875
},
{
"ttft": 1.0526770440046675503253936767578125,
"latency": 0,
"completed": false,
"input_tokens": 78,
"output_tokens": 874,
"output_tps_per_user": 25.3379512755903988363570533692836761474609375,
"time_to_second_token": 0.10646082204766571521759033203125,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.039466489974797659689986772946213022805750370025634765625,
"chunk_inter_token_latency_avg": 0.0396025813195383380449499099995591677725315093994140625
}
],
"warmup_duration": 5.51400000000000023447910280083306133747100830078125,
"aggregate_source": "openai_continuous_usage",
"avg_running_reqs": 0,
"capacity_limited": false,
"hardware_summary": {
"samples": 14,
"gpu_count": 2,
"temp_avg_c": 53.8599999999999994315658113919198513031005859375,
"temp_max_c": 55,
"vram_total_mb": 131072,
"cpu_temp_max_c": 67.8799999999999954525264911353588104248046875,
"cpu_util_avg_pct": 22.25,
"duration_seconds": 28.77499999999999857891452847979962825775146484375,
"gpu_util_avg_pct": 77.1400000000000005684341886080801486968994140625,
"gpu_util_max_pct": 88,
"mem_util_avg_pct": 8.6400000000000005684341886080801486968994140625,
"mem_util_max_pct": 10,
"pcie_rx_avg_mb_s": 545.7899999999999636202119290828704833984375,
"pcie_rx_max_mb_s": 935,
"pcie_tx_avg_mb_s": 530.1399999999999863575794734060764312744140625,
"pcie_tx_max_mb_s": 845,
"vram_used_avg_mb": 126742,
"vram_used_max_mb": 126742,
"power_total_avg_w": 299.08999999999997498889570124447345733642578125,
"power_total_max_w": 353.740000000000009094947017729282379150390625,
"vram_used_avg_pct": 96.7000000000000028421709430404007434844970703125,
"vram_used_max_pct": 96.7000000000000028421709430404007434844970703125,
"power_limit_total_w": 360
},
"loop_diagnostics": [],
"max_running_reqs": 0,
"warmup_timed_out": false,
"input_seq_len_avg": 78,
"output_seq_len_avg": 874,
"output_seq_len_p50": 874,
"output_seq_len_p90": 874,
"output_seq_len_p99": 874,
"server_spec_drafts": 0,
"server_steps_per_s": 0,
"server_utilization": 0,
"accept_norm_ref_len": 0,
"measurement_seconds": 29.994288000000000948830347624607384204864501953125,
"per_request_avg_tps": 25.30481767377576574062914005480706691741943359375,
"request_latency_avg": 0,
"request_latency_p50": 0,
"request_latency_p90": 0,
"request_latency_p99": 0,
"server_engine_steps": 0,
"client_output_tokens": 3036,
"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": 0.0394672462849382787997143395841703750193119049072265625,
"inter_token_latency_p50": 0.03946648936655093586889364587477757595479488372802734375,
"inter_token_latency_p90": 0.039468610575457314937342090388483484275639057159423828125,
"inter_token_latency_p99": 0.039469428521426042077990814505028538405895233154296875,
"output_tps_per_user_avg": 25.33746575289483615733843180350959300994873046875,
"output_tps_per_user_p50": 25.3379516660919676951380097307264804840087890625,
"output_tps_per_user_p90": 25.3379528457385134743162780068814754486083984375,
"output_tps_per_user_p99": 25.33795315012300619628149433992803096771240234375,
"server_spec_accept_rate": 0,
"loop_diagnostics_omitted": 0,
"measurement_wall_seconds": 30.0013569999999987203409546054899692535400390625,
"server_spec_draft_tokens": 0,
"time_to_second_token_avg": 0.10643311150488443672657012939453125,
"time_to_second_token_p50": 0.1064383090124465525150299072265625,
"time_to_second_token_p90": 0.1064648049883544500548993028132827021181583404541015625,
"time_to_second_token_p99": 0.1064663412654772434162708805160946212708950042724609375,
"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": 0.03960334023764496003128243728497182019054889678955078125,
"chunk_inter_token_latency_p50": 0.03960258070919421558020445672809728421270847320556640625,
"chunk_inter_token_latency_p90": 0.039604709232614061786836856526861083693802356719970703125,
"chunk_inter_token_latency_p99": 0.03960552999908613369672849557900917716324329376220703125
},
{
"ttft_avg": 0.3714770359802059829235076904296875,
"ttft_p50": 0.3714770359802059829235076904296875,
"ttft_p90": 0.3714770359802059829235076904296875,
"ttft_p99": 0.3714770359802059829235076904296875,
"wall_time": 56.6061966260313056409358978271484375,
"num_errors": 0,
"concurrency": 1,
"underfilled": false,
"ready_reason": "metrics_unavailable, active_streams=1/1, stable=3.0s",
"total_tokens": 1407,
"aggregate_tps": 46.91853713865066310972906649112701416015625,
"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": 0.3714770359802059829235076904296875,
"latency": 0,
"completed": false,
"input_tokens": 32319,
"output_tokens": 1588,
"output_tps_per_user": 47.912160770966949030480463989078998565673828125,
"time_to_second_token": 0.061618866981007158756256103515625,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.0208715278941450747829389200660443748347461223602294921875,
"chunk_inter_token_latency_avg": 0.020911057303035503196486644128526677377521991729736328125
}
],
"warmup_duration": 26.553999999999998493649400188587605953216552734375,
"aggregate_source": "openai_continuous_usage",
"avg_running_reqs": 0,
"capacity_limited": false,
"hardware_summary": {
"samples": 13,
"gpu_count": 2,
"temp_avg_c": 55.88000000000000255795384873636066913604736328125,
"temp_max_c": 57,
"vram_total_mb": 131072,
"cpu_temp_max_c": 67.25,
"cpu_util_avg_pct": 10.1300000000000007815970093361102044582366943359375,
"duration_seconds": 26.4849999999999994315658113919198513031005859375,
"gpu_util_avg_pct": 87.349999999999994315658113919198513031005859375,
"gpu_util_max_pct": 92,
"mem_util_avg_pct": 13.6500000000000003552713678800500929355621337890625,
"mem_util_max_pct": 15,
"pcie_rx_avg_mb_s": 494.6200000000000045474735088646411895751953125,
"pcie_rx_max_mb_s": 557,
"pcie_tx_avg_mb_s": 440.45999999999997953636921010911464691162109375,
"pcie_tx_max_mb_s": 495,
"vram_used_avg_mb": 128040,
"vram_used_max_mb": 128040,
"power_total_avg_w": 345.30000000000001136868377216160297393798828125,
"power_total_max_w": 367.1200000000000045474735088646411895751953125,
"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": 1588,
"output_seq_len_p50": 1588,
"output_seq_len_p90": 1588,
"output_seq_len_p99": 1588,
"server_spec_drafts": 0,
"server_steps_per_s": 0,
"server_utilization": 0,
"accept_norm_ref_len": 0,
"measurement_seconds": 29.988147000000001440866981283761560916900634765625,
"per_request_avg_tps": 46.91853713865066310972906649112701416015625,
"request_latency_avg": 0,
"request_latency_p50": 0,
"request_latency_p90": 0,
"request_latency_p99": 0,
"server_engine_steps": 0,
"client_output_tokens": 1407,
"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": 0.0208715278941450747829389200660443748347461223602294921875,
"inter_token_latency_p50": 0.0208715278941450747829389200660443748347461223602294921875,
"inter_token_latency_p90": 0.0208715278941450747829389200660443748347461223602294921875,
"inter_token_latency_p99": 0.0208715278941450747829389200660443748347461223602294921875,
"output_tps_per_user_avg": 47.912160770966949030480463989078998565673828125,
"output_tps_per_user_p50": 47.912160770966949030480463989078998565673828125,
"output_tps_per_user_p90": 47.912160770966949030480463989078998565673828125,
"output_tps_per_user_p99": 47.912160770966949030480463989078998565673828125,
"server_spec_accept_rate": 0,
"loop_diagnostics_omitted": 0,
"measurement_wall_seconds": 30.00031299999999845340425963513553142547607421875,
"server_spec_draft_tokens": 0,
"time_to_second_token_avg": 0.061618866981007158756256103515625,
"time_to_second_token_p50": 0.061618866981007158756256103515625,
"time_to_second_token_p90": 0.061618866981007158756256103515625,
"time_to_second_token_p99": 0.061618866981007158756256103515625,
"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": 0.020911057303035503196486644128526677377521991729736328125,
"chunk_inter_token_latency_p50": 0.020911057303035503196486644128526677377521991729736328125,
"chunk_inter_token_latency_p90": 0.020911057303035503196486644128526677377521991729736328125,
"chunk_inter_token_latency_p99": 0.020911057303035503196486644128526677377521991729736328125
},
{
"ttft_avg": 0.7320088319829665124416351318359375,
"ttft_p50": 0.7320088319829665124416351318359375,
"ttft_p90": 0.7320263855974189493025505726109258830547332763671875,
"ttft_p99": 0.73203033516067084196521363992360420525074005126953125,
"wall_time": 35.1140899729798547923564910888671875,
"num_errors": 0,
"concurrency": 2,
"underfilled": false,
"ready_reason": "metrics_unavailable, active_streams=2/2, stable=3.0s",
"total_tokens": 1390,
"aggregate_tps": 46.3532810475466021671309135854244232177734375,
"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": 0.7320307740010321140289306640625,
"latency": 0,
"completed": false,
"input_tokens": 32319,
"output_tokens": 809,
"output_tps_per_user": 23.9280364493352379895441117696464061737060546875,
"time_to_second_token": 0.123393650981597602367401123046875,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.041791979133656897982262989899027161300182342529296875,
"chunk_inter_token_latency_avg": 0.041791979133656897982262989899027161300182342529296875
},
{
"ttft": 0.731986889964900910854339599609375,
"latency": 0,
"completed": false,
"input_tokens": 32319,
"output_tokens": 809,
"output_tps_per_user": 23.929012066990953400136277195997536182403564453125,
"time_to_second_token": 0.123399773030541837215423583984375,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.04179027521907004161239029826901969499886035919189453125,
"chunk_inter_token_latency_avg": 0.04179027521907004161239029826901969499886035919189453125
}
],
"warmup_duration": 5.01400000000000023447910280083306133747100830078125,
"aggregate_source": "openai_continuous_usage",
"avg_running_reqs": 0,
"capacity_limited": false,
"hardware_summary": {
"samples": 14,
"gpu_count": 2,
"temp_avg_c": 52.71000000000000085265128291212022304534912109375,
"temp_max_c": 54,
"vram_total_mb": 131072,
"cpu_temp_max_c": 69,
"cpu_util_avg_pct": 19.410000000000000142108547152020037174224853515625,
"duration_seconds": 28.687000000000001165290086646564304828643798828125,
"gpu_util_avg_pct": 83.1400000000000005684341886080801486968994140625,
"gpu_util_max_pct": 91,
"mem_util_avg_pct": 8.3599999999999994315658113919198513031005859375,
"mem_util_max_pct": 9,
"pcie_rx_avg_mb_s": 314,
"pcie_rx_max_mb_s": 508,
"pcie_tx_avg_mb_s": 279.70999999999997953636921010911464691162109375,
"pcie_tx_max_mb_s": 385,
"vram_used_avg_mb": 128040,
"vram_used_max_mb": 128040,
"power_total_avg_w": 274.44999999999998863131622783839702606201171875,
"power_total_max_w": 311.1100000000000136424205265939235687255859375,
"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": 809,
"output_seq_len_p50": 809,
"output_seq_len_p90": 809,
"output_seq_len_p99": 809,
"server_spec_drafts": 0,
"server_steps_per_s": 0,
"server_utilization": 0,
"accept_norm_ref_len": 0,
"measurement_seconds": 29.987089999999998468638295889832079410552978515625,
"per_request_avg_tps": 23.17664052377330108356545679271221160888671875,
"request_latency_avg": 0,
"request_latency_p50": 0,
"request_latency_p90": 0,
"request_latency_p99": 0,
"server_engine_steps": 0,
"client_output_tokens": 1390,
"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": 0.041791127176363469797326644084023428149521350860595703125,
"inter_token_latency_p50": 0.041791127176363469797326644084023428149521350860595703125,
"inter_token_latency_p90": 0.041791808742198212345275720736026414670050144195556640625,
"inter_token_latency_p99": 0.041791962094511025949117311029112897813320159912109375,
"output_tps_per_user_avg": 23.928524258163093918483355082571506500244140625,
"output_tps_per_user_p50": 23.928524258163093918483355082571506500244140625,
"output_tps_per_user_p90": 23.928914505225382214348428533412516117095947265625,
"output_tps_per_user_p99": 23.9290023108143969921002280898392200469970703125,
"server_spec_accept_rate": 0,
"loop_diagnostics_omitted": 0,
"measurement_wall_seconds": 30.00121299999999990859578247182071208953857421875,
"server_spec_draft_tokens": 0,
"time_to_second_token_avg": 0.123396712006069719791412353515625,
"time_to_second_token_p50": 0.123396712006069719791412353515625,
"time_to_second_token_p90": 0.1233991608256474192817364610164077021181583404541015625,
"time_to_second_token_p99": 0.12339971181005239542205487168757827021181583404541015625,
"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": 0.041791127176363469797326644084023428149521350860595703125,
"chunk_inter_token_latency_p50": 0.041791127176363469797326644084023428149521350860595703125,
"chunk_inter_token_latency_p90": 0.041791808742198212345275720736026414670050144195556640625,
"chunk_inter_token_latency_p99": 0.041791962094511025949117311029112897813320159912109375
},
{
"ttft_avg": 1.26812282274477183818817138671875,
"ttft_p50": 1.1695273354998789727687835693359375,
"ttft_p90": 1.4486634040717036864265310214250348508358001708984375,
"ttft_p99": 1.556132773878052599769716834998689591884613037109375,
"wall_time": 36.14038230502046644687652587890625,
"num_errors": 0,
"concurrency": 4,
"underfilled": false,
"ready_reason": "metrics_unavailable, active_streams=4/4, stable=3.0s",
"total_tokens": 1921,
"aggregate_tps": 64.14413122957694213255308568477630615234375,
"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": 1.165362805011682212352752685546875,
"latency": 0,
"completed": false,
"input_tokens": 32319,
"output_tokens": 571,
"output_tps_per_user": 16.620031963054540113944312906824052333831787109375,
"time_to_second_token": 0.4031608839868567883968353271484375,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.060168356006952784797991995446864166297018527984619140625,
"chunk_inter_token_latency_avg": 0.060168356006952784797991995446864166297018527984619140625
},
{
"ttft": 1.169015559018589556217193603515625,
"latency": 0,
"completed": false,
"input_tokens": 32319,
"output_tokens": 571,
"output_tps_per_user": 16.62240255062214799863795633427798748016357421875,
"time_to_second_token": 0.400978200021199882030487060546875,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.060159775156123369843630399600442615337669849395751953125,
"chunk_inter_token_latency_avg": 0.060159775156123369843630399600442615337669849395751953125
},
{
"ttft": 1.17003911198116838932037353515625,
"latency": 0,
"completed": false,
"input_tokens": 32319,
"output_tokens": 571,
"output_tps_per_user": 16.62347726870728337189575540833175182342529296875,
"time_to_second_token": 0.3987725630286149680614471435546875,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.060155885789457609613695154848755919374525547027587890625,
"chunk_inter_token_latency_avg": 0.060155885789457609613695154848755919374525547027587890625
},
{
"ttft": 1.56807381496764719486236572265625,
"latency": 0,
"completed": false,
"input_tokens": 32319,
"output_tokens": 564,
"output_tps_per_user": 16.612869581893381365489403833635151386260986328125,
"time_to_second_token": 0.1875255590421147644519805908203125,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.060194296660819823774346559730474837124347686767578125,
"chunk_inter_token_latency_avg": 0.060194296660819823774346559730474837124347686767578125
}
],
"warmup_duration": 6.0160000000000000142108547152020037174224853515625,
"aggregate_source": "openai_continuous_usage",
"avg_running_reqs": 0,
"capacity_limited": false,
"hardware_summary": {
"samples": 13,
"gpu_count": 2,
"temp_avg_c": 53.46000000000000085265128291212022304534912109375,
"temp_max_c": 55,
"vram_total_mb": 131072,
"cpu_temp_max_c": 66,
"cpu_util_avg_pct": 15.199999999999999289457264239899814128875732421875,
"duration_seconds": 26.542000000000001591615728102624416351318359375,
"gpu_util_avg_pct": 84.150000000000005684341886080801486968994140625,
"gpu_util_max_pct": 90,
"mem_util_avg_pct": 11.5800000000000000710542735760100185871124267578125,
"mem_util_max_pct": 13,
"pcie_rx_avg_mb_s": 358,
"pcie_rx_max_mb_s": 625,
"pcie_tx_avg_mb_s": 331.07999999999998408384271897375583648681640625,
"pcie_tx_max_mb_s": 588,
"vram_used_avg_mb": 128054,
"vram_used_max_mb": 128054,
"power_total_avg_w": 297.58999999999997498889570124447345733642578125,
"power_total_max_w": 326.44999999999998863131622783839702606201171875,
"vram_used_avg_pct": 97.7000000000000028421709430404007434844970703125,
"vram_used_max_pct": 97.7000000000000028421709430404007434844970703125,
"power_limit_total_w": 360
},
"loop_diagnostics": [],
"max_running_reqs": 0,
"warmup_timed_out": false,
"input_seq_len_avg": 32319,
"output_seq_len_avg": 569.25,
"output_seq_len_p50": 571,
"output_seq_len_p90": 571,
"output_seq_len_p99": 571,
"server_spec_drafts": 0,
"server_steps_per_s": 0,
"server_utilization": 0,
"accept_norm_ref_len": 0,
"measurement_seconds": 29.94818000000000068894223659299314022064208984375,
"per_request_avg_tps": 16.0360328073942355331382714211940765380859375,
"request_latency_avg": 0,
"request_latency_p50": 0,
"request_latency_p90": 0,
"request_latency_p99": 0,
"server_engine_steps": 0,
"client_output_tokens": 1921,
"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": 0.060169578403338398742139503383441478945314884185791015625,
"inter_token_latency_p50": 0.060164065581538077320811197523653390817344188690185546875,
"inter_token_latency_p90": 0.06018651446465970999977201927322312258183956146240234375,
"inter_token_latency_p99": 0.06019351844120381100911032490330399014055728912353515625,
"output_tps_per_user_avg": 16.61969534106933821249185712076723575592041015625,
"output_tps_per_user_p50": 16.62121725683834227993429522030055522918701171875,
"output_tps_per_user_p90": 16.623154853281743470461151446215808391571044921875,
"output_tps_per_user_p99": 16.62344502716472760539545561186969280242919921875,
"server_spec_accept_rate": 0,
"loop_diagnostics_omitted": 0,
"measurement_wall_seconds": 30.001310000000000144382283906452357769012451171875,
"server_spec_draft_tokens": 0,
"time_to_second_token_avg": 0.347609301519696600735187530517578125,
"time_to_second_token_p50": 0.39987538152490742504596710205078125,
"time_to_second_token_p90": 0.40250607879715971648693084716796875,
"time_to_second_token_p99": 0.403095403467887081205844879150390625,
"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": 0.060169578403338398742139503383441478945314884185791015625,
"chunk_inter_token_latency_p50": 0.060164065581538077320811197523653390817344188690185546875,
"chunk_inter_token_latency_p90": 0.06018651446465970999977201927322312258183956146240234375,
"chunk_inter_token_latency_p99": 0.06019351844120381100911032490330399014055728912353515625
},
{
"ttft_avg": 0.72743101301603019237518310546875,
"ttft_p50": 0.72743101301603019237518310546875,
"ttft_p90": 0.72743101301603019237518310546875,
"ttft_p99": 0.72743101301603019237518310546875,
"wall_time": 124.209481501020491123199462890625,
"num_errors": 0,
"concurrency": 1,
"underfilled": false,
"ready_reason": "metrics_unavailable, active_streams=1/1, stable=3.0s",
"total_tokens": 1492,
"aggregate_tps": 49.82240937091719246154752909205853939056396484375,
"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": 0.72743101301603019237518310546875,
"latency": 0,
"completed": false,
"input_tokens": 128878,
"output_tokens": 1718,
"output_tps_per_user": 50.90352829142697288489216589368879795074462890625,
"time_to_second_token": 0.06584455701522529125213623046875,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.0196450036680152310963887174466435681097209453582763671875,
"chunk_inter_token_latency_avg": 0.019667913293284051967635406299450551159679889678955078125
}
],
"warmup_duration": 94.201999999999998181010596454143524169921875,
"aggregate_source": "openai_continuous_usage",
"avg_running_reqs": 0,
"capacity_limited": false,
"hardware_summary": {
"samples": 14,
"gpu_count": 2,
"temp_avg_c": 55.8599999999999994315658113919198513031005859375,
"temp_max_c": 57,
"vram_total_mb": 131072,
"cpu_temp_max_c": 66.6200000000000045474735088646411895751953125,
"cpu_util_avg_pct": 10.199999999999999289457264239899814128875732421875,
"duration_seconds": 28.724000000000000198951966012828052043914794921875,
"gpu_util_avg_pct": 88,
"gpu_util_max_pct": 93,
"mem_util_avg_pct": 13.92999999999999971578290569595992565155029296875,
"mem_util_max_pct": 15,
"pcie_rx_avg_mb_s": 481.93000000000000682121026329696178436279296875,
"pcie_rx_max_mb_s": 551,
"pcie_tx_avg_mb_s": 440.3600000000000136424205265939235687255859375,
"pcie_tx_max_mb_s": 509,
"vram_used_avg_mb": 128168,
"vram_used_max_mb": 128168,
"power_total_avg_w": 344.56999999999999317878973670303821563720703125,
"power_total_max_w": 366.220000000000027284841053187847137451171875,
"vram_used_avg_pct": 97.780000000000001136868377216160297393798828125,
"vram_used_max_pct": 97.780000000000001136868377216160297393798828125,
"power_limit_total_w": 360
},
"loop_diagnostics": [],
"max_running_reqs": 0,
"warmup_timed_out": false,
"input_seq_len_avg": 128878,
"output_seq_len_avg": 1718,
"output_seq_len_p50": 1718,
"output_seq_len_p90": 1718,
"output_seq_len_p99": 1718,
"server_spec_drafts": 0,
"server_steps_per_s": 0,
"server_utilization": 0,
"accept_norm_ref_len": 0,
"measurement_seconds": 29.946363999999999094825398060493171215057373046875,
"per_request_avg_tps": 49.82240937091719246154752909205853939056396484375,
"request_latency_avg": 0,
"request_latency_p50": 0,
"request_latency_p90": 0,
"request_latency_p99": 0,
"server_engine_steps": 0,
"client_output_tokens": 1492,
"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": 0.0196450036680152310963887174466435681097209453582763671875,
"inter_token_latency_p50": 0.0196450036680152310963887174466435681097209453582763671875,
"inter_token_latency_p90": 0.0196450036680152310963887174466435681097209453582763671875,
"inter_token_latency_p99": 0.0196450036680152310963887174466435681097209453582763671875,
"output_tps_per_user_avg": 50.90352829142697288489216589368879795074462890625,
"output_tps_per_user_p50": 50.90352829142697288489216589368879795074462890625,
"output_tps_per_user_p90": 50.90352829142697288489216589368879795074462890625,
"output_tps_per_user_p99": 50.90352829142697288489216589368879795074462890625,
"server_spec_accept_rate": 0,
"loop_diagnostics_omitted": 0,
"measurement_wall_seconds": 30.0006130000000013069438864476978778839111328125,
"server_spec_draft_tokens": 0,
"time_to_second_token_avg": 0.06584455701522529125213623046875,
"time_to_second_token_p50": 0.06584455701522529125213623046875,
"time_to_second_token_p90": 0.06584455701522529125213623046875,
"time_to_second_token_p99": 0.06584455701522529125213623046875,
"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": 0.019667913293284051967635406299450551159679889678955078125,
"chunk_inter_token_latency_p50": 0.019667913293284051967635406299450551159679889678955078125,
"chunk_inter_token_latency_p90": 0.019667913293284051967635406299450551159679889678955078125,
"chunk_inter_token_latency_p99": 0.019667913293284051967635406299450551159679889678955078125
},
{
"ttft_avg": 1.42072024897788651287555694580078125,
"ttft_p50": 1.42072024897788651287555694580078125,
"ttft_p90": 1.42173497137264348566532135009765625,
"ttft_p99": 1.4219632839114637601340973560581915080547332763671875,
"wall_time": 35.613381503033451735973358154296875,
"num_errors": 0,
"concurrency": 2,
"underfilled": false,
"ready_reason": "metrics_unavailable, active_streams=2/2, stable=3.0s",
"total_tokens": 1552,
"aggregate_tps": 51.77047493200232253229842172004282474517822265625,
"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": 1.421988651971332728862762451171875,
"latency": 0,
"completed": false,
"input_tokens": 128878,
"output_tokens": 888,
"output_tps_per_user": 26.824943967638763098193521727807819843292236328125,
"time_to_second_token": 0.133111602975986897945404052734375,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.037278735836555186577978560080737224780023097991943359375,
"chunk_inter_token_latency_avg": 0.037278735836555186577978560080737224780023097991943359375
},
{
"ttft": 1.4194518459844402968883514404296875,
"latency": 0,
"completed": false,
"input_tokens": 128878,
"output_tokens": 888,
"output_tps_per_user": 26.826229117743849883481743745505809783935546875,
"time_to_second_token": 0.1331652650260366499423980712890625,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.03727694994368639658244291013033944182097911834716796875,
"chunk_inter_token_latency_avg": 0.03727694994368639658244291013033944182097911834716796875
}
],
"warmup_duration": 5.51400000000000023447910280083306133747100830078125,
"aggregate_source": "openai_continuous_usage",
"avg_running_reqs": 0,
"capacity_limited": false,
"hardware_summary": {
"samples": 14,
"gpu_count": 2,
"temp_avg_c": 51.8599999999999994315658113919198513031005859375,
"temp_max_c": 54,
"vram_total_mb": 131072,
"cpu_temp_max_c": 69.8799999999999954525264911353588104248046875,
"cpu_util_avg_pct": 20.6700000000000017053025658242404460906982421875,
"duration_seconds": 28.745999999999998664179656771011650562286376953125,
"gpu_util_avg_pct": 81.8900000000000005684341886080801486968994140625,
"gpu_util_max_pct": 90,
"mem_util_avg_pct": 7.92999999999999971578290569595992565155029296875,
"mem_util_max_pct": 9,
"pcie_rx_avg_mb_s": 323.6399999999999863575794734060764312744140625,
"pcie_rx_max_mb_s": 538,
"pcie_tx_avg_mb_s": 277.56999999999999317878973670303821563720703125,
"pcie_tx_max_mb_s": 387,
"vram_used_avg_mb": 128168,
"vram_used_max_mb": 128168,
"power_total_avg_w": 247.1299999999999954525264911353588104248046875,
"power_total_max_w": 309.69999999999998863131622783839702606201171875,
"vram_used_avg_pct": 97.780000000000001136868377216160297393798828125,
"vram_used_max_pct": 97.780000000000001136868377216160297393798828125,
"power_limit_total_w": 360
},
"loop_diagnostics": [],
"max_running_reqs": 0,
"warmup_timed_out": false,
"input_seq_len_avg": 128878,
"output_seq_len_avg": 888,
"output_seq_len_p50": 888,
"output_seq_len_p90": 888,
"output_seq_len_p99": 888,
"server_spec_drafts": 0,
"server_steps_per_s": 0,
"server_utilization": 0,
"accept_norm_ref_len": 0,
"measurement_seconds": 29.978477000000001595481080585159361362457275390625,
"per_request_avg_tps": 25.885237466001161266149210860021412372589111328125,
"request_latency_avg": 0,
"request_latency_p50": 0,
"request_latency_p90": 0,
"request_latency_p99": 0,
"server_engine_steps": 0,
"client_output_tokens": 1552,
"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": 0.0372778428901207881107637831519241444766521453857421875,
"inter_token_latency_p50": 0.0372778428901207881107637831519241444766521453857421875,
"inter_token_latency_p90": 0.037278557247268306884535604694974608719348907470703125,
"inter_token_latency_p99": 0.03727871797762649652696609336999244987964630126953125,
"output_tps_per_user_avg": 26.82558654269130471448079333640635013580322265625,
"output_tps_per_user_p50": 26.82558654269130471448079333640635013580322265625,
"output_tps_per_user_p90": 26.82610060273334084968155366368591785430908203125,
"output_tps_per_user_p99": 26.826216266242798980101724737323820590972900390625,
"server_spec_accept_rate": 0,
"loop_diagnostics_omitted": 0,
"measurement_wall_seconds": 30.000654000000000820591594674624502658843994140625,
"server_spec_draft_tokens": 0,
"time_to_second_token_avg": 0.13313843400101177394390106201171875,
"time_to_second_token_p50": 0.13313843400101177394390106201171875,
"time_to_second_token_p90": 0.133159898821031685844928915685159154236316680908203125,
"time_to_second_token_p99": 0.1331647284055361424304209094771067611873149871826171875,
"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": 0.0372778428901207881107637831519241444766521453857421875,
"chunk_inter_token_latency_p50": 0.0372778428901207881107637831519241444766521453857421875,
"chunk_inter_token_latency_p90": 0.037278557247268306884535604694974608719348907470703125,
"chunk_inter_token_latency_p99": 0.03727871797762649652696609336999244987964630126953125
},
{
"ttft_avg": 3.067665966736967675387859344482421875,
"ttft_p50": 2.94989748249645344913005828857421875,
"ttft_p90": 3.28095987036358582855655185994692146778106689453125,
"ttft_p99": 3.40841787719284194935198684106580913066864013671875,
"wall_time": 37.5983978839940391480922698974609375,
"num_errors": 0,
"concurrency": 4,
"underfilled": false,
"ready_reason": "metrics_unavailable, active_streams=4/4, stable=3.0s",
"total_tokens": 2031,
"aggregate_tps": 67.970195838417197364833555184304714202880859375,
"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": 2.9482890240033157169818878173828125,
"latency": 0,
"completed": false,
"input_tokens": 128878,
"output_tokens": 577,
"output_tps_per_user": 17.220026598499433845290695899166166782379150390625,
"time_to_second_token": 0.4823734969832003116607666015625,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.058071919591990685260984861315591842867434024810791015625,
"chunk_inter_token_latency_avg": 0.058172914234759363905080675749559304676949977874755859375
},
{
"ttft": 2.949281779001466929912567138671875,
"latency": 0,
"completed": false,
"input_tokens": 128878,
"output_tokens": 577,
"output_tps_per_user": 17.22268381694050987107402761466801166534423828125,
"time_to_second_token": 0.4800375929917208850383758544921875,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.058062959909673535208174399713243474252521991729736328125,
"chunk_inter_token_latency_avg": 0.058163938970386012961011346078521455638110637664794921875
},
{
"ttft": 2.9505131859914399683475494384765625,
"latency": 0,
"completed": false,
"input_tokens": 128878,
"output_tokens": 577,
"output_tps_per_user": 17.225073221679100043957078014500439167022705078125,
"time_to_second_token": 0.4751977749983780086040496826171875,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.0580549056094241677783429622650146484375,
"chunk_inter_token_latency_avg": 0.058155870662657947745000086570144048891961574554443359375
},
{
"ttft": 3.4225798779516480863094329833984375,
"latency": 0,
"completed": false,
"input_tokens": 128878,
"output_tokens": 571,
"output_tps_per_user": 17.2919730028701366109089576639235019683837890625,
"time_to_second_token": 0.1872160620405338704586029052734375,
"e2e_output_tps_per_user": 0,
"inter_token_latency_avg": 0.0578303007895061560450500337537960149347782135009765625,
"chunk_inter_token_latency_avg": 0.057931935764531650168152054902748204767704010009765625
}
],
"warmup_duration": 7.52099999999999990762944435118697583675384521484375,
"aggregate_source": "openai_continuous_usage",
"avg_running_reqs": 0,
"capacity_limited": false,
"hardware_summary": {
"samples": 14,
"gpu_count": 2,
"temp_avg_c": 51.8599999999999994315658113919198513031005859375,
"temp_max_c": 54,
"vram_total_mb": 131072,
"cpu_temp_max_c": 66.1200000000000045474735088646411895751953125,
"cpu_util_avg_pct": 15.550000000000000710542735760100185871124267578125,
"duration_seconds": 28.833999999999999630517777404747903347015380859375,
"gpu_util_avg_pct": 83.8599999999999994315658113919198513031005859375,
"gpu_util_max_pct": 89,
"mem_util_avg_pct": 11.25,
"mem_util_max_pct": 12,
"pcie_rx_avg_mb_s": 370.8600000000000136424205265939235687255859375,
"pcie_rx_max_mb_s": 666,
"pcie_tx_avg_mb_s": 356.5,
"pcie_tx_max_mb_s": 560,
"vram_used_avg_mb": 128168,
"vram_used_max_mb": 128168,
"power_total_avg_w": 298.009999999999990905052982270717620849609375,
"power_total_max_w": 348.23000000000001818989403545856475830078125,
"vram_used_avg_pct": 97.780000000000001136868377216160297393798828125,
"vram_used_max_pct": 97.780000000000001136868377216160297393798828125,
"power_limit_total_w": 360
},
"loop_diagnostics": [],
"max_running_reqs": 0,
"warmup_timed_out": false,
"input_seq_len_avg": 128878,
"output_seq_len_avg": 575.5,
"output_seq_len_p50": 577,
"output_seq_len_p90": 577,
"output_seq_len_p99": 577,
"server_spec_drafts": 0,
"server_steps_per_s": 0,
"server_utilization": 0,
"accept_norm_ref_len": 0,
"measurement_seconds": 29.88074399999999997135091689415276050567626953125,
"per_request_avg_tps": 16.99254895960429934120838879607617855072021484375,
"request_latency_avg": 0,
"request_latency_p50": 0,
"request_latency_p90": 0,
"request_latency_p99": 0,
"server_engine_steps": 0,
"client_output_tokens": 2031,
"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": 0.0580050214751486326036911123082973062992095947265625,
"inter_token_latency_p50": 0.0580589327595488480238117290355148725211620330810546875,
"inter_token_latency_p90": 0.05806923168729553885736294205344165675342082977294921875,
"inter_token_latency_p99": 0.05807165080152117131451205978009966202080249786376953125,
"output_tps_per_user_avg": 17.239939159997295092807689798064529895782470703125,
"output_tps_per_user_p50": 17.223878519309806733872392214834690093994140625,
"output_tps_per_user_p90": 17.2719030685128274171802331693470478057861328125,
"output_tps_per_user_p99": 17.289966009434404980993349454365670680999755859375,
"server_spec_accept_rate": 0,
"loop_diagnostics_omitted": 0,
"measurement_wall_seconds": 30.001022999999999996134647517465054988861083984375,
"server_spec_draft_tokens": 0,
"time_to_second_token_avg": 0.406206231753458268940448760986328125,
"time_to_second_token_p50": 0.4776176839950494468212127685546875,
"time_to_second_token_p90": 0.48167272578575648367404937744140625,
"time_to_second_token_p99": 0.482303419863455928862094879150390625,
"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": 0.058106164908083741960087564848436159081757068634033203125,
"chunk_inter_token_latency_p50": 0.058159904816521980353005716324332752265036106109619140625,
"chunk_inter_token_latency_p90": 0.0581702216554473572340810960668022744357585906982421875,
"chunk_inter_token_latency_p99": 0.058172644976828162544091327390560763888061046600341796875
}
],
"metadata": {
"model": "GLM-5.3-Flash-max",
"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": "# GLM-5.3-Flash, max 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"
},
"recipe_slug": "glm-5-3-flash-max-effort-on-2x-cmp170hx-exl3-tp2-four-slots",
"source_runs": [
{
"gates": {
"observed_efforts": [
"max"
],
"cpu_psi_avg10_max": 0.92000000000000003996802888650563545525074005126953125,
"public_router_posts": 0,
"peak_active_sequences": 4
},
"label": "max-ctx0",
"sha256": "cb914159280b0e76559042f9c4ad01fe994b85271b142978ca8f8f3d2888d108",
"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-max",
"--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": "2026-10-17T12:35:51.871540+00:00",
"finished": "2026-10-17T12:38:01.341463+00:00",
"raw_file": "runs/max-ctx0-attempt1/benchmark_results.json"
},
{
"gates": {
"observed_efforts": [
"max"
],
"cpu_psi_avg10_max": 0.179999999999999993338661852249060757458209991455078125,
"public_router_posts": 0,
"peak_active_sequences": 4
},
"label": "max-ctx32768",
"sha256": "c7bb465903a9af4f9d5203fcbe36e82ac412183d3991aec6977a6fcaf3a6ef68",
"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-max",
"--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": "2026-10-17T12:44:07.579597+00:00",
"finished": "2026-10-17T12:47:07.149833+00:00",
"raw_file": "runs/max-ctx32768-attempt1/benchmark_results.json"
},
{
"gates": {
"observed_efforts": [
"max"
],
"cpu_psi_avg10_max": 0.939999999999999946709294817992486059665679931640625,
"public_router_posts": 0,
"peak_active_sequences": 4
},
"label": "max-ctx131072",
"sha256": "035bb2d9b690b09d9e1cb88937f19d65c53de188104d76a088121712c4020f21",
"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-max",
"--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": "2026-10-17T12:57:50.049761+00:00",
"finished": "2026-10-17T13:03:06.229023+00:00",
"raw_file": "runs/max-ctx131072-attempt1/benchmark_results.json"
},
{
"gates": {
"observed_efforts": [
"max"
],
"cpu_psi_avg10_max": 0.56999999999999995115018691649311222136020660400390625,
"public_router_posts": 0,
"peak_active_sequences": 4
},
"label": "max-finite-ctx0",
"sha256": "4db0a4deede5241a8c16539b0f5520405e6985212524f5a6f66f0c33a50e81ca",
"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-max",
"--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": "2026-10-17T13:17:07.872745+00:00",
"finished": "2026-10-17T13:19:34.082951+00:00",
"raw_file": "runs/max-finite-ctx0-attempt1/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": [
1242
],
"tokens_per_second": [
41.4733142771386127378718811087310314178466796875
],
"requested_duration": 30,
"measurement_seconds": 29.94696762599050998687744140625,
"request_count_target": 0,
"aggregate_tokens_per_second": 41.4733142771386127378718811087310314178466796875,
"measurement_start_monotonic": 508836.6206766120158135890960693359375,
"reported_measurement_seconds_rounded": 29.946967999999998255589161999523639678955078125
},
{
"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": [
978,
978
],
"tokens_per_second": [
32.67909085284836834262023330666124820709228515625,
32.67909085284836834262023330666124820709228515625
],
"requested_duration": 30,
"measurement_seconds": 29.927393158024646169224070035852491855621337890625,
"request_count_target": 0,
"aggregate_tokens_per_second": 65.3581817056967366852404666133224964141845703125,
"measurement_start_monotonic": 508874.1486469259834848344326019287109375,
"reported_measurement_seconds_rounded": 29.927392999999998579596649506129324436187744140625
},
{
"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": [
759,
759,
759,
759
],
"tokens_per_second": [
25.30481767377576574062914005480706691741943359375,
25.30481767377576574062914005480706691741943359375,
25.30481767377576574062914005480706691741943359375,
25.30481767377576574062914005480706691741943359375
],
"requested_duration": 30,
"measurement_seconds": 29.9942884309566579759120941162109375,
"request_count_target": 0,
"aggregate_tokens_per_second": 101.219270695103062962516560219228267669677734375,
"measurement_start_monotonic": 508911.6807014960213564336299896240234375,
"reported_measurement_seconds_rounded": 29.994288000000000948830347624607384204864501953125
},
{
"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": [
1407
],
"tokens_per_second": [
46.91853713865066310972906649112701416015625
],
"requested_duration": 30,
"measurement_seconds": 29.988147240015678107738494873046875,
"request_count_target": 0,
"aggregate_tokens_per_second": 46.91853713865066310972906649112701416015625,
"measurement_start_monotonic": 509382.44761947100050747394561767578125,
"reported_measurement_seconds_rounded": 29.988147000000001440866981283761560916900634765625
},
{
"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": [
695,
695
],
"tokens_per_second": [
23.17664052377330108356545679271221160888671875,
23.17664052377330108356545679271221160888671875
],
"requested_duration": 30,
"measurement_seconds": 29.98708977201022207736968994140625,
"request_count_target": 0,
"aggregate_tokens_per_second": 46.3532810475466021671309135854244232177734375,
"measurement_start_monotonic": 509419.5234910459839738905429840087890625,
"reported_measurement_seconds_rounded": 29.987089999999998468638295889832079410552978515625
},
{
"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": [
481,
481,
481,
478
],
"tokens_per_second": [
16.061076065292301251474782475270330905914306640625,
16.061076065292301251474782475270330905914306640625,
16.061076065292301251474782475270330905914306640625,
15.96090303370004193084241705946624279022216796875
],
"requested_duration": 30,
"measurement_seconds": 29.948180186969690197429372346960008144378662109375,
"request_count_target": 0,
"aggregate_tokens_per_second": 64.14413122957694213255308568477630615234375,
"measurement_start_monotonic": 509457.649736259016208350658416748046875,
"reported_measurement_seconds_rounded": 29.94818000000000068894223659299314022064208984375
},
{
"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": [
1492
],
"tokens_per_second": [
49.82240937091719246154752909205853939056396484375
],
"requested_duration": 30,
"measurement_seconds": 29.9463638719753362238407135009765625,
"request_count_target": 0,
"aggregate_tokens_per_second": 49.82240937091719246154752909205853939056396484375,
"measurement_start_monotonic": 510339.53973444900475442409515380859375,
"reported_measurement_seconds_rounded": 29.946363999999999094825398060493171215057373046875
},
{
"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": [
776,
776
],
"tokens_per_second": [
25.885237466001161266149210860021412372589111328125,
25.885237466001161266149210860021412372589111328125
],
"requested_duration": 30,
"measurement_seconds": 29.9784771539852954447269439697265625,
"request_count_target": 0,
"aggregate_tokens_per_second": 51.77047493200232253229842172004282474517822265625,
"measurement_start_monotonic": 510377.074038758990354835987091064453125,
"reported_measurement_seconds_rounded": 29.978477000000001595481080585159361362457275390625
},
{
"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": [
509,
509,
509,
504
],
"tokens_per_second": [
17.034381921100123946644089301116764545440673828125,
17.034381921100123946644089301116764545440673828125,
17.034381921100123946644089301116764545440673828125,
16.8670500751168219721876084804534912109375
],
"requested_duration": 30,
"measurement_seconds": 29.88074368401430547237396240234375,
"request_count_target": 0,
"aggregate_tokens_per_second": 67.970195838417197364833555184304714202880859375,
"measurement_start_monotonic": 510416.70560424798168241977691650390625,
"reported_measurement_seconds_rounded": 29.88074399999999997135091689415276050567626953125
}
],
"slot_ceiling": 4,
"tool_version": "0.8.0",
"burst_results": [
{
"ttft_avg": 0.291426673749811016023159027099609375,
"ttft_p50": 0.2913828105083666741847991943359375,
"ttft_p90": 0.29326906150672582551663936101249419152736663818359375,
"ttft_p99": 0.29360551254590971925750864102155901491641998291015625,
"wall_time": 39.7239213399589061737060546875,
"num_errors": 0,
"concurrency": 1,
"underfilled": false,
"ready_reason": "request_count=4, warmup_request_count=1",
"total_tokens": 2048,
"aggregate_tps": 51.5558366575427982070323196239769458770751953125,
"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": 0.29364289599470794200897216796875,
"latency": 9.87666594400070607662200927734375,
"completed": true,
"input_tokens": 78,
"output_tokens": 512,
"output_tps_per_user": 53.32346561624174086091443314217031002044677734375,
"time_to_second_token": 0.0586799230077303946018218994140625,
"e2e_output_tps_per_user": 51.83935580113444530070410110056400299072265625,
"inter_token_latency_avg": 0.018753469761264185411420157834072597324848175048828125,
"chunk_inter_token_latency_avg": 0.018753469761264185411420157834072597324848175048828125
},
{
"ttft": 0.290368839981965720653533935546875,
"latency": 9.88814019598066806793212890625,
"completed": true,
"input_tokens": 78,
"output_tokens": 512,
"output_tps_per_user": 53.24152670928339148304075933992862701416015625,
"time_to_second_token": 0.0588920160080306231975555419921875,
"e2e_output_tps_per_user": 51.77920112905739102870938950218260288238525390625,
"inter_token_latency_avg": 0.018782331420741101013494045446350355632603168487548828125,
"chunk_inter_token_latency_avg": 0.018782331420741101013494045446350355632603168487548828125
},
{
"ttft": 0.2892981779878027737140655517578125,
"latency": 9.8989601450157351791858673095703125,
"completed": true,
"input_tokens": 78,
"output_tokens": 512,
"output_tps_per_user": 53.1756477754692156167948269285261631011962890625,
"time_to_second_token": 0.058949825004674494266510009765625,
"e2e_output_tps_per_user": 51.722604445255711880236049182713031768798828125,
"inter_token_latency_avg": 0.0188056007182542707367733925138963968493044376373291015625,
"chunk_inter_token_latency_avg": 0.0188056007182542707367733925138963968493044376373291015625
},
{
"ttft": 0.292396781034767627716064453125,
"latency": 9.8831332009867765009403228759765625,
"completed": true,
"input_tokens": 78,
"output_tokens": 512,
"output_tps_per_user": 53.28058009569999597943024127744138240814208984375,
"time_to_second_token": 0.0577116939821280539035797119140625,
"e2e_output_tps_per_user": 51.80543351868207224697471247054636478424072265625,
"inter_token_latency_avg": 0.0187685644226066698780908126309441286139190196990966796875,
"chunk_inter_token_latency_avg": 0.0187685644226066698780908126309441286139190196990966796875
}
],
"warmup_duration": 0,
"aggregate_source": "openai_completed_usage",
"avg_running_reqs": 0,
"capacity_limited": false,
"hardware_summary": {
"samples": 18,
"gpu_count": 2,
"temp_avg_c": 54.8599999999999994315658113919198513031005859375,
"temp_max_c": 57,
"vram_total_mb": 131072,
"cpu_temp_max_c": 68.5,
"cpu_util_avg_pct": 10.230000000000000426325641456060111522674560546875,
"duration_seconds": 37.566000000000002501110429875552654266357421875,
"gpu_util_avg_pct": 85.6400000000000005684341886080801486968994140625,
"gpu_util_max_pct": 92,
"mem_util_avg_pct": 13.75,
"mem_util_max_pct": 15,
"pcie_rx_avg_mb_s": 542.1699999999999590727384202182292938232421875,
"pcie_rx_max_mb_s": 671,
"pcie_tx_avg_mb_s": 446.5,
"pcie_tx_max_mb_s": 527,
"vram_used_avg_mb": 128774,
"vram_used_max_mb": 128774,
"power_total_avg_w": 324.6200000000000045474735088646411895751953125,
"power_total_max_w": 369.1499999999999772626324556767940521240234375,
"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": 39.72392099999999715009835199452936649322509765625,
"per_request_avg_tps": 51.5558366575427982070323196239769458770751953125,
"request_latency_avg": 9.88672487149597145617008209228515625,
"request_latency_p50": 9.88563669848372228443622589111328125,
"request_latency_p90": 9.895714160305214335267010028474032878875732421875,
"request_latency_p99": 9.8986355465446838053367173415608704090118408203125,
"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": 0.018777491580716558494668078083122964017093181610107421875,
"inter_token_latency_p50": 0.018775447921673883711068953061840147711336612701416015625,
"inter_token_latency_p90": 0.0187986199290003201667342835889940033666789531707763671875,
"inter_token_latency_p99": 0.0188049026393288763736588720121289952658116817474365234375,
"output_tps_per_user_avg": 53.255305049173585985045065172016620635986328125,
"output_tps_per_user_p50": 53.2610534024916972839491791091859340667724609375,
"output_tps_per_user_p90": 53.3105999600792159753837040625512599945068359375,
"output_tps_per_user_p99": 53.32217905062548624073315295390784740447998046875,
"server_spec_accept_rate": 0,
"loop_diagnostics_omitted": 0,
"measurement_wall_seconds": 0,
"server_spec_draft_tokens": 0,
"time_to_second_token_avg": 0.058558364500640891492366790771484375,
"time_to_second_token_p50": 0.058785969507880508899688720703125,
"time_to_second_token_p90": 0.05893248230568133572138123099648510105907917022705078125,
"time_to_second_token_p99": 0.05894809073477517979977591267015668563544750213623046875,
"server_spec_accept_length": 0,
"e2e_output_tps_per_user_avg": 51.786648723532408666869741864502429962158203125,
"e2e_output_tps_per_user_p50": 51.79231732386973163784205098636448383331298828125,
"e2e_output_tps_per_user_p90": 51.82917911639873409512802027165889739990234375,
"e2e_output_tps_per_user_p99": 51.83833813266087275906102149747312068939208984375,
"server_accept_len_effective": 0,
"server_spec_accepted_tokens": 0,
"chunk_inter_token_latency_avg": 0.018777491580716558494668078083122964017093181610107421875,
"chunk_inter_token_latency_p50": 0.018775447921673883711068953061840147711336612701416015625,
"chunk_inter_token_latency_p90": 0.0187986199290003201667342835889940033666789531707763671875,
"chunk_inter_token_latency_p99": 0.0188049026393288763736588720121289952658116817474365234375
},
{
"ttft_avg": 0.49450309851090423762798309326171875,
"ttft_p50": 0.53021741300472058355808258056640625,
"ttft_p90": 0.59000928134191787943763074508751742541790008544921875,
"ttft_p99": 0.61287002717028371545637810413609258830547332763671875,
"wall_time": 32.4523503180244006216526031494140625,
"num_errors": 0,
"concurrency": 2,
"underfilled": false,
"ready_reason": "request_count=4, warmup_request_count=1",
"total_tokens": 2048,
"aggregate_tps": 63.1079099026771501712573808617889881134033203125,
"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": 0.529694144963286817073822021484375,
"latency": 14.5874083319795317947864532470703125,
"completed": true,
"input_tokens": 78,
"output_tokens": 512,
"output_tps_per_user": 36.3501486231638892832052079029381275177001953125,
"time_to_second_token": 0.0791451320401392877101898193359375,
"e2e_output_tps_per_user": 35.0987638343925709705217741429805755615234375,
"inter_token_latency_avg": 0.027510203888485802725671902635440346784889698028564453125,
"chunk_inter_token_latency_avg": 0.027510203888485802725671902635440346784889698028564453125
},
{
"ttft": 0.3021674579940736293792724609375,
"latency": 17.7700171690084971487522125244140625,
"completed": true,
"input_tokens": 78,
"output_tokens": 512,
"output_tps_per_user": 29.25374378952819398591600474901497364044189453125,
"time_to_second_token": 0.3127748449915088713169097900390625,
"e2e_output_tps_per_user": 28.8125776767928556409970042295753955841064453125,
"inter_token_latency_avg": 0.03418365892566423147602705512326792813837528228759765625,
"chunk_inter_token_latency_avg": 0.03418365892566423147602705512326792813837528228759765625
},
{
"ttft": 0.5307406810461543500423431396484375,
"latency": 14.5873405340244062244892120361328125,
"completed": true,
"input_tokens": 78,
"output_tokens": 512,
"output_tps_per_user": 36.35303027365693395722701097838580608367919921875,
"time_to_second_token": 0.0795378879993222653865814208984375,
"e2e_output_tps_per_user": 35.09892696381357524160193861462175846099853515625,
"inter_token_latency_avg": 0.0275080231956521571723417451948989764787256717681884765625,
"chunk_inter_token_latency_avg": 0.0275080231956521571723417451948989764787256717681884765625
},
{
"ttft": 0.6154101100401021540164947509765625,
"latency": 16.75653035403229296207427978515625,
"completed": true,
"input_tokens": 78,
"output_tokens": 512,
"output_tps_per_user": 31.658273544563730439449500408954918384552001953125,
"time_to_second_token": 0.08915618597529828548431396484375,
"e2e_output_tps_per_user": 30.55525154566334578021269408054649829864501953125,
"inter_token_latency_avg": 0.03158731945986729960917926973706926219165325164794921875,
"chunk_inter_token_latency_avg": 0.03177385874801612331008726641812245361506938934326171875
}
],
"warmup_duration": 0,
"aggregate_source": "openai_completed_usage",
"avg_running_reqs": 0,
"capacity_limited": false,
"hardware_summary": {
"samples": 14,
"gpu_count": 2,
"temp_avg_c": 54.5,
"temp_max_c": 57,
"vram_total_mb": 131072,
"cpu_temp_max_c": 67,
"cpu_util_avg_pct": 23.510000000000001563194018672220408916473388671875,
"duration_seconds": 28.760999999999999232613845379091799259185791015625,
"gpu_util_avg_pct": 78.8900000000000005684341886080801486968994140625,
"gpu_util_max_pct": 89,
"mem_util_avg_pct": 11.28999999999999914734871708787977695465087890625,
"mem_util_max_pct": 14,
"pcie_rx_avg_mb_s": 406.3600000000000136424205265939235687255859375,
"pcie_rx_max_mb_s": 624,
"pcie_tx_avg_mb_s": 353.29000000000002046363078989088535308837890625,
"pcie_tx_max_mb_s": 596,
"vram_used_avg_mb": 128786,
"vram_used_max_mb": 128786,
"power_total_avg_w": 314.740000000000009094947017729282379150390625,
"power_total_max_w": 361.3799999999999954525264911353588104248046875,
"vram_used_avg_pct": 98.2600000000000051159076974727213382720947265625,
"vram_used_max_pct": 98.2600000000000051159076974727213382720947265625,
"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": 32.45235000000000269437805400229990482330322265625,
"per_request_avg_tps": 31.55395495133857508562869043089449405670166015625,
"request_latency_avg": 15.925324097261182032525539398193359375,
"request_latency_p50": 15.67196934300591237843036651611328125,
"request_latency_p90": 17.465971124515636603291568462736904621124267578125,
"request_latency_p99": 17.7396125645592093178493087179958820343017578125,
"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": 0.0301973013674173736131667311610726756043732166290283203125,
"inter_token_latency_p50": 0.0295487616741765511674255861862548044882714748382568359375,
"inter_token_latency_p90": 0.03340475708592514914041515794451697729527950286865234375,
"inter_token_latency_p99": 0.034105768741690321854687084623947157524526119232177734375,
"output_tps_per_user_avg": 33.40379905772818602827101130969822406768798828125,
"output_tps_per_user_p50": 34.0042110838638080849705147556960582733154296875,
"output_tps_per_user_p90": 36.3521657785090184233922627754509449005126953125,
"output_tps_per_user_p99": 36.3529438241421445354717434383928775787353515625,
"server_spec_accept_rate": 0,
"loop_diagnostics_omitted": 0,
"measurement_wall_seconds": 0,
"server_spec_draft_tokens": 0,
"time_to_second_token_avg": 0.140153512751567177474498748779296875,
"time_to_second_token_p50": 0.08434703698731027543544769287109375,
"time_to_second_token_p90": 0.2456892472866457344249369043609476648271083831787109375,
"time_to_second_token_p99": 0.3060662852210225270965793242794461548328399658203125,
"server_spec_accept_length": 0,
"e2e_output_tps_per_user_avg": 32.391380005165586908333352766931056976318359375,
"e2e_output_tps_per_user_p50": 32.827007690027954822653555311262607574462890625,
"e2e_output_tps_per_user_p90": 35.098878024987271828649681992828845977783203125,
"e2e_output_tps_per_user_p99": 35.09892206993094276867850567214190959930419921875,
"server_accept_len_effective": 0,
"server_spec_accepted_tokens": 0,
"chunk_inter_token_latency_avg": 0.0302439361894545795383937303313359734602272510528564453125,
"chunk_inter_token_latency_p50": 0.0296420313182509630178795845267814001999795436859130859375,
"chunk_inter_token_latency_p90": 0.03346071887236980180180268007461563684046268463134765625,
"chunk_inter_token_latency_p99": 0.034111364920334785733047056055511347949504852294921875
},
{
"ttft_avg": 1.0208510140073485672473907470703125,
"ttft_p50": 1.02123252299497835338115692138671875,
"ttft_p90": 1.021816280722851022488839589641429483890533447265625,
"ttft_p99": 1.0219671246007784315423805310274474322795867919921875,
"wall_time": 21.363349313964135944843292236328125,
"num_errors": 0,
"concurrency": 4,
"underfilled": false,
"ready_reason": "request_count=4, warmup_request_count=1",
"total_tokens": 2048,
"aggregate_tps": 95.8651178662011744791016099043190479278564453125,
"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": 1.0219838850316591560840606689453125,
"latency": 21.2249176180339418351650238037109375,
"completed": true,
"input_tokens": 78,
"output_tokens": 512,
"output_tps_per_user": 25.293356239903982185524000669829547405242919921875,
"time_to_second_token": 0.1240382529795169830322265625,
"e2e_output_tps_per_user": 24.12259068393154137766032363288104534149169921875,
"inter_token_latency_avg": 0.039536073841491749247811782197459251619875431060791015625,
"chunk_inter_token_latency_avg": 0.03976955459252418112470195410423912107944488525390625
},
{
"ttft": 1.0189551250077784061431884765625,
"latency": 21.2243325469898991286754608154296875,
"completed": true,
"input_tokens": 78,
"output_tokens": 512,
"output_tps_per_user": 25.2902971980154944731111754663288593292236328125,
"time_to_second_token": 0.124111031997017562389373779296875,
"e2e_output_tps_per_user": 24.1232556485086462316758115775883197784423828125,
"inter_token_latency_avg": 0.039540856011706697337171334538652445189654827117919921875,
"chunk_inter_token_latency_avg": 0.03977436500390181539632550311580416746437549591064453125
},
{
"ttft": 1.021425204002298414707183837890625,
"latency": 21.224562883959151804447174072265625,
"completed": true,
"input_tokens": 78,
"output_tokens": 512,
"output_tps_per_user": 25.293100908130387693972807028330862522125244140625,
"time_to_second_token": 0.1240409580059349536895751953125,
"e2e_output_tps_per_user": 24.122993853831179222879654844291508197784423828125,
"inter_token_latency_avg": 0.0395364729549057825064295457195839844644069671630859375,
"chunk_inter_token_latency_avg": 0.03976995606290718854314292229901184327900409698486328125
},
{
"ttft": 1.0210398419876582920551300048828125,
"latency": 21.224061010987497866153717041015625,
"completed": true,
"input_tokens": 78,
"output_tokens": 512,
"output_tps_per_user": 25.293246773611002709003514610230922698974609375,
"time_to_second_token": 0.1240398859954439103603363037109375,
"e2e_output_tps_per_user": 24.12356427617421417153309448622167110443115234375,
"inter_token_latency_avg": 0.039536244949119059288733524226699955761432647705078125,
"chunk_inter_token_latency_avg": 0.03976972671062960207333247808492160402238368988037109375
}
],
"warmup_duration": 0,
"aggregate_source": "openai_completed_usage",
"avg_running_reqs": 0,
"capacity_limited": false,
"hardware_summary": {
"samples": 10,
"gpu_count": 2,
"temp_avg_c": 54.0499999999999971578290569595992565155029296875,
"temp_max_c": 56,
"vram_total_mb": 131072,
"cpu_temp_max_c": 63.75,
"cpu_util_avg_pct": 19.660000000000000142108547152020037174224853515625,
"duration_seconds": 19.900999999999999801048033987171947956085205078125,
"gpu_util_avg_pct": 79.2000000000000028421709430404007434844970703125,
"gpu_util_max_pct": 94,
"mem_util_avg_pct": 9.25,
"mem_util_max_pct": 13,
"pcie_rx_avg_mb_s": 586.799999999999954525264911353588104248046875,
"pcie_rx_max_mb_s": 925,
"pcie_tx_avg_mb_s": 554.299999999999954525264911353588104248046875,
"pcie_tx_max_mb_s": 786,
"vram_used_avg_mb": 128786,
"vram_used_max_mb": 128786,
"power_total_avg_w": 289.68000000000000682121026329696178436279296875,
"power_total_max_w": 326.68000000000000682121026329696178436279296875,
"vram_used_avg_pct": 98.2600000000000051159076974727213382720947265625,
"vram_used_max_pct": 98.2600000000000051159076974727213382720947265625,
"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": 21.363348999999999477950041182339191436767578125,
"per_request_avg_tps": 23.966279466550293619775402476079761981964111328125,
"request_latency_avg": 21.22446851499262265861034393310546875,
"request_latency_p50": 21.22444771547452546656131744384765625,
"request_latency_p90": 21.22481119781150482594966888427734375,
"request_latency_p99": 21.224906976011698844786224071867763996124267578125,
"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": 0.03953741193930582209503654667059890925884246826171875,
"inter_token_latency_p50": 0.03953635895201242089758153497314197011291980743408203125,
"inter_token_latency_p90": 0.0395395410946664249696169690651004202663898468017578125,
"inter_token_latency_p99": 0.039540724520002672182084069163465755991637706756591796875,
"output_tps_per_user_avg": 25.2925002799152167654028744436800479888916015625,
"output_tps_per_user_p50": 25.2931738408706934251313214190304279327392578125,
"output_tps_per_user_p90": 25.293323400016088697839222732000052928924560546875,
"output_tps_per_user_p99": 25.293352955915192836755522876046597957611083984375,
"server_spec_accept_rate": 0,
"loop_diagnostics_omitted": 0,
"measurement_wall_seconds": 0,
"server_spec_draft_tokens": 0,
"time_to_second_token_avg": 0.124057532244478352367877960205078125,
"time_to_second_token_p50": 0.12404042200068943202495574951171875,
"time_to_second_token_p90": 0.1240900097996927853305493272273452021181583404541015625,
"time_to_second_token_p99": 0.12410892977728509023460645721570472232997417449951171875,
"server_spec_accept_length": 0,
"e2e_output_tps_per_user_avg": 24.123101115611394362758801435120403766632080078125,
"e2e_output_tps_per_user_p50": 24.123124751169910950920893810689449310302734375,
"e2e_output_tps_per_user_p90": 24.123471687874545210661381133832037448883056640625,
"e2e_output_tps_per_user_p99": 24.123555017344248341260026791132986545562744140625,
"server_accept_len_effective": 0,
"server_spec_accepted_tokens": 0,
"chunk_inter_token_latency_avg": 0.0397709005924906933149287624473799951374530792236328125,
"chunk_inter_token_latency_p50": 0.0397698413867683953082377001919667236506938934326171875,
"chunk_inter_token_latency_p90": 0.039773042321603425952591948089320794679224491119384765625,
"chunk_inter_token_latency_p99": 0.03977423273567197437028397644098731689155101776123046875
}
],
"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": "max",
"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": [
51.5558366575427982070323196239769458770751953125
],
"requested_duration": 30,
"measurement_seconds": 39.7239213399589061737060546875,
"request_count_target": 4,
"aggregate_tokens_per_second": 51.5558366575427982070323196239769458770751953125,
"measurement_start_monotonic": 511317.3059726040228269994258880615234375,
"reported_measurement_seconds_rounded": 39.72392099999999715009835199452936649322509765625
},
{
"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": [
31.55395495133857508562869043089449405670166015625,
31.55395495133857508562869043089449405670166015625
],
"requested_duration": 30,
"measurement_seconds": 32.4523503180244006216526031494140625,
"request_count_target": 4,
"aggregate_tokens_per_second": 63.1079099026771501712573808617889881134033203125,
"measurement_start_monotonic": 511369.085347656975500285625457763671875,
"reported_measurement_seconds_rounded": 32.45235000000000269437805400229990482330322265625
},
{
"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": [
23.966279466550293619775402476079761981964111328125,
23.966279466550293619775402476079761981964111328125,
23.966279466550293619775402476079761981964111328125,
23.966279466550293619775402476079761981964111328125
],
"requested_duration": 30,
"measurement_seconds": 21.363349313964135944843292236328125,
"request_count_target": 4,
"aggregate_tokens_per_second": 95.8651178662011744791016099043190479278564453125,
"measurement_start_monotonic": 511413.59733222401700913906097412109375,
"reported_measurement_seconds_rounded": 21.363348999999999477950041182339191436767578125
}
],
"engine_source_commit": "fa0adf1899015f692ddce30391ee7e740a0910c9"
}Run file preserved verbatim in the corpus (unknown keys never dropped) · derived rows regenerate from it on import.