{"id":"run_9QHXZMzYBddmjC6T","owner":"anonymous-TWTejS","publishedAt":"2026-08-23T21:37:31.205Z","manifest":{"schema_version":"1.0","kind":"run","client_run_id":"vllm-2500d1d39b74-1","name":"NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 | 1x RTX 5090 | 1024->256 | C8","notes":"Text-only synthetic serving measurement. Exact token lengths, greedy decoding, streaming responses, and EOS ignored. No prompts, generated text, request bodies, or response bodies are retained in the aggregate artifact.","visibility":"public","license":"CC-BY-4.0","outcome":"completed","subject":{"type":"model","source":"huggingface","adapters":[],"revision":"e8f3c7c4de75ad84fe1bcef95d38eca76214480b","identifier":"nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4","modalities":["text"],"architecture":"NemotronHForCausalLM","quantization":{"format":"NVFP4","checkpoint":"modelopt"},"modifications":[],"revision_kind":"commit","parameter_count_total":30000000000,"parameter_count_active":3000000000},"environment":{"os":"Ubuntu 24.04.4 LTS","cpu":{"model":"AMD Ryzen Threadripper PRO 9955WX 16-Cores","sockets":1,"numa_nodes":1,"logical_cpus":32,"physical_cores":16},"kernel":"6.8.0-136-generic","topology":"One process restricted to GPU 0; other host GPUs idle.","accelerators":[{"count":1,"model":"NVIDIA GeForce RTX 5090","driver":"580.173.02","vendor":"NVIDIA","architecture":"Blackwell SM120","interconnect":"single device","memory_bytes_per_device":34190917632}],"driver_versions":{"nvidia":"580.173.02","cuda_compiler":"Build cuda_13.1.r13.1/compiler.37061995_0"},"host_architecture":"x86_64","system_memory_bytes":269585743872},"runtime":{"engine":"vllm","version":"0.27.1","parameters":{"server":{"dtype":"bfloat16","max_num_seqs":"95","max_model_len":"16384","kv_cache_dtype":"fp8","prefix_caching":false,"gpu_memory_utilization":"0.90","max_num_batched_tokens":"16384"},"dependencies":{"pytorch":"2.13.0+cu130","flashinfer":"0.6.16.post3","transformers":"5.15.1"},"load_balancer":{"type":"none","retry_other_replica_on_connect_failure":false}},"parallelism":{"tensor":1,"pipeline":1,"replicas":1},"launch_command":"CUDA_VISIBLE_DEVICES=0 vllm serve nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 --revision e8f3c7c4de75ad84fe1bcef95d38eca76214480b --served-model-name nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 --host 127.0.0.1 --dtype bfloat16 --kv-cache-dtype fp8 --gpu-memory-utilization 0.90 --max-model-len 16384 --max-num-batched-tokens 16384 --max-num-seqs 95 --no-enable-prefix-caching --disable-uvicorn-access-log --mamba-backend flashinfer --mamba-cache-mode align --port 9000 --tensor-parallel-size 1","safe_environment":{"HF_HUB_OFFLINE":"1","CUDA_VISIBLE_DEVICES":"0","VLLM_MARLIN_USE_ATOMIC_ADD":"1"}},"workload":{"mode":"online","task":"text_generation","client":{"name":"vllm bench serve","backend":"vllm","version":"0.27.1"},"dataset":{"name":"vllm-random","range_ratio":0,"random_prefix_tokens":0},"sampling":{"mode":"greedy","temperature":0},"stopping":{"ignore_eos":true,"max_tokens":256},"streaming":true,"tokenizer":{"revision":"e8f3c7c4de75ad84fe1bcef95d38eca76214480b","identifier":"nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4"},"concurrency":8,"random_seed":20260823,"input_tokens":{"kind":"fixed","value":1024},"endpoint_type":"openai_completions","output_tokens":{"kind":"fixed","value":256},"request_count":200,"arrival_process":"closed_loop","prefix_behavior":{"shared_prefix_tokens":0,"server_prefix_caching":false},"warmup_request_count":16,"request_rate_per_second":null},"measurement":{"notes":"Aggregate-only output; --save-detailed was not enabled.","warmup":"16 explicit requests before the curve; the harness also issues one unmeasured endpoint test immediately before each timed point.","harness":"vllm bench serve","version":"0.27.1","repetitions":1,"timing_scope":"Client-observed, streaming online serving interval after the harness initial test request.","benchmark_command":"vllm bench serve --backend vllm --host localhost --port 9000 --endpoint /v1/completions --model nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 --tokenizer nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 --dataset-name random --random-input-len 1024 --random-output-len 256 --random-range-ratio 0 --num-prompts 200 --request-rate inf --max-concurrency 8 --ignore-eos --temperature 0 --seed 20260823 --percentile-metrics ttft,tpot,itl,e2el --metric-percentiles 50,90,95,99 --disable-tqdm --save-result --result-dir results --result-filename nemotron35-lightning-30b-a3b-nvfp4__r1__i1024__o256__c8.json","retry_failure_treatment":"No client retries; failed requests are counted as failures.","client_overhead_included":true},"results":{"latency_ms":{"end_to_end":{"p50":1583.6592429550365,"p90":1693.830194650218,"p95":1703.427686507348,"p99":1715.5872783437371,"mean":1588.690427877009,"stddev":75.18482962633028,"sample_count":200,"measurement_status":"measured"},"inter_token":{"p50":5.175547092221677,"p90":5.659801955334842,"p95":5.796038976404815,"p99":6.077988415490837,"mean":5.2508729549910065,"stddev":4.17494292037529,"sample_count":200,"measurement_status":"measured"},"time_to_first_token":{"p50":270.9805004997179,"p90":272.55046302452683,"p95":272.9516806313768,"p99":319.01711547281593,"mean":249.71798035665415,"stddev":62.42725980815819,"sample_count":200,"measurement_status":"measured"},"time_per_output_token":{"p50":5.244095792409543,"p90":5.665034476905039,"p95":5.971588884325077,"p99":6.243736107477589,"mean":5.250872343217076,"stddev":0.3741959102208384,"sample_count":200,"measurement_status":"measured"}},"throughput":{"requests_per_second":{"value":5.034548533550064,"sample_count":200,"measurement_status":"measured"},"total_tokens_per_second":{"value":6444.222122944082,"sample_count":200,"measurement_status":"measured"},"output_tokens_per_second":{"value":1288.8444245888163,"sample_count":200,"measurement_status":"measured"}},"failed_requests":0,"duration_seconds":39.72550838813186,"total_input_tokens":204800,"successful_requests":200,"total_output_tokens":51200},"artifacts":[{"kind":"vllm_bench_serve_json","sha256":"2500d1d39b740270a3603107af5532a05ffd5abde44f40b2ab11775582eb345b","license":"CC-BY-4.0","filename":"nemotron35-lightning-30b-a3b-nvfp4__r1__i1024__o256__c8.json","media_type":"application/json","size_bytes":1715,"visibility":"private","client_artifact_id":"raw-vllm-serve","raw_requests_included":false}],"provenance":{"tool":"Runpile vllm importer completed from local evidence","source":"import","generated_at":"2026-08-23T21:37:30.653578+00:00","tool_version":"1.0"},"warnings":["vLLM reported uncalibrated FP8 attention q/prob scales of 1.0; serving performance is measured, but model quality was not evaluated.","vLLM auto-selected weight-only Marlin FP4 kernels and reported no native FP4 support for this ModelOpt W4A16 checkpoint on RTX 5090."]},"indexingState":"indexed","resultKey":"res_K9O8SBsdBkd9i5m1PS_0nl7y","eligibilityReasons":[],"artifactBacked":true,"downloads":[]}