{"id":"run_AHKS717Z228rrvoH","owner":"anonymous-EVjB2S","publishedAt":"2026-08-30T21:33:20.734Z","manifest":{"schema_version":"1.0","kind":"run","client_run_id":"rps-fbf0587d7b1d12d54b91ecb6d0a5e859202c8ccb","name":"nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 | baseline | 8192->256 | C4","notes":"Aggregate-only synthetic online serving measurement. It reports serving performance and does not evaluate model quality.","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":"W4A16 NVFP4 mixed precision","checkpoint":"ModelOpt"},"modifications":[],"revision_kind":"commit","parameter_count_total":30000000000,"parameter_count_active":3000000000},"environment":{"os":"Linux-6.8.0-136-generic-x86_64-with-glibc2.39","cpu":{"model":"x86_64"},"kernel":"6.8.0-136-generic","topology":"One server engine on one GPU.","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"},"host_architecture":"x86_64","system_memory_bytes":269585743872},"runtime":{"engine":"vllm","version":"0.28.0","extensions":{"runpile.server_variant":"baseline","runpile.auxiliary_models":[]},"parameters":{"server":{"dtype":"bfloat16","max_num_seqs":32,"enforce_eager":false,"max_model_len":9216,"kv_cache_dtype":"fp8","prefix_caching":false,"chunked_prefill":true,"expert_parallel":false,"async_scheduling":true,"gpu_memory_utilization":0.88,"max_num_batched_tokens":16384}},"parallelism":{"data":1,"expert":1,"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 --port 9000 --host 127.0.0.1 --dtype bfloat16 --gpu-memory-utilization 0.88 --max-model-len 9216 --max-num-batched-tokens 16384 --max-num-seqs 32 --tensor-parallel-size 1 --pipeline-parallel-size 1 --kv-cache-dtype fp8 --distributed-executor-backend mp --no-enable-prefix-caching --enable-chunked-prefill --async-scheduling --no-enforce-eager --served-model-name nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 --disable-uvicorn-access-log --mamba-backend flashinfer --mamba-cache-mode align --mamba-ssm-cache-dtype float16 --enable-mamba-cache-stochastic-rounding --mamba-cache-philox-rounds 5","build_identifier":"vllm-0.28.0-cu130-gguf-0.0.5","safe_environment":{"HF_HUB_OFFLINE":"1","CUDA_VISIBLE_DEVICES":"0","TOKENIZERS_PARALLELISM":"false"}},"workload":{"mode":"online","task":"text_generation","client":{"name":"vllm bench serve","backend":"vllm","version":"0.28.0","placement":"same-host"},"dataset":{"sha256":"834037d264dccfd0f703d2b0878332ae34da85e0bc229ab820909a51f6609acb","generator":"runpile-sweeps-tokenizer-verified-jsonl","synthetic":true,"generator_revision":"cab078b273cceada260983886369dfc896718627d072e44f74a1cca08b67675c","shared_prefix_tokens":0},"sampling":{"mode":"greedy","temperature":0},"stopping":{"ignore_eos":true,"max_output_tokens":256},"streaming":true,"tokenizer":{"revision":"e8f3c7c4de75ad84fe1bcef95d38eca76214480b","identifier":"nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4"},"extensions":{"runpile.profile_id":"serve-fixed-8k-256-v1"},"concurrency":4,"random_seed":0,"input_tokens":{"kind":"fixed","value":8192},"endpoint_type":"openai-compatible-completions","output_tokens":{"kind":"fixed","value":256},"request_count":128,"arrival_process":"closed_loop","prefix_behavior":{"cache_state":"warm-host","cache_enabled":false,"shared_prefix_tokens":0},"warmup_request_count":8,"request_rate_per_second":null},"measurement":{"notes":"Aggregate-only output.","warmup":"8 unmeasured warmups per invocation after one readiness request.","harness":"vllm bench serve","version":"0.28.0","extensions":{"runpile.declared_warmup_count":8,"runpile.max_gpu_temperature_c":56,"runpile.preemption_log_mentions":0,"runpile.readiness_request_count":1,"runpile.controller_source_sha256":"cab078b273cceada260983886369dfc896718627d072e44f74a1cca08b67675c","runpile.independent_repetition_index":2},"repetitions":1,"timing_scope":"client-observed streaming request latency","benchmark_command":"vllm bench serve --backend vllm --base-url http://127.0.0.1:9000 --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 custom --dataset-path <EXACT_SYNTHETIC_DATASET> --custom-output-len 256 --skip-chat-template --disable-shuffle --num-prompts 128 --max-concurrency 4 --request-rate inf --seed 0 --ignore-eos --temperature 0 --percentile-metrics ttft,tpot,itl,e2el --metric-percentiles 50,90,95,99 --disable-tqdm --save-result --result-dir <local-path> --result-filename nemotron35-nvfp4-target-tp1__baseline__serve-fixed-8k-256-v1__c4__r2__attempt1.json --num-warmups 8 --metadata campaign=runpile-rtx5090-nemotron-gemma4-specdecode-pilot-20260830-v2 config_id=nemotron35-nvfp4-target-tp1 server_variant=baseline workload_key=serve-fixed-8k-256-v1 repetition=2 aggregate_only=true","retry_failure_treatment":"Failed requests counted; no client retries.","client_overhead_included":true},"results":{"latency_ms":{"end_to_end":{"p50":2085.009931586683,"p90":2112.9252176731825,"p95":2116.43687391188,"p99":2125.018481905572,"mean":2084.3452544577303,"stddev":21.644253877250076,"sample_count":128,"measurement_status":"measured"},"inter_token":{"p50":4.162037977948785,"p90":4.355160146951675,"p95":4.416138934902847,"p99":4.583545993082227,"mean":5.135447650336929,"stddev":19.441180766344765,"sample_count":128,"measurement_status":"measured"},"time_to_first_token":{"p50":900.3191587980837,"p90":1030.7886733673513,"p95":1032.097917678766,"p99":1036.9537807861343,"mean":774.8062731407117,"stddev":309.4708026656812,"sample_count":128,"measurement_status":"measured"},"time_per_output_token":{"p50":4.640999215417633,"p90":7.138081983681403,"p95":7.189572918064454,"p99":7.237043550767589,"mean":5.135446985556936,"stddev":1.216037353778188,"sample_count":128,"measurement_status":"measured"}},"throughput":{"requests_per_second":{"value":1.918884695151418,"sample_count":128,"measurement_status":"measured"},"total_tokens_per_second":{"value":16210.73790463918,"sample_count":128,"measurement_status":"measured"},"output_tokens_per_second":{"value":491.234481958763,"sample_count":128,"measurement_status":"measured"}},"failed_requests":0,"duration_seconds":66.70541503792629,"total_input_tokens":1048576,"successful_requests":128,"total_output_tokens":32768},"artifacts":[{"kind":"vllm_bench_serve_json","sha256":"6ceb3b12ab571bd758bec6b5e3a603a9a578896806a41d1f33840c4a64897509","license":"CC-BY-4.0","filename":"nemotron35-nvfp4-target-tp1__baseline__serve-fixed-8k-256-v1__c4__r2.json","media_type":"application/json","size_bytes":1973,"visibility":"private","client_artifact_id":"raw-vllm-serve","raw_requests_included":false}],"provenance":{"tool":"runpile HTTP importer","source":"import","tool_version":"1.0"},"warnings":["Review imported facts against the local model, hardware, launch command, and benchmark command before publication."]},"indexingState":"indexed","resultKey":"res_E99gp0OkkNYbeZYMKsOhE6Gw","eligibilityReasons":[],"artifactBacked":true,"downloads":[]}