{"id":"run_UFGlXBUP9NAQD_sQ","owner":"anonymous-EVjB2S","publishedAt":"2026-08-30T21:26:45.977Z","manifest":{"schema_version":"1.0","kind":"run","client_run_id":"rps-5535ac7f2fe38b48d74d7bdcaca69c3d5482e70e","name":"nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 | baseline | 1024->256 | C1","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 with embedded NemotronHMTPModel head","quantization":{"format":"W4A16 NVFP4 mixed precision with checkpoint-native MTP tensors","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 --speculative-config '{\"method\":\"mtp\",\"num_speculative_tokens\":1}'","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":"84dd4d29ed5b7f3dce95f6533eca5bf01876b237ac81467c83e658f75a47b217","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-1k-256-v1"},"concurrency":1,"random_seed":0,"input_tokens":{"kind":"fixed","value":1024},"endpoint_type":"openai-compatible-completions","output_tokens":{"kind":"fixed","value":256},"request_count":256,"arrival_process":"closed_loop","prefix_behavior":{"cache_state":"warm-host","cache_enabled":false,"shared_prefix_tokens":0},"warmup_request_count":16,"request_rate_per_second":null},"measurement":{"notes":"Aggregate-only output.","warmup":"16 unmeasured warmups per invocation after one readiness request.","harness":"vllm bench serve","version":"0.28.0","extensions":{"runpile.declared_warmup_count":16,"runpile.max_gpu_temperature_c":58,"runpile.preemption_log_mentions":0,"runpile.readiness_request_count":1,"runpile.controller_source_sha256":"cab078b273cceada260983886369dfc896718627d072e44f74a1cca08b67675c","runpile.speculative_accept_length":1.975532911356476,"runpile.independent_repetition_index":1},"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 256 --max-concurrency 1 --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-mtp1-tp1__baseline__serve-fixed-1k-256-v1__c1__r1__attempt1.json --num-warmups 16 --metadata campaign=runpile-rtx5090-nemotron-gemma4-specdecode-pilot-20260830-v2 config_id=nemotron35-nvfp4-mtp1-tp1 server_variant=baseline workload_key=serve-fixed-1k-256-v1 repetition=1 aggregate_only=true","retry_failure_treatment":"Failed requests counted; no client retries.","client_overhead_included":true},"results":{"latency_ms":{"end_to_end":{"p50":518.8446380198002,"p90":570.975448936224,"p95":575.3920415882021,"p99":626.0756714269519,"mean":530.3019640650746,"stddev":41.27543716982169,"sample_count":256,"measurement_status":"measured"},"inter_token":{"p50":3.7228879518806934,"p90":4.040371440351009,"p95":4.11106301471591,"p99":4.229722041636705,"mean":3.7653353975468016,"stddev":0.21708410480127335,"sample_count":256,"measurement_status":"measured"},"time_to_first_token":{"p50":43.81207260303199,"p90":46.74316896125674,"p95":48.23842435143888,"p99":51.80679487530142,"mean":44.206180251421756,"stddev":1.8850266296744829,"sample_count":256,"measurement_status":"measured"},"time_per_output_token":{"p50":1.8622496347947448,"p90":2.064270611998497,"p95":2.0842190379934276,"p99":2.267105639912188,"mean":1.906257975739815,"stddev":0.16117789954197045,"sample_count":256,"measurement_status":"measured"}},"throughput":{"requests_per_second":{"value":1.8854694456557788,"sample_count":256,"measurement_status":"measured"},"total_tokens_per_second":{"value":2413.400890439397,"sample_count":256,"measurement_status":"measured"},"output_tokens_per_second":{"value":482.68017808787937,"sample_count":256,"measurement_status":"measured"}},"failed_requests":0,"duration_seconds":135.77520473208278,"total_input_tokens":262144,"successful_requests":256,"total_output_tokens":65536},"artifacts":[{"kind":"vllm_bench_serve_json","sha256":"b051d22459e16c8c1a7e873e186c97522dad00a38bda0a63ccae0351204afacf","license":"CC-BY-4.0","filename":"nemotron35-nvfp4-mtp1-tp1__baseline__serve-fixed-1k-256-v1__c1__r1.json","media_type":"application/json","size_bytes":2663,"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_K9O8SBsdBkd9i5m1PS_0nl7y","eligibilityReasons":[],"artifactBacked":true,"downloads":[]}