{"id":"run_RGHlaSQzZ43srC1g","owner":"anonymous-EVjB2S","publishedAt":"2026-08-30T21:26:19.989Z","manifest":{"schema_version":"1.0","kind":"run","client_run_id":"rps-fab07c114c2ab2eb0a2e7373d82b2ecd5dfa2600","name":"nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 | baseline | 1024->256 | C16","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":16,"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":61,"runpile.preemption_log_mentions":0,"runpile.readiness_request_count":1,"runpile.controller_source_sha256":"cab078b273cceada260983886369dfc896718627d072e44f74a1cca08b67675c","runpile.speculative_accept_length":1.976206700738217,"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 256 --max-concurrency 16 --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__c16__r2__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=2 aggregate_only=true","retry_failure_treatment":"Failed requests counted; no client retries.","client_overhead_included":true},"results":{"latency_ms":{"end_to_end":{"p50":1706.2360348645598,"p90":1794.8214244097471,"p95":1849.528340739198,"p99":2412.0641669956944,"mean":1705.2602466246753,"stddev":195.5022227386788,"sample_count":256,"measurement_status":"measured"},"inter_token":{"p50":8.15631914883852,"p90":8.77093467861414,"p95":9.309618640691033,"p99":76.01884737610817,"mean":10.83644445984766,"stddev":17.194759629580783,"sample_count":256,"measurement_status":"measured"},"time_to_first_token":{"p50":331.99934917502105,"p90":538.9096545986831,"p95":575.1529417466372,"p99":576.5457437373698,"mean":307.23202547596884,"stddev":177.55676788099925,"sample_count":256,"measurement_status":"measured"},"time_per_output_token":{"p50":5.32480085944282,"p90":6.517861417366886,"p95":6.61178618879514,"p99":7.391374697726136,"mean":5.4824636123478685,"stddev":0.86142063234637,"sample_count":256,"measurement_status":"measured"}},"throughput":{"requests_per_second":{"value":9.201755454013853,"sample_count":256,"measurement_status":"measured"},"total_tokens_per_second":{"value":11778.24698113773,"sample_count":256,"measurement_status":"measured"},"output_tokens_per_second":{"value":2355.6493962275463,"sample_count":256,"measurement_status":"measured"}},"failed_requests":0,"duration_seconds":27.820778467692435,"total_input_tokens":262144,"successful_requests":256,"total_output_tokens":65536},"artifacts":[{"kind":"vllm_bench_serve_json","sha256":"b4f1c672f80cd8bdc797336abf117a10e8a7f55e356987e972f30a5fa00058e2","license":"CC-BY-4.0","filename":"nemotron35-nvfp4-mtp1-tp1__baseline__serve-fixed-1k-256-v1__c16__r2.json","media_type":"application/json","size_bytes":2659,"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":[]}