{"id":"run_APmGBIAv0Col8Qxz","owner":"anonymous-EVjB2S","publishedAt":"2026-08-28T04:31:10.947Z","manifest":{"schema_version":"1.0","kind":"run","client_run_id":"rps-8f74e8efa82f5d2e231d729a13940bc422636389","name":"nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 | nemotron-humming | 256->1024 | 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","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":"nemotron-humming"},"parameters":{"server":{"dtype":"bfloat16","max_num_seqs":64,"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":0.92,"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 0.92 --max-model-len 9216 --max-num-batched-tokens 16384 --max-num-seqs 64 --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 --moe-backend humming --linear-backend humming","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":"b685595ca55c529f1b8fda05ff0f8363f4179ccc5d7b7efbd5ae371193379fac","generator":"runpile-sweeps-tokenizer-verified-jsonl","synthetic":true,"generator_revision":"f08cb0bccfbd93f7ea4cee52d5b921a8250619c0ed7abd572cd2159a829f7d28","shared_prefix_tokens":0},"sampling":{"mode":"greedy","temperature":0},"stopping":{"ignore_eos":true,"max_output_tokens":1024},"streaming":true,"tokenizer":{"revision":"e8f3c7c4de75ad84fe1bcef95d38eca76214480b","identifier":"nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4"},"extensions":{"runpile.profile_id":"serve-fixed-256-1k-v1"},"concurrency":16,"random_seed":0,"input_tokens":{"kind":"fixed","value":256},"endpoint_type":"openai-compatible-completions","output_tokens":{"kind":"fixed","value":1024},"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":50,"runpile.preemption_log_mentions":0,"runpile.readiness_request_count":1,"runpile.controller_source_sha256":"f08cb0bccfbd93f7ea4cee52d5b921a8250619c0ed7abd572cd2159a829f7d28","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 1024 --skip-chat-template --disable-shuffle --num-prompts 128 --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-lightning-30b-a3b-nvfp4-tp1__nemotron-humming__serve-fixed-256-1k-v1__c16__r1__attempt1.json --num-warmups 8 --metadata campaign=runpile-rtx5090-remaining-models-20260827-2 config_id=nemotron35-lightning-30b-a3b-nvfp4-tp1 server_variant=nemotron-humming workload_key=serve-fixed-256-1k-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":8040.725730825216,"p90":8250.470422860235,"p95":8250.763475405984,"p99":8251.019786135294,"mean":8042.113672043342,"stddev":150.0216370140322,"sample_count":128,"measurement_status":"measured"},"inter_token":{"p50":7.714477134868503,"p90":8.021041797474027,"p95":8.112678374163806,"p99":8.292237543500958,"mean":7.706416265293267,"stddev":1.2995980280750035,"sample_count":128,"measurement_status":"measured"},"time_to_first_token":{"p50":171.7656732071191,"p90":188.90738431364298,"p95":189.48163378518075,"p99":190.00253181438893,"mean":158.4499350283295,"stddev":41.66357697793689,"sample_count":128,"measurement_status":"measured"},"time_per_output_token":{"p50":7.740874634327355,"p90":7.895858118033814,"p95":7.8959983010855535,"p99":8.004958879440215,"mean":7.7064161652150664,"stddev":0.1491760684764102,"sample_count":128,"measurement_status":"measured"}},"throughput":{"requests_per_second":{"value":1.9893051766774976,"sample_count":128,"measurement_status":"measured"},"total_tokens_per_second":{"value":2546.3106261471967,"sample_count":128,"measurement_status":"measured"},"output_tokens_per_second":{"value":2037.0485009177576,"sample_count":128,"measurement_status":"measured"}},"failed_requests":0,"duration_seconds":64.34407425299287,"total_input_tokens":32768,"successful_requests":128,"total_output_tokens":131072},"artifacts":[{"kind":"vllm_bench_serve_json","sha256":"f2be1e884e8ddcf6c52fa2abf1de4bf64d904fe65b30fccd224d90b19be51dd3","license":"CC-BY-4.0","filename":"nemotron35-lightning-30b-a3b-nvfp4-tp1__nemotron-humming__serve-fixed-256-1k-v1__c16__r1.json","media_type":"application/json","size_bytes":1983,"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_Iu1JbBKm_1pdTdxPs7hwlM-c","eligibilityReasons":[],"artifactBacked":true,"downloads":[]}