{"id":"run_YU6G5_Z0LYBxsQ5g","owner":"anonymous-EVjB2S","publishedAt":"2026-08-30T21:20:08.236Z","manifest":{"schema_version":"1.0","kind":"run","client_run_id":"rps-daa1235f941b75de40cf5695193249e55cf08562","name":"nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 | baseline | 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 with DFlashDraftModel proposer","quantization":{"format":"target W4A16 NVFP4 plus ModelOpt W4A16-NVFP4 DFlash proposer","checkpoint":"ModelOpt"},"modifications":["Speculative dflash-draft checkpoint: nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4-DFlash@7fc1f1ff4b82b917efbd0710df0872c2bb89caa5."],"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":[{"role":"dflash-draft","revision":"7fc1f1ff4b82b917efbd0710df0872c2bb89caa5","repository":"nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4-DFlash","quantization":"ModelOpt W4A16_NVFP4; BF16 activations","weight_bytes":1177089556}]},"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\":\"dflash\",\"model\":\"nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4-DFlash\",\"revision\":\"7fc1f1ff4b82b917efbd0710df0872c2bb89caa5\",\"num_speculative_tokens\":3}'","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":"3e16e2ad994c5c1d9a4a8da014269d4c27123438a8026b79e3706fcf76186a36","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":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":61,"runpile.preemption_log_mentions":0,"runpile.readiness_request_count":1,"runpile.controller_source_sha256":"cab078b273cceada260983886369dfc896718627d072e44f74a1cca08b67675c","runpile.speculative_accept_length":2.988990235003756,"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-nvfp4-dflash3-tp1__baseline__serve-fixed-256-1k-v1__c16__r1__attempt1.json --num-warmups 8 --metadata campaign=runpile-rtx5090-nemotron-gemma4-specdecode-pilot-20260830-v2 config_id=nemotron35-nvfp4-dflash3-tp1 server_variant=baseline 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":3367.521662497893,"p90":6157.831379026174,"p95":11172.710040537631,"p99":13096.592812165618,"mean":4285.529614702682,"stddev":2345.4044300851674,"sample_count":128,"measurement_status":"measured"},"inter_token":{"p50":11.314709205180407,"p90":12.004746124148369,"p95":12.306404300034046,"p99":42.569405771791935,"mean":12.307380559882272,"stddev":6.481667449329652,"sample_count":128,"measurement_status":"measured"},"time_to_first_token":{"p50":65.71388454176486,"p90":292.09277583286166,"p95":292.9138721898198,"p99":293.35905513726175,"mean":108.6163014297199,"stddev":76.91776422596537,"sample_count":128,"measurement_status":"measured"},"time_per_output_token":{"p50":3.214654304680732,"p90":5.799620211914632,"p95":10.857062341799207,"p99":12.699853842081147,"mean":4.0830042162981055,"stddev":2.289351619493074,"sample_count":128,"measurement_status":"measured"}},"throughput":{"requests_per_second":{"value":3.4833424891639573,"sample_count":128,"measurement_status":"measured"},"total_tokens_per_second":{"value":4458.678386129865,"sample_count":128,"measurement_status":"measured"},"output_tokens_per_second":{"value":3566.9427089038923,"sample_count":128,"measurement_status":"measured"}},"failed_requests":0,"duration_seconds":36.746314896736294,"total_input_tokens":32768,"successful_requests":128,"total_output_tokens":131072},"artifacts":[{"kind":"vllm_bench_serve_json","sha256":"c07395c50e1d3c1f1c2663658c31e41e85a12ff38c1191e335fb28fdadb30a8b","license":"CC-BY-4.0","filename":"nemotron35-nvfp4-dflash3-tp1__baseline__serve-fixed-256-1k-v1__c16__r1.json","media_type":"application/json","size_bytes":2820,"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_H_0tcV_sSkSaHCdXHPQOvGFl","eligibilityReasons":[],"artifactBacked":true,"downloads":[]}