{"id":"run_EbSvHpdEliAjuYVw","owner":"anonymous-EVjB2S","publishedAt":"2026-08-28T04:34:08.742Z","manifest":{"schema_version":"1.0","kind":"run","client_run_id":"rps-b90c00e55b75f3c7f9e3e9f07415439f2c0daa31","name":"nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 | nemotron-ssu-horizontal | 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","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-ssu-horizontal"},"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 --mamba-ssu-algorithm horizontal","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":"0e0945825728f70eee1188087607793934a8cbde9fe12a4cc0c6192429ce7fd3","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":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":53,"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 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-lightning-30b-a3b-nvfp4-tp1__nemotron-ssu-horizontal__serve-fixed-1k-256-v1__c16__r1__attempt1.json --num-warmups 16 --metadata campaign=runpile-rtx5090-remaining-models-20260827-2 config_id=nemotron35-lightning-30b-a3b-nvfp4-tp1 server_variant=nemotron-ssu-horizontal 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":2296.7996620573103,"p90":2324.363566469401,"p95":2369.425434502773,"p99":2370.949684153311,"mean":2296.551465164157,"stddev":29.43435933779433,"sample_count":256,"measurement_status":"measured"},"inter_token":{"p50":6.990380119532347,"p90":7.295199064537883,"p95":7.40725330542773,"p99":7.642740183509886,"mean":7.084633895219132,"stddev":7.463082558630159,"sample_count":256,"measurement_status":"measured"},"time_to_first_token":{"p50":518.2820507325232,"p90":520.6379075534642,"p95":532.1276988834143,"p99":533.2514608511701,"mean":489.96992095635505,"stddev":113.86469883297589,"sample_count":256,"measurement_status":"measured"},"time_per_output_token":{"p50":6.980804257167906,"p90":7.205049857935484,"p95":8.727708813624785,"p99":8.880006973266454,"mean":7.084633506697262,"stddev":0.4585290016553236,"sample_count":256,"measurement_status":"measured"}},"throughput":{"requests_per_second":{"value":6.965322430077304,"sample_count":256,"measurement_status":"measured"},"total_tokens_per_second":{"value":8915.61271049895,"sample_count":256,"measurement_status":"measured"},"output_tokens_per_second":{"value":1783.1225420997898,"sample_count":256,"measurement_status":"measured"}},"failed_requests":0,"duration_seconds":36.75350316800177,"total_input_tokens":262144,"successful_requests":256,"total_output_tokens":65536},"artifacts":[{"kind":"vllm_bench_serve_json","sha256":"0600979115d6dbc4b282f8e384152334132337b90f5b11fe1c5cb70f30845cbd","license":"CC-BY-4.0","filename":"nemotron35-lightning-30b-a3b-nvfp4-tp1__nemotron-ssu-horizontal__serve-fixed-1k-256-v1__c16__r1.json","media_type":"application/json","size_bytes":1984,"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":[]}