{"id":"run_rT8CVflR0Kd8APoq","owner":"anonymous-EVjB2S","publishedAt":"2026-08-27T12:23:45.554Z","manifest":{"schema_version":"1.0","kind":"run","client_run_id":"rps-7b4b0b4c0c4c80f7589feb36d1e25decce14e70d","name":"nvidia/Gemma-4-26B-A4B-NVFP4 | sync-scheduling | 1024->256 | C32","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":"a19cfe00be84568a6867111c9a68c9c44fdcffe6","identifier":"nvidia/Gemma-4-26B-A4B-NVFP4","modalities":["text"],"architecture":"Gemma4ForConditionalGeneration","quantization":{"format":"NVFP4","checkpoint":"modelopt/compressed-tensors"},"modifications":["Multimodal inputs disabled at serving time."],"revision_kind":"commit","parameter_count_total":26000000000,"parameter_count_active":4000000000},"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":"0.23.0+cu129","extensions":{"runpile.server_variant":"sync-scheduling"},"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":false,"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/Gemma-4-26B-A4B-NVFP4 --revision a19cfe00be84568a6867111c9a68c9c44fdcffe6 --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 --no-async-scheduling --no-enforce-eager --served-model-name nvidia/Gemma-4-26B-A4B-NVFP4 --disable-uvicorn-access-log --language-model-only","build_identifier":"vllm-0.23.0-cu129","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":"0.23.0+cu129","placement":"same-host"},"dataset":{"generator":"runpile-sweeps-tokenizer-verified-jsonl","synthetic":true,"shared_prefix_tokens":0},"sampling":{"mode":"greedy","temperature":0},"stopping":{"ignore_eos":true,"max_output_tokens":256},"streaming":true,"tokenizer":{"revision":"a19cfe00be84568a6867111c9a68c9c44fdcffe6","identifier":"nvidia/Gemma-4-26B-A4B-NVFP4"},"extensions":{"runpile.profile_id":"serve-fixed-1k-256-v1"},"concurrency":32,"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":"0.23.0+cu129","extensions":{"runpile.declared_warmup_count":16,"runpile.max_gpu_temperature_c":53,"runpile.preemption_log_mentions":0,"runpile.readiness_request_count":1,"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/Gemma-4-26B-A4B-NVFP4 --tokenizer nvidia/Gemma-4-26B-A4B-NVFP4 --dataset-name custom --dataset-path <EXACT_SYNTHETIC_DATASET> --custom-output-len 256 --skip-chat-template --disable-shuffle --num-prompts 256 --max-concurrency 32 --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 gemma4-26b-a4b-nvfp4-tp1__sync-scheduling__serve-fixed-1k-256-v1__c32__r1__attempt1.json --num-warmups 16 --metadata campaign=runpile-rtx5090-param-sweeps-20260826-retry2 config_id=gemma4-26b-a4b-nvfp4-tp1 server_variant=sync-scheduling 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":4360.025390516967,"p90":4486.026896396652,"p95":4487.192935193889,"p99":4487.4232520814985,"mean":4388.239404388514,"stddev":58.86600339853516,"sample_count":256,"measurement_status":"measured"},"inter_token":{"p50":13.18185799755156,"p90":13.792525930330157,"p95":13.938486110419035,"p99":14.42349455319345,"mean":14.38286805332966,"stddev":22.26371619412703,"sample_count":256,"measurement_status":"measured"},"time_to_first_token":{"p50":618.3948169928044,"p90":953.3059759996831,"p95":1064.3789143068716,"p99":1065.7723166747019,"mean":720.6081664880912,"stddev":256.36116280287865,"sample_count":256,"measurement_status":"measured"},"time_per_output_token":{"p50":14.587546195652262,"p90":15.431239459590584,"p95":16.675763870315517,"p99":17.22378999524403,"mean":14.3828675996095,"stddev":1.0288375528048972,"sample_count":256,"measurement_status":"measured"}},"throughput":{"requests_per_second":{"value":7.289559417953441,"sample_count":256,"measurement_status":"measured"},"total_tokens_per_second":{"value":9330.636054980405,"sample_count":256,"measurement_status":"measured"},"output_tokens_per_second":{"value":1866.127210996081,"sample_count":256,"measurement_status":"measured"}},"failed_requests":0,"duration_seconds":35.118720531929284,"total_input_tokens":262144,"successful_requests":256,"total_output_tokens":65536},"artifacts":[{"kind":"vllm_bench_serve_json","sha256":"03bb64057ffea0f9a24143b89c0349bf2c8598266d17d29ae0094459fceee650","license":"CC-BY-4.0","filename":"gemma4-26b-a4b-nvfp4-tp1__sync-scheduling__serve-fixed-1k-256-v1__c32__r1.json","media_type":"application/json","size_bytes":1926,"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_QulgfeHNK01g5oAcPxfWv0ws","eligibilityReasons":[],"artifactBacked":true,"downloads":[]}