{"id":"run_MzxyZ1RYLi8O29jg","owner":"anonymous-EVjB2S","publishedAt":"2026-08-30T21:14:34.041Z","manifest":{"schema_version":"1.0","kind":"run","client_run_id":"rps-2d3354a6bed9f63969b316838df14485b4f0b95d","name":"RedHatAI/gemma-4-31B-it-NVFP4 | baseline | 8192->256 | C1","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":"5df75ccad9b74546f5be758a2a68acc2e16bbc9b","identifier":"RedHatAI/gemma-4-31B-it-NVFP4","modalities":["text"],"architecture":"Gemma4ForConditionalGeneration with Gemma4AssistantForCausalLM","quantization":{"format":"NVFP4 target plus BF16 MTP assistant","checkpoint":"compressed-tensors"},"modifications":["Multimodal inputs disabled at serving time.","Speculative mtp-assistant checkpoint: google/gemma-4-31B-it-assistant@627c5ec1458b9086b841a91e0512fd31fd2fbbf1."],"revision_kind":"commit","parameter_count_total":31000000000,"parameter_count_active":31000000000},"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":"mtp-assistant","revision":"627c5ec1458b9086b841a91e0512fd31fd2fbbf1","repository":"google/gemma-4-31B-it-assistant","quantization":"BF16","weight_bytes":939042560}]},"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=3 vllm serve RedHatAI/gemma-4-31B-it-NVFP4 --revision 5df75ccad9b74546f5be758a2a68acc2e16bbc9b --port 9003 --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 RedHatAI/gemma-4-31B-it-NVFP4 --disable-uvicorn-access-log --language-model-only --speculative-config '{\"method\":\"mtp\",\"model\":\"google/gemma-4-31B-it-assistant\",\"revision\":\"627c5ec1458b9086b841a91e0512fd31fd2fbbf1\",\"num_speculative_tokens\":1}'","build_identifier":"vllm-0.28.0-cu130-gguf-0.0.5","safe_environment":{"HF_HUB_OFFLINE":"1","CUDA_VISIBLE_DEVICES":"3","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":"2a478dc9220ebfc98b746ea9af2d673901ee229887b9c876e177567309a9439b","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":"5df75ccad9b74546f5be758a2a68acc2e16bbc9b","identifier":"RedHatAI/gemma-4-31B-it-NVFP4"},"extensions":{"runpile.profile_id":"serve-fixed-8k-256-v1"},"concurrency":1,"random_seed":0,"input_tokens":{"kind":"fixed","value":8192},"endpoint_type":"openai-compatible-completions","output_tokens":{"kind":"fixed","value":256},"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":53,"runpile.preemption_log_mentions":0,"runpile.readiness_request_count":1,"runpile.controller_source_sha256":"cab078b273cceada260983886369dfc896718627d072e44f74a1cca08b67675c","runpile.speculative_accept_length":1.6868142538110495,"runpile.independent_repetition_index":2},"repetitions":1,"timing_scope":"client-observed streaming request latency","benchmark_command":"vllm bench serve --backend vllm --base-url http://127.0.0.1:9003 --endpoint /v1/completions --model RedHatAI/gemma-4-31B-it-NVFP4 --tokenizer RedHatAI/gemma-4-31B-it-NVFP4 --dataset-name custom --dataset-path <EXACT_SYNTHETIC_DATASET> --custom-output-len 256 --skip-chat-template --disable-shuffle --num-prompts 128 --max-concurrency 1 --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 gemma4-31b-nvfp4-mtp1-tp1__baseline__serve-fixed-8k-256-v1__c1__r2__attempt1.json --num-warmups 8 --metadata campaign=runpile-rtx5090-nemotron-gemma4-specdecode-pilot-20260830-v2 config_id=gemma4-31b-nvfp4-mtp1-tp1 server_variant=baseline workload_key=serve-fixed-8k-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":3473.458589753136,"p90":6088.022520300001,"p95":6140.271225152537,"p99":6146.915779625997,"mean":3970.9726212604437,"stddev":961.2550515109667,"sample_count":128,"measurement_status":"measured"},"inter_token":{"p50":21.228738827630877,"p90":21.375338546931744,"p95":21.42627048306167,"p99":21.745781064964827,"mean":21.439135616601806,"stddev":2.5517940539507302,"sample_count":128,"measurement_status":"measured"},"time_to_first_token":{"p50":692.340831970796,"p90":701.909999223426,"p95":702.4796405807137,"p99":702.9849179554731,"mean":693.8000046611705,"stddev":6.099312021645084,"sample_count":128,"measurement_status":"measured"},"time_per_output_token":{"p50":10.871171234540789,"p90":21.155809689057513,"p95":21.352936182645898,"p99":21.358652463821947,"mean":12.85165731999715,"stddev":3.7679889718402655,"sample_count":128,"measurement_status":"measured"}},"throughput":{"requests_per_second":{"value":0.25182319752668453,"sample_count":128,"measurement_status":"measured"},"total_tokens_per_second":{"value":2127.402372705431,"sample_count":128,"measurement_status":"measured"},"output_tokens_per_second":{"value":64.46673856683124,"sample_count":128,"measurement_status":"measured"}},"failed_requests":0,"duration_seconds":508.2931249272078,"total_input_tokens":1048576,"successful_requests":128,"total_output_tokens":32768},"artifacts":[{"kind":"vllm_bench_serve_json","sha256":"3ddac0aa5d849d1c1b3c2b761d4e28bca92a2db3b9645303c7cfbb14b421d22c","license":"CC-BY-4.0","filename":"gemma4-31b-nvfp4-mtp1-tp1__baseline__serve-fixed-8k-256-v1__c1__r2.json","media_type":"application/json","size_bytes":2622,"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_-dWLu8sXZGSHpysM4ZEc5df5","eligibilityReasons":[],"artifactBacked":true,"downloads":[]}