{"id":"run_hNcnWISmq5KnfZNj","owner":"anonymous-EVjB2S","publishedAt":"2026-08-30T21:15:20.796Z","manifest":{"schema_version":"1.0","kind":"run","client_run_id":"rps-83cb63eb55750427d52f4901bda292c36ca5288d","name":"RedHatAI/gemma-4-31B-it-NVFP4 | baseline | 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":"5df75ccad9b74546f5be758a2a68acc2e16bbc9b","identifier":"RedHatAI/gemma-4-31B-it-NVFP4","modalities":["text"],"architecture":"Gemma4ForConditionalGeneration","quantization":{"format":"NVFP4","publisher":"Red Hat AI","checkpoint":"compressed-tensors"},"modifications":["Multimodal inputs disabled at serving time."],"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":[]},"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","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":"7b3a595f0a42c9e568ebe49422d9a636f6f46f8c66bc8cb4557f4f2a17650e52","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-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":59,"runpile.preemption_log_mentions":0,"runpile.readiness_request_count":1,"runpile.controller_source_sha256":"cab078b273cceada260983886369dfc896718627d072e44f74a1cca08b67675c","runpile.independent_repetition_index":1},"repetitions":1,"timing_scope":"client-observed streaming request latency","benchmark_command":"vllm bench serve --backend vllm --base-url http://127.0.0.1:9003 --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 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 gemma4-31b-nvfp4-target-tp1__baseline__serve-fixed-1k-256-v1__c16__r1__attempt1.json --num-warmups 16 --metadata campaign=runpile-rtx5090-nemotron-gemma4-specdecode-pilot-20260830-v2 config_id=gemma4-31b-nvfp4-target-tp1 server_variant=baseline 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":6585.456859320402,"p90":12788.528912700713,"p95":12804.283127305098,"p99":12983.282615803182,"mean":8599.385174653435,"stddev":2856.2676995091633,"sample_count":256,"measurement_status":"measured"},"inter_token":{"p50":22.261866834014654,"p90":22.501366771757603,"p95":22.602269402705133,"p99":78.9434453425929,"mean":23.731064302567166,"stddev":16.892793297539757,"sample_count":256,"measurement_status":"measured"},"time_to_first_token":{"p50":708.1313389353454,"p90":6480.654504150152,"p95":6487.311004078947,"p99":6846.886362298392,"mean":2547.9639879667957,"stddev":2756.446443135029,"sample_count":256,"measurement_status":"measured"},"time_per_output_token":{"p50":23.747287982819127,"p90":24.797019685673362,"p95":25.003814378607213,"p99":25.04539136610487,"mean":23.731063477202504,"stddev":0.7267838767287409,"sample_count":256,"measurement_status":"measured"}},"throughput":{"requests_per_second":{"value":1.8065552027445515,"sample_count":256,"measurement_status":"measured"},"total_tokens_per_second":{"value":2312.390659513026,"sample_count":256,"measurement_status":"measured"},"output_tokens_per_second":{"value":462.4781319026052,"sample_count":256,"measurement_status":"measured"}},"failed_requests":0,"duration_seconds":141.70615966292098,"total_input_tokens":262144,"successful_requests":256,"total_output_tokens":65536},"artifacts":[{"kind":"vllm_bench_serve_json","sha256":"f605f75a93d5b181277b6850986e0ba2a6cc94e13cb819ae58e304cb0f20618c","license":"CC-BY-4.0","filename":"gemma4-31b-nvfp4-target-tp1__baseline__serve-fixed-1k-256-v1__c16__r1.json","media_type":"application/json","size_bytes":1946,"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_JaZlsAJTCIrW9Hxly32CYsvk","eligibilityReasons":[],"artifactBacked":true,"downloads":[]}