{"id":"run_0-WR4SiHyciBiBso","owner":"anonymous-EVjB2S","publishedAt":"2026-08-28T04:22:44.095Z","manifest":{"schema_version":"1.0","kind":"run","client_run_id":"rps-474b99be31e6ff3204abadb583ec44e95c892c22","name":"RedHatAI/gemma-4-31B-it-NVFP4 | sync-scheduling | 8192->256 | C4","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":"sync-scheduling"},"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":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 RedHatAI/gemma-4-31B-it-NVFP4 --revision 5df75ccad9b74546f5be758a2a68acc2e16bbc9b --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 32 --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 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":"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":"56e8e142448f2b694160f4a9a9baa643f7a421657883703a11779fea2a1f8183","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":"5df75ccad9b74546f5be758a2a68acc2e16bbc9b","identifier":"RedHatAI/gemma-4-31B-it-NVFP4"},"extensions":{"runpile.profile_id":"serve-fixed-8k-256-v1"},"concurrency":4,"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":59,"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 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 4 --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-tp1__sync-scheduling__serve-fixed-8k-256-v1__c4__r1__attempt1.json --num-warmups 8 --metadata campaign=runpile-rtx5090-remaining-models-20260827-2 config_id=gemma4-31b-nvfp4-tp1 server_variant=sync-scheduling workload_key=serve-fixed-8k-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":7131.3763812649995,"p90":7140.0438914541155,"p95":7148.118565324694,"p99":7823.704882729799,"mean":7129.910754880257,"stddev":207.65024713642825,"sample_count":128,"measurement_status":"measured"},"inter_token":{"p50":17.23249116912484,"p90":17.50827906653285,"p95":17.592061078175902,"p99":18.135656458325684,"mean":23.039084353242252,"stddev":62.48216215540924,"sample_count":128,"measurement_status":"measured"},"time_to_first_token":{"p50":1385.9550645574927,"p90":1391.0612702835351,"p95":1399.56661243923,"p99":2080.025688125752,"mean":1254.9444867254351,"stddev":362.2080094627225,"sample_count":128,"measurement_status":"measured"},"time_per_output_token":{"p50":22.530809081360403,"p90":25.19198063687951,"p95":25.20802112332746,"p99":25.226968609910532,"mean":23.039083404528714,"stddev":1.4115806728553544,"sample_count":128,"measurement_status":"measured"}},"throughput":{"requests_per_second":{"value":0.5609480451636755,"sample_count":128,"measurement_status":"measured"},"total_tokens_per_second":{"value":4738.88908554273,"sample_count":128,"measurement_status":"measured"},"output_tokens_per_second":{"value":143.60269956190092,"sample_count":128,"measurement_status":"measured"}},"failed_requests":0,"duration_seconds":228.1851253490895,"total_input_tokens":1048576,"successful_requests":128,"total_output_tokens":32768},"artifacts":[{"kind":"vllm_bench_serve_json","sha256":"bba6f919e1d44f94ce06275b3fefc33644b9399e3567ea562cf5d83f294181e6","license":"CC-BY-4.0","filename":"gemma4-31b-nvfp4-tp1__sync-scheduling__serve-fixed-8k-256-v1__c4__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_5HRejR4Q0jwyaH1xv7KFUPgU","eligibilityReasons":[],"artifactBacked":true,"downloads":[]}