{"id":"run_UeB5ytUp-kIr2xJ1","owner":"anonymous-EVjB2S","publishedAt":"2026-08-30T21:13:34.093Z","manifest":{"schema_version":"1.0","kind":"run","client_run_id":"rps-32e26e108c78299977ef78d299ad134396d7de90","name":"RedHatAI/gemma-4-31B-it-NVFP4 | baseline | 256->1024 | C8","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":"43dfac229922b03b51294ab8ee6cb5744d397d2ce10e25a67a45582b31cf678c","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":1024},"streaming":true,"tokenizer":{"revision":"5df75ccad9b74546f5be758a2a68acc2e16bbc9b","identifier":"RedHatAI/gemma-4-31B-it-NVFP4"},"extensions":{"runpile.profile_id":"serve-fixed-256-1k-v1"},"concurrency":8,"random_seed":0,"input_tokens":{"kind":"fixed","value":256},"endpoint_type":"openai-compatible-completions","output_tokens":{"kind":"fixed","value":1024},"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":1.3412490362374712,"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 1024 --skip-chat-template --disable-shuffle --num-prompts 128 --max-concurrency 8 --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-256-1k-v1__c8__r1__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-256-1k-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":12030.367152066901,"p90":18256.12351349555,"p95":18265.975025645457,"p99":18303.561056242324,"mean":13598.726615095075,"stddev":4434.529658001455,"sample_count":128,"measurement_status":"measured"},"inter_token":{"p50":17.707477789372206,"p90":17.9713468067348,"p95":18.078036326915026,"p99":18.42159595340491,"mean":17.708673535811208,"stddev":1.251891247697448,"sample_count":128,"measurement_status":"measured"},"time_to_first_token":{"p50":65.13132597319782,"p90":77.234459342435,"p95":98.10038940049708,"p99":146.5987750934437,"mean":71.37544911165605,"stddev":18.24947890167315,"sample_count":128,"measurement_status":"measured"},"time_per_output_token":{"p50":11.656535929569634,"p90":17.773176619431638,"p95":17.790143655154232,"p99":17.82813755928731,"mean":13.223217171049285,"stddev":4.336394333299593,"sample_count":128,"measurement_status":"measured"}},"throughput":{"requests_per_second":{"value":0.5647120729526374,"sample_count":128,"measurement_status":"measured"},"total_tokens_per_second":{"value":722.8314533793758,"sample_count":128,"measurement_status":"measured"},"output_tokens_per_second":{"value":578.2651627035007,"sample_count":128,"measurement_status":"measured"}},"failed_requests":0,"duration_seconds":226.66418185597286,"total_input_tokens":32768,"successful_requests":128,"total_output_tokens":131072},"artifacts":[{"kind":"vllm_bench_serve_json","sha256":"1f67ada866e7da6b1d22d90c8f6b6993cf7280e9a5dc574bbff1188a799c0e15","license":"CC-BY-4.0","filename":"gemma4-31b-nvfp4-mtp1-tp1__baseline__serve-fixed-256-1k-v1__c8__r1.json","media_type":"application/json","size_bytes":2623,"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_bqqCBzml0aonQDvpAkMmHrSD","eligibilityReasons":[],"artifactBacked":true,"downloads":[]}