{"id":"run_fZEvnfa2idDpkHzh","owner":"anonymous-EVjB2S","publishedAt":"2026-08-27T12:10:30.929Z","manifest":{"schema_version":"1.0","kind":"run","client_run_id":"rps-a500fba9ef4e8f61eea3f7ef225ee74b09d21c08","name":"google/gemma-4-12B-it | 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":"5926caa4ec0cac5cbfadaf4077420520de1d5205","identifier":"google/gemma-4-12B-it","modalities":["text"],"architecture":"Gemma4UnifiedForConditionalGeneration","quantization":{"format":"none","weights_dtype":"bf16"},"modifications":["Multimodal inputs disabled at serving time."],"revision_kind":"commit","parameter_count_total":12000000000,"parameter_count_active":12000000000},"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":"baseline"},"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":true,"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 google/gemma-4-12B-it --revision 5926caa4ec0cac5cbfadaf4077420520de1d5205 --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 --async-scheduling --no-enforce-eager --served-model-name google/gemma-4-12B-it --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":"5926caa4ec0cac5cbfadaf4077420520de1d5205","identifier":"google/gemma-4-12B-it"},"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":"0.23.0+cu129","extensions":{"runpile.declared_warmup_count":16,"runpile.max_gpu_temperature_c":59,"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 google/gemma-4-12B-it --tokenizer google/gemma-4-12B-it --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 gemma4-12b-bf16-tp1__baseline__serve-fixed-1k-256-v1__c16__r1__attempt1.json --num-warmups 16 --metadata campaign=runpile-rtx5090-param-sweeps-20260826-retry2 config_id=gemma4-12b-bf16-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":6771.070312475786,"p90":6794.358297483996,"p95":6798.9593733800575,"p99":6799.681630311534,"mean":6773.73219202309,"stddev":10.729545324700288,"sample_count":256,"measurement_status":"measured"},"inter_token":{"p50":19.964219070971012,"p90":20.103559968993068,"p95":20.195930427871644,"p99":20.419131191447374,"mean":21.65865740823267,"stddev":38.980088939907965,"sample_count":256,"measurement_status":"measured"},"time_to_first_token":{"p50":1686.8535843677819,"p90":1688.8580322265625,"p95":1689.097142778337,"p99":1711.9114941451699,"mean":1250.7746800583845,"stddev":517.3665419827317,"sample_count":256,"measurement_status":"measured"},"time_per_output_token":{"p50":20.026117506637878,"p90":23.60513703167146,"p95":25.903760440939784,"p99":25.953976570025965,"mean":21.65865690966551,"stddev":2.026416856790976,"sample_count":256,"measurement_status":"measured"}},"throughput":{"requests_per_second":{"value":2.3617259629871996,"sample_count":256,"measurement_status":"measured"},"total_tokens_per_second":{"value":3023.0092326236154,"sample_count":256,"measurement_status":"measured"},"output_tokens_per_second":{"value":604.6018465247231,"sample_count":256,"measurement_status":"measured"}},"failed_requests":0,"duration_seconds":108.39530242374167,"total_input_tokens":262144,"successful_requests":256,"total_output_tokens":65536},"artifacts":[{"kind":"vllm_bench_serve_json","sha256":"2b44446c8e16bfdb953771787ae3dd1a9d7d0b052f2eac35ea4bcb1abf9a8504","license":"CC-BY-4.0","filename":"gemma4-12b-bf16-tp1__baseline__serve-fixed-1k-256-v1__c16__r1.json","media_type":"application/json","size_bytes":1907,"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_wPP0prC_V-yPrh1eiYGCOlhN","eligibilityReasons":[],"artifactBacked":true,"downloads":[]}