{"id":"run_GufrPFUj3gf9V8QH","owner":"anonymous-TWTejS","publishedAt":"2026-08-23T21:02:46.524Z","manifest":{"schema_version":"1.0","kind":"run","client_run_id":"vllm-b1a43cb13066-1","name":"Qwen3.8-27B-NVFP4 | 1x RTX 5090 | 1024->256 | C1","notes":"Text-only synthetic serving measurement. Exact token lengths, greedy decoding, streaming responses, and EOS ignored. No prompts, generated text, request bodies, or response bodies are retained in the aggregate artifact.","visibility":"public","license":"CC-BY-4.0","outcome":"completed","subject":{"type":"model","source":"huggingface","revision":"7d6f8d4d72f56b92b3cdbf22f156b90e1bab0108","identifier":"unsloth/Qwen3.8-27B-NVFP4","modalities":["text"],"architecture":"Qwen3_5ForConditionalGeneration","quantization":{"format":"NVFP4","recipe":"Unsloth Dynamic V3.0 preview","checkpoint":"compressed-tensors"},"modifications":["Multimodal inputs disabled at serving time."],"revision_kind":"commit","parameter_count_total":27000000000,"parameter_count_active":27000000000},"environment":{"os":"Ubuntu 24.04.4 LTS","cpu":{"model":"AMD Ryzen Threadripper PRO 9955WX 16-Cores","sockets":1,"numa_nodes":1,"logical_cpus":32,"physical_cores":16},"kernel":"6.8.0-136-generic","topology":"One process restricted to GPU 0; other host GPUs idle.","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","cuda_compiler":"Build cuda_13.1.r13.1/compiler.37061995_0"},"host_architecture":"x86_64","system_memory_bytes":269585743872},"runtime":{"engine":"vllm","version":"0.27.1","parameters":{"server":{"dtype":"bfloat16","max_num_seqs":"26","max_model_len":"16384","kv_cache_dtype":"fp8","prefix_caching":false,"gpu_memory_utilization":"0.90","max_num_batched_tokens":"16384"},"dependencies":{"pytorch":"2.13.0+cu130","flashinfer":"0.6.16.post3","transformers":"5.15.1"},"load_balancer":{"type":"none","retry_other_replica_on_connect_failure":false}},"parallelism":{"tensor":1,"pipeline":1,"replicas":1},"launch_command":"CUDA_VISIBLE_DEVICES=0 vllm serve unsloth/Qwen3.8-27B-NVFP4 --revision 7d6f8d4d72f56b92b3cdbf22f156b90e1bab0108 --served-model-name unsloth/Qwen3.8-27B-NVFP4 --host 127.0.0.1 --dtype bfloat16 --kv-cache-dtype fp8 --gpu-memory-utilization 0.90 --max-model-len 16384 --max-num-batched-tokens 16384 --max-num-seqs 26 --no-enable-prefix-caching --disable-uvicorn-access-log --language-model-only --port 9000 --tensor-parallel-size 1","safe_environment":{"HF_HUB_OFFLINE":"1","CUDA_VISIBLE_DEVICES":"0","VLLM_MARLIN_USE_ATOMIC_ADD":"1"}},"workload":{"mode":"online","task":"text_generation","client":{"name":"vllm bench serve","backend":"vllm","version":"0.27.1"},"dataset":{"name":"vllm-random","range_ratio":0,"random_prefix_tokens":0},"sampling":{"mode":"greedy","temperature":0},"stopping":{"ignore_eos":true,"max_tokens":256},"streaming":true,"tokenizer":{"revision":"7d6f8d4d72f56b92b3cdbf22f156b90e1bab0108","identifier":"unsloth/Qwen3.8-27B-NVFP4"},"concurrency":1,"random_seed":20260823,"input_tokens":{"kind":"fixed","value":1024},"endpoint_type":"openai_completions","output_tokens":{"kind":"fixed","value":256},"request_count":200,"arrival_process":"closed_loop","prefix_behavior":{"shared_prefix_tokens":0,"server_prefix_caching":false},"warmup_request_count":16,"request_rate_per_second":null},"measurement":{"notes":"Aggregate-only output; --save-detailed was not enabled.","warmup":"16 explicit requests before the curve; the harness also issues one unmeasured endpoint test immediately before each timed point.","harness":"vllm bench serve","version":"0.27.1","repetitions":1,"timing_scope":"Client-observed, streaming online serving interval after the harness initial test request.","benchmark_command":"vllm bench serve --backend vllm --host localhost --port 9000 --endpoint /v1/completions --model unsloth/Qwen3.8-27B-NVFP4 --tokenizer unsloth/Qwen3.8-27B-NVFP4 --dataset-name random --random-input-len 1024 --random-output-len 256 --random-range-ratio 0 --num-prompts 200 --request-rate inf --max-concurrency 1 --ignore-eos --temperature 0 --seed 20260823 --percentile-metrics ttft,tpot,itl,e2el --metric-percentiles 50,90,95,99 --disable-tqdm --save-result --result-dir results --result-filename qwen38-27b-nvfp4__r1__i1024__o256__c1.json","retry_failure_treatment":"No client retries; failed requests are counted as failures.","client_overhead_included":true},"results":{"latency_ms":{"end_to_end":{"p50":3797.9976835194975,"p90":3798.9388219779357,"p95":3799.240821506828,"p99":3799.5833589229733,"mean":3797.9939120879862,"stddev":0.7425472378724276,"sample_count":200,"measurement_status":"measured"},"inter_token":{"p50":14.528246596455574,"p90":14.818546338938177,"p95":15.370776259806005,"p99":16.03026847122237,"mean":14.530949570370984,"stddev":0.43659794626935283,"sample_count":200,"measurement_status":"measured"},"time_to_first_token":{"p50":92.55417052190751,"p90":93.5234519885853,"p95":93.91606360441074,"p99":94.46411974728107,"mean":92.60196813149378,"stddev":0.7403230739463629,"sample_count":200,"measurement_status":"measured"},"time_per_output_token":{"p50":14.530868214262904,"p90":14.53315981061143,"p95":14.535543932396845,"p99":14.537234403430393,"mean":14.530948799829382,"stddev":0.002169624059170554,"sample_count":200,"measurement_status":"measured"}},"throughput":{"requests_per_second":{"value":0.2632912824548962,"sample_count":200,"measurement_status":"measured"},"total_tokens_per_second":{"value":337.0128415422671,"sample_count":200,"measurement_status":"measured"},"output_tokens_per_second":{"value":67.40256830845343,"sample_count":200,"measurement_status":"measured"}},"failed_requests":0,"duration_seconds":759.6149714309722,"total_input_tokens":204800,"successful_requests":200,"total_output_tokens":51200},"artifacts":[{"kind":"vllm_bench_serve_json","sha256":"b1a43cb13066206385bf76949704886d95214a9bda0157f48ded68fed2e1b9a1","license":"CC-BY-4.0","filename":"qwen38-27b-nvfp4__r1__i1024__o256__c1.json","media_type":"application/json","size_bytes":1658,"visibility":"private","client_artifact_id":"raw-vllm-serve","raw_requests_included":false}],"provenance":{"tool":"Runpile vllm importer completed from local evidence","source":"import","generated_at":"2026-08-23T21:02:45.925212+00:00","tool_version":"1.0"}},"indexingState":"indexed","resultKey":"res_URiduwnYYa2G-wiSqVSkP0ZH","eligibilityReasons":[],"artifactBacked":true,"downloads":[]}