{"id":"run_FW-OGBWRqx7n-e74","owner":"anonymous-EVjB2S","publishedAt":"2026-08-28T04:43:34.036Z","manifest":{"schema_version":"1.0","kind":"run","client_run_id":"rps-d30924b382a7cdd8a4131af9c7f24c6c42419aff","name":"unsloth/Qwen3.8-27B-NVFP4 | baseline | 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":"7d6f8d4d72f56b92b3cdbf22f156b90e1bab0108","identifier":"unsloth/Qwen3.8-27B-NVFP4","modalities":["text"],"architecture":"Qwen3_5ForConditionalGeneration","quantization":{"format":"NVFP4 mixed precision","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":"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"},"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":0.92,"max_num_batched_tokens":16384}},"parallelism":{"data":1,"expert":1,"tensor":1,"pipeline":1,"replicas":1},"launch_command":"CUDA_VISIBLE_DEVICES=0 vllm serve unsloth/Qwen3.8-27B-NVFP4 --revision 7d6f8d4d72f56b92b3cdbf22f156b90e1bab0108 --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 --async-scheduling --no-enforce-eager --served-model-name unsloth/Qwen3.8-27B-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":"cbdfc3a46fb4a6226c9325854d4822a9715b87c5f90978d2373938c9ab16008b","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":"7d6f8d4d72f56b92b3cdbf22f156b90e1bab0108","identifier":"unsloth/Qwen3.8-27B-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":60,"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 unsloth/Qwen3.8-27B-NVFP4 --tokenizer unsloth/Qwen3.8-27B-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 qwen38-27b-nvfp4-tp1__baseline__serve-fixed-8k-256-v1__c4__r1__attempt1.json --num-warmups 8 --metadata campaign=runpile-rtx5090-remaining-models-20260827-2 config_id=qwen38-27b-nvfp4-tp1 server_variant=baseline 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":8040.498492540792,"p90":8075.507495831697,"p95":8085.301309148781,"p99":8834.098854167387,"mean":8048.7340659419715,"stddev":215.5219094419861,"sample_count":128,"measurement_status":"measured"},"inter_token":{"p50":19.48888460174203,"p90":19.629741786047816,"p95":19.699753052555025,"p99":19.97008142527193,"mean":24.826801869166246,"stddev":66.99577838784852,"sample_count":128,"measurement_status":"measured"},"time_to_first_token":{"p50":1565.618614666164,"p90":3087.407494429499,"p95":3109.8413981031626,"p99":3119.996009990573,"mean":1717.8997853225155,"stddev":703.279248982032,"sample_count":128,"measurement_status":"measured"},"time_per_output_token":{"p50":25.390547712096105,"p90":28.364895960206493,"p95":28.45407498927385,"p99":28.476868046374587,"mean":24.826801100468455,"stddev":2.707175076253997,"sample_count":128,"measurement_status":"measured"}},"throughput":{"requests_per_second":{"value":0.4968901672249191,"sample_count":128,"measurement_status":"measured"},"total_tokens_per_second":{"value":4197.728132716116,"sample_count":128,"measurement_status":"measured"},"output_tokens_per_second":{"value":127.20388280957928,"sample_count":128,"measurement_status":"measured"}},"failed_requests":0,"duration_seconds":257.602199526038,"total_input_tokens":1048576,"successful_requests":128,"total_output_tokens":32768},"artifacts":[{"kind":"vllm_bench_serve_json","sha256":"11dadcdf5f0469d3fb8ed8d5c2e59c974c1c8922f843fc348e38cd4a0c05c0de","license":"CC-BY-4.0","filename":"qwen38-27b-nvfp4-tp1__baseline__serve-fixed-8k-256-v1__c4__r1.json","media_type":"application/json","size_bytes":1906,"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_NQkelp8ar_NiHyMC2NhDUKS-","eligibilityReasons":[],"artifactBacked":true,"downloads":[]}