{"id":"run_fHL70cqgxKpV-z06","owner":"anonymous-EVjB2S","publishedAt":"2026-08-29T19:25:01.768Z","manifest":{"schema_version":"1.0","kind":"run","client_run_id":"rps-f237a08c3f4735b737e56051f9129f144861206a","name":"gittensor-model-hub/Qwen3.8-27B-NVFP4-RTX5090 | baseline | 256->1024 | C1","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":"0cc27958cefbbe231782ec8511de8c4eb5233348","identifier":"gittensor-model-hub/Qwen3.8-27B-NVFP4-RTX5090","modalities":["text"],"architecture":"Qwen3_5ForConditionalGeneration","quantization":{"format":"NVFP4 W4A4","recipe":"gittensor-model-hub RTX 5090 target checkpoint; BF16 speculative drafter is a runtime modification","checkpoint":"NVIDIA ModelOpt"},"modifications":["Speculative draft-bf16 checkpoint: incoai/Qwen3.8-27B-DFlash2@dedf8df68adfb1afeaf7b7480c0a0243108177b4."],"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":"sglang","version":"0.0.0.dev1+g5f55db35e","extensions":{"runpile.server_variant":"baseline","runpile.auxiliary_models":[{"role":"draft-bf16","revision":"dedf8df68adfb1afeaf7b7480c0a0243108177b4","repository":"incoai/Qwen3.8-27B-DFlash2","quantization":"BF16","weight_bytes":3849104785}]},"parameters":{"server":{"dtype":"bfloat16","context_length":9216,"kv_cache_dtype":"fp8_e4m3","attention_backend":"flashinfer","cuda_graph_max_bs":1,"mem_fraction_static":0.9,"chunked_prefill_size":1024,"max_running_requests":1}},"parallelism":{"data":1,"expert":1,"tensor":1,"pipeline":1,"replicas":1},"launch_command":"CUDA_VISIBLE_DEVICES=0 sglang serve --model-path gittensor-model-hub/Qwen3.8-27B-NVFP4-RTX5090 --revision 0cc27958cefbbe231782ec8511de8c4eb5233348 --served-model-name gittensor-model-hub/Qwen3.8-27B-NVFP4-RTX5090 --port 9000 --host 127.0.0.1 --dtype bfloat16 --mem-fraction-static 0.9 --context-length 9216 --max-running-requests 1 --tp 1 --kv-cache-dtype fp8_e4m3 --chunked-prefill-size 1024 --attention-backend flashinfer --cuda-graph-max-bs 1 --trust-remote-code --reasoning-parser qwen3 --tool-call-parser qwen3_coder --mamba-full-memory-ratio 5.61 --mamba-radix-cache-strategy extra_buffer_lazy --mamba-ssm-dtype bfloat16 --cuda-graph-bs-prefill 64 128 256 512 1024 --enable-memory-saver --enable-hierarchical-cache --hicache-ratio 2 --hicache-write-policy write_through --speculative-algorithm DFLASH --speculative-draft-model-path incoai/Qwen3.8-27B-DFlash2 --speculative-draft-model-revision dedf8df68adfb1afeaf7b7480c0a0243108177b4 --speculative-draft-model-quantization unquant --speculative-num-draft-tokens 8","build_identifier":"lmsysorg/sglang@sha256:616a3e97f45191af975896cfa644279096cb31bd408a071c2e99ca7209c3cafe","safe_environment":{"HF_HUB_OFFLINE":"1","CUDA_VISIBLE_DEVICES":"0","TOKENIZERS_PARALLELISM":"false"}},"workload":{"mode":"online","task":"text_generation","client":{"name":"sglang benchmark serving","backend":"sglang","version":"0.0.0.dev1+g5f55db35e","placement":"same-host"},"dataset":{"sha256":"20954ec817c4b70aa26473fd6b50c397ddc8acd8ccd2e58b1c74128586b39b62","generator":"runpile-sweeps-tokenizer-verified-jsonl","synthetic":true,"generator_revision":"acfe836377740b09ca2b148328ed3e98b8989baa77300bb70a8c1df9048a9199","shared_prefix_tokens":0},"sampling":{"mode":"greedy","temperature":0},"stopping":{"ignore_eos":true,"max_output_tokens":1024},"streaming":true,"tokenizer":{"revision":"0cc27958cefbbe231782ec8511de8c4eb5233348","identifier":"gittensor-model-hub/Qwen3.8-27B-NVFP4-RTX5090"},"extensions":{"runpile.profile_id":"serve-fixed-256-1k-v1"},"concurrency":1,"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":"sglang benchmark serving","version":"0.0.0.dev1+g5f55db35e","extensions":{"runpile.declared_warmup_count":8,"runpile.max_gpu_temperature_c":50,"runpile.preemption_log_mentions":2,"runpile.readiness_request_count":1,"runpile.controller_source_sha256":"acfe836377740b09ca2b148328ed3e98b8989baa77300bb70a8c1df9048a9199","runpile.speculative_accept_length":7.014409722222222,"runpile.independent_repetition_index":1},"repetitions":1,"timing_scope":"client-observed streaming request latency","benchmark_command":"sglang bench serving --backend sglang-oai --base-url http://127.0.0.1:9000 --model gittensor-model-hub/Qwen3.8-27B-NVFP4-RTX5090 --served-model-name gittensor-model-hub/Qwen3.8-27B-NVFP4-RTX5090 --tokenizer gittensor-model-hub/Qwen3.8-27B-NVFP4-RTX5090 --dataset-name custom --dataset-path <EXACT_SYNTHETIC_DATASET> --sharegpt-output-len 1024 --num-prompts 128 --max-concurrency 1 --request-rate inf --seed 0 --temperature 0 --disable-tqdm --output-file <local-path> --warmup-requests 8 --tag runpile-rtx5090-qwen38-dflash2-pilot-20260829-v2:qwen38-27b-nvfp4-sglang-dflash2-bf16-tp1:baseline:serve-fixed-256-1k-v1:c1:r1","retry_failure_treatment":"Failed requests counted; no client retries.","client_overhead_included":true},"results":{"latency_ms":{"end_to_end":{"sample_count":128,"measurement_status":"measured"},"inter_token":{"p50":2.035790530499071,"p90":2.0705918432213366,"p95":2.690367323035995,"p99":16.248925030231476,"mean":2.360254199739371,"stddev":2.4319463586698675,"sample_count":128,"measurement_status":"measured"},"time_to_first_token":{"p50":43.955194065347314,"p90":44.27056387066841,"p95":44.352819374762475,"p99":44.468281362205744,"mean":43.777589013188845,"stddev":1.4233878057678868,"sample_count":128,"measurement_status":"measured"},"time_per_output_token":{"p50":2.0353493323041634,"p90":2.1468091607257676,"p95":2.9933321546446328,"p99":11.782495341735315,"mean":2.346547797824919,"stddev":1.56996089280687,"sample_count":128,"measurement_status":"measured"}},"throughput":{"requests_per_second":{"value":0.4090622422225122,"sample_count":128,"measurement_status":"measured"},"input_tokens_per_second":{"value":104.71993400896312,"sample_count":128,"measurement_status":"measured"},"output_tokens_per_second":{"value":418.8797360358525,"sample_count":128,"measurement_status":"measured"}},"failed_requests":0,"duration_seconds":312.9108159788884,"total_input_tokens":32768,"successful_requests":128,"total_output_tokens":131072},"artifacts":[{"kind":"sglang_bench_serving_json","sha256":"af69c94ad7defc2db1bcdb5e11009a27aeaabf465eff39e38ad0a339394bdcdf","license":"CC-BY-4.0","filename":"qwen38-27b-nvfp4-sglang-dflash2-bf16-tp1__baseline__serve-fixed-256-1k-v1__c1__r1.json","media_type":"application/json","size_bytes":43598,"visibility":"private","client_artifact_id":"raw-sglang-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_vezlK4JRfkkXbZjSaP9G7mwz","eligibilityReasons":[],"artifactBacked":true,"downloads":[]}