{"id":"run_fyUaMp-rKzwely1c","owner":"anonymous-TWTejS","publishedAt":"2026-08-24T00:13:49.879Z","manifest":{"schema_version":"1.0","kind":"run","client_run_id":"vllm-58c6ffd8008a-1","name":"Muse-Glimmer-30B-GGUF | 4x RTX 5090 | 1024->256 | C128","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","adapters":[],"revision":"70bf1b61ac09f91b24d39038091b41c582bc5d7a","identifier":"meta-models/Muse-Glimmer-30B-GGUF","modalities":["text"],"architecture":"MuseGlimmerForConditionalGeneration","quantization":{"format":"Dynamic Q4_K_XL","checkpoint":"GGUF"},"modifications":[],"revision_kind":"commit","parameter_count_total":30000000000,"parameter_count_active":30000000000},"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":"Four independent one-GPU replicas, one per GPU, behind the recorded round-robin HTTP proxy.","accelerators":[{"count":4,"model":"NVIDIA GeForce RTX 5090","driver":"580.173.02","vendor":"NVIDIA","architecture":"Blackwell SM120","interconnect":"PCIe within one NUMA node; no NVLink","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":"llama.cpp","version":"b10353","parameters":{"server":{"dflash":false,"kv_cache":"Q8_0","slots_per_replica":64,"context_tokens_per_slot":2048,"context_tokens_per_replica":131072},"load_balancer":{"type":"round_robin_llama_cpp_compat_proxy","cache_prompt":false,"sse_usage_split":true,"tokenizer_schema_translation":true}},"parallelism":{"tensor":1,"pipeline":1,"replicas":4},"launch_command":"CUDA_VISIBLE_DEVICES=0 llama-server -m Muse-Glimmer-30B-KQuant-Dynamic-Q4_K_XL.gguf -a meta-models/Muse-Glimmer-30B-GGUF -ngl 99 -c 131072 -np 64 --flash-attn on -ctk q8_0 -ctv q8_0 --host 127.0.0.1 --port 9001 --jinja --temp 1.0 --top-p 0.95 --top-k 64 -t 16 -tb 16\nCUDA_VISIBLE_DEVICES=1 llama-server -m Muse-Glimmer-30B-KQuant-Dynamic-Q4_K_XL.gguf -a meta-models/Muse-Glimmer-30B-GGUF -ngl 99 -c 131072 -np 64 --flash-attn on -ctk q8_0 -ctv q8_0 --host 127.0.0.1 --port 9002 --jinja --temp 1.0 --top-p 0.95 --top-k 64 -t 16 -tb 16\nCUDA_VISIBLE_DEVICES=2 llama-server -m Muse-Glimmer-30B-KQuant-Dynamic-Q4_K_XL.gguf -a meta-models/Muse-Glimmer-30B-GGUF -ngl 99 -c 131072 -np 64 --flash-attn on -ctk q8_0 -ctv q8_0 --host 127.0.0.1 --port 9003 --jinja --temp 1.0 --top-p 0.95 --top-k 64 -t 16 -tb 16\nCUDA_VISIBLE_DEVICES=3 llama-server -m Muse-Glimmer-30B-KQuant-Dynamic-Q4_K_XL.gguf -a meta-models/Muse-Glimmer-30B-GGUF -ngl 99 -c 131072 -np 64 --flash-attn on -ctk q8_0 -ctv q8_0 --host 127.0.0.1 --port 9004 --jinja --temp 1.0 --top-p 0.95 --top-k 64 -t 16 -tb 16\nRUNPILE_BACKENDS=<replica-urls> RUNPILE_PROXY_PORT=9000 python llama_compat_proxy.py","source_revision":"f8def7fe168bab245fbf15d3f18b26dbb1ef73c8","safe_environment":{"CUDA_VISIBLE_DEVICES":"one distinct GPU index per replica"}},"workload":{"mode":"online","task":"text_generation","client":{"name":"vllm bench serve","backend":"openai","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":"a4e59da52a7bc87ae7251dd5545c0dd437c44b68","identifier":"meta-models/Muse-Glimmer-30B"},"concurrency":128,"random_seed":20260823,"input_tokens":{"kind":"fixed","value":1024},"endpoint_type":"openai_completions","output_tokens":{"kind":"fixed","value":256},"request_count":512,"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. For llama.cpp, the recorded compatibility proxy disables prompt caching, translates tokenize/detokenize schemas, and splits combined SSE usage so the harness verifies exact server token counts.","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 openai --host localhost --port 9000 --endpoint /v1/completions --model meta-models/Muse-Glimmer-30B-GGUF --tokenizer meta-models/Muse-Glimmer-30B --dataset-name random --random-input-len 1024 --random-output-len 256 --random-range-ratio 0 --num-prompts 512 --request-rate inf --max-concurrency 128 --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 muse-glimmer-30b-dynamic-q4-k-xl__r4__i1024__o256__c128.json","retry_failure_treatment":"No client retries; failed requests are counted as failures.","client_overhead_included":true},"results":{"latency_ms":{"end_to_end":{"p50":26168.193052057177,"p90":35196.18870350532,"p95":36491.099596815184,"p99":38653.61566632054,"mean":25484.543009984918,"stddev":7971.726491528232,"sample_count":512,"measurement_status":"measured"},"inter_token":{"p50":74.88128799013793,"p90":201.7206327756867,"p95":226.97045478271315,"p99":413.3422682550736,"mean":95.92038245909947,"stddev":80.57160014963311,"sample_count":512,"measurement_status":"measured"},"time_to_first_token":{"p50":657.3868159903213,"p90":7493.745647859762,"p95":9298.947637865786,"p99":10158.012563667726,"mean":2081.4685939476476,"stddev":2809.682146605481,"sample_count":512,"measurement_status":"measured"},"time_per_output_token":{"p50":96.22322077607262,"p90":124.05349988800785,"p95":129.37168541752422,"p99":137.1881454760921,"mean":91.77676241583244,"stddev":27.70315101188698,"sample_count":512,"measurement_status":"measured"}},"throughput":{"requests_per_second":{"value":4.751532776552276,"sample_count":512,"measurement_status":"measured"},"total_tokens_per_second":{"value":6081.961953986913,"sample_count":512,"measurement_status":"measured"},"output_tokens_per_second":{"value":1216.3923907973826,"sample_count":512,"measurement_status":"measured"}},"failed_requests":0,"duration_seconds":107.75470234081149,"total_input_tokens":524288,"successful_requests":512,"total_output_tokens":131072},"artifacts":[{"kind":"vllm_bench_serve_json","sha256":"58c6ffd8008a5bbc5b14963cfa7a08e2bcac737a4e3f728a322a049a6a443fec","license":"CC-BY-4.0","filename":"muse-glimmer-30b-dynamic-q4-k-xl__r4__i1024__o256__c128.json","media_type":"application/json","size_bytes":1685,"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":"2026-08-24T00:13:49.438927+00:00","tool_version":"1.0"}},"indexingState":"indexed","resultKey":"res_Afr6481TGJdtta0GOICA_Uzz","eligibilityReasons":[],"artifactBacked":true,"downloads":[]}