{"run":"bench-ollama-qwen3-4b","concurrency":1,"wall_s":8.33,"total_gen_tokens":200,"aggregate_tps":24.0,"per_req_decode_tps":[25.7],"prefill_tps":[159],"max_latency_s":8.33} {"run":"bench-ollama-qwen3-4b","concurrency":1,"wall_s":7.89,"total_gen_tokens":200,"aggregate_tps":25.3,"per_req_decode_tps":[25.7],"prefill_tps":[1982],"max_latency_s":7.89} {"run":"bench-ollama-qwen3-4b","concurrency":2,"wall_s":15.83,"total_gen_tokens":400,"aggregate_tps":25.3,"per_req_decode_tps":[25.4,25.6],"prefill_tps":[1965,1997],"max_latency_s":15.83} {"run":"bench-ollama-qwen3-4b","concurrency":4,"wall_s":33.14,"total_gen_tokens":800,"aggregate_tps":24.1,"per_req_decode_tps":[25.3,25.4,22.3,24.4],"prefill_tps":[1950,1952,1977,1985],"max_latency_s":33.14} {"run":"bench-ollama-qwen3-4b","concurrency":8,"wall_s":63.99,"total_gen_tokens":1600,"aggregate_tps":25.0,"per_req_decode_tps":[25.1,24.9,25.3,25.3,25.2,25.1,25.1,25.0],"prefill_tps":[1932,1868,1924,1934,1952,1991,1938,1966],"max_latency_s":63.99} {"run":"bench-llamaserver-qwen3-4b-gpu","slots":1,"concurrency":1,"wall_s":6.5,"total_gen_tokens":200,"aggregate_tps":30.8,"per_req_decode_tps":[33.3],"max_latency_s":6.5} {"run":"bench-llamaserver-qwen3-4b-gpu","slots":1,"concurrency":4,"wall_s":25.96,"total_gen_tokens":800,"aggregate_tps":30.8,"per_req_decode_tps":[33.1,33.1,33.1,33.0],"max_latency_s":25.96} {"run":"bench-llamaserver-qwen3-4b-gpu","slots":1,"concurrency":8,"wall_s":52.18,"total_gen_tokens":1600,"aggregate_tps":30.7,"per_req_decode_tps":[33.0,33.0,33.0,33.0,33.0,33.0,32.9,32.6],"max_latency_s":52.18} {"run":"bench-llamaserver-qwen3-4b-gpu","slots":1,"concurrency":16,"wall_s":105.57,"total_gen_tokens":3200,"aggregate_tps":30.3,"per_req_decode_tps":[32.2,32.3,32.5,32.6,32.6,32.8,32.6,32.8,32.7,32.7,32.7,32.4,32.6,32.4,32.7,32.7],"max_latency_s":105.5} {"run":"bench-llamaserver-qwen3-4b-gpu","slots":4,"concurrency":1,"wall_s":6.65,"total_gen_tokens":200,"aggregate_tps":30.1,"per_req_decode_tps":[32.5],"max_latency_s":6.65} {"run":"bench-llamaserver-qwen3-4b-gpu","slots":4,"concurrency":4,"wall_s":12.26,"total_gen_tokens":800,"aggregate_tps":65.3,"per_req_decode_tps":[18.7,18.7,18.7,18.7],"max_latency_s":12.26} {"run":"bench-llamaserver-qwen3-4b-gpu","slots":4,"concurrency":8,"wall_s":24.51,"total_gen_tokens":1600,"aggregate_tps":65.3,"per_req_decode_tps":[18.8,18.8,18.8,18.8,18.8,18.8,18.8,18.8],"max_latency_s":24.51} {"run":"bench-llamaserver-qwen3-4b-gpu","slots":4,"concurrency":16,"wall_s":49.32,"total_gen_tokens":3200,"aggregate_tps":64.9,"per_req_decode_tps":[18.8,18.8,18.8,18.8,18.8,18.8,18.8,18.8,18.8,18.8,18.8,18.8,18.8,18.8,18.8,18.8],"max_latency_s":49.32} {"run":"bench-llamaserver-qwen3-4b-gpu","slots":8,"concurrency":1,"wall_s":6.62,"total_gen_tokens":200,"aggregate_tps":30.2,"per_req_decode_tps":[32.6],"max_latency_s":6.62} {"run":"bench-llamaserver-qwen3-4b-gpu","slots":8,"concurrency":4,"wall_s":12.21,"total_gen_tokens":800,"aggregate_tps":65.5,"per_req_decode_tps":[18.8,18.8,18.8,18.8],"max_latency_s":12.21} {"run":"bench-llamaserver-qwen3-4b-gpu","slots":8,"concurrency":8,"wall_s":19.69,"total_gen_tokens":1600,"aggregate_tps":81.2,"per_req_decode_tps":[12.0,12.0,12.0,12.0,12.0,12.0,12.0,12.0],"max_latency_s":19.69} {"run":"bench-llamaserver-qwen3-4b-gpu","slots":8,"concurrency":16,"wall_s":39.64,"total_gen_tokens":3200,"aggregate_tps":80.7,"per_req_decode_tps":[11.9,11.9,11.9,11.9,11.9,11.9,11.9,11.9,11.9,11.9,11.9,11.9,11.9,11.9,11.9,11.9],"max_latency_s":39.63} {"run":"bench-llamaserver-qwen3-4b-cpu","slots":1,"concurrency":1,"wall_s":20.01,"total_gen_tokens":200,"aggregate_tps":10.0,"per_req_decode_tps":[10.8],"max_latency_s":20.01} {"run":"bench-llamaserver-qwen3-4b-cpu","slots":1,"concurrency":4,"wall_s":80.55,"total_gen_tokens":800,"aggregate_tps":9.9,"per_req_decode_tps":[10.6,10.9,10.9,10.5],"max_latency_s":80.55} {"run":"bench-llamaserver-qwen3-4b-cpu","slots":1,"concurrency":8,"wall_s":160.32,"total_gen_tokens":1600,"aggregate_tps":10.0,"per_req_decode_tps":[10.7,11.0,10.6,10.8,11.0,10.4,10.8,10.9],"max_latency_s":160.32} {"run":"bench-llamaserver-qwen3-4b-cpu","slots":4,"concurrency":1,"wall_s":20.12,"total_gen_tokens":200,"aggregate_tps":9.9,"per_req_decode_tps":[10.8],"max_latency_s":20.12} {"run":"bench-llamaserver-qwen3-4b-cpu","slots":4,"concurrency":4,"wall_s":36.19,"total_gen_tokens":800,"aggregate_tps":22.1,"per_req_decode_tps":[6.6,6.6,6.6,6.6],"max_latency_s":36.19} {"run":"bench-llamaserver-qwen3-4b-cpu","slots":4,"concurrency":8,"wall_s":71.44,"total_gen_tokens":1600,"aggregate_tps":22.4,"per_req_decode_tps":[6.7,6.7,6.7,6.7,6.5,6.5,6.5,6.5],"max_latency_s":71.44} {"run":"bench-llamaserver-qwen3-4b-cpu","slots":8,"concurrency":1,"wall_s":20.0,"total_gen_tokens":200,"aggregate_tps":10.0,"per_req_decode_tps":[10.8],"max_latency_s":20.0} {"run":"bench-llamaserver-qwen3-4b-cpu","slots":8,"concurrency":4,"wall_s":35.91,"total_gen_tokens":800,"aggregate_tps":22.3,"per_req_decode_tps":[6.7,6.7,6.7,6.7],"max_latency_s":35.91} {"run":"bench-llamaserver-qwen3-4b-cpu","slots":8,"concurrency":8,"wall_s":78.49,"total_gen_tokens":1600,"aggregate_tps":20.4,"per_req_decode_tps":[2.9,2.9,2.9,2.9,2.9,2.9,2.9,2.9],"max_latency_s":78.49}