GUIDELLM__MP_CONTEXT_TYPE=forkserver
guidellm benchmark run --target http://localhost:8003 --model Qwen3-4B-GPTQ-W4A8-zpfix --processor /home/ckl/projects/arle/infer/models/Qwen3-4B-GPTQ-W4A8-zpfix --profile concurrent --data prompt_tokens=4096\,prompt_tokens_min=4096\,prompt_tokens_max=4096\,output_tokens=256\,output_tokens_min=256\,output_tokens_max=256 --max-seconds 42 --random-seed 20260416 --output-dir /home/ckl/projects/arle/bench-output/2026-05-09-baseline-B7-w4a8-c4-4k-r1 --backend openai_http --backend-kwargs \{\"validate_backend\":\ \"/v1/models\"\,\ \"request_format\":\ \"/v1/completions\"\} --disable-console-interactive --outputs json --outputs csv --outputs html --rate 4 --warmup 10
