aiperf profile --model Qwen/Qwen3-30B-A3B-FP8 --url http://localhost:8000 --endpoint-type chat --tokenizer Qwen/Qwen3-30B-A3B-FP8 --streaming --synthetic-input-tokens-mean 1024 --synthetic-input-tokens-stddev 0 --output-tokens-mean 256 --output-tokens-stddev 0 --concurrency 32 --request-count 300 --warmup-request-count 30 --artifact-dir results/sglang_Qwen3-30B-A3B-FP8_c32_isl1024_osl256_20260828_171951