nsys profile --output /root/arle-local-expert-segment/docs/trace-artifacts/2026-05-15-dsv4-deepep/nsys-single-token-segment-input/trace --force-overwrite=true --trace cuda,nvtx,osrt --capture-range=cudaProfilerApi --capture-range-end=stop --export=sqlite --kill=none /root/arle/target/release/infer --model-path /root/DeepSeek-V4-Flash --port 18142 --num-slots 1 --max-seq-len 4096 --mem-fraction-static 0.10 --kv-cache-dtype fp8 --deepseek-distributed-layers 43
