expert distributions

[CUHKSZzxy]

Refer to the dlBLAS. Not compatible with CUDA graph, used with --eager-mode

LMDEPLOY_DUMP_EXPERT_DISTRIBUTION=1 \
LMDEPLOY_EXPERT_DUMP_DIR="your_expert_distribution_dir" \
LMDEPLOY_DP_MASTER_ADDR=0.0.0.0 \
LMDEPLOY_DP_MASTER_PORT=29555 \
lmdeploy serve api_server \
    Qwen/Qwen3-235B-A22B-FP8 \
    --backend pytorch \
    --tp 1 \
    --dp 4 \
    --ep 4 \
    --proxy-url http://0.0.0.0:8001 \
    --nnodes 1 \
    --node-rank 0 \
    --eager-mode \
    --log-level INFO