Quick Reference
| Config Field | SGLang CLI Argument | Description |
|---|---|---|
tp | --tp-size, --tensor-parallel-size | Tensor Parallelism - splits model across GPUs |
dp | --dp-size, --data-parallel-size | Data Parallelism - runs multiple model replicas |
ep | --ep-size, --expert-parallel-size, --ep | Expert Parallelism - distributes MoE experts |
attn_dp_size | --attn-dp-size, --attention-data-parallel-size | Attention Data Parallelism - data-parallel attention groups inside the TP group, TP for FFN (hybrid) |
enable_dp_attention | --enable-dp-attention | Deprecated. Use —attn-dp-size <dp-size> instead of —dp-size <dp-size> —enable-dp-attention. |