Checklist

• I searched related issues but found no solution.
• The bug persists in the latest version.
• Issues without environment info and a minimal reproducible demo are hard to resolve and may receive no feedback.
• If this is not a bug report but a general question, please start a discussion at https://github.com/sgl-project/sglang/discussions. Otherwise, it will be closed.
• Please use English. Otherwise, it will be closed.

Describe the bug

SGLang fails to load some Qwen3 MoE checkpoints because qwen3_moe.py assumes config.rope_parameters["rope_theta"] always exists.
However, some valid Qwen3 configs expose RoPE as top-level rope_theta (and optional rope_scaling) without rope_parameters, which
triggers an AttributeError.

Expected behavior is to support both config styles:

• new style: config.rope_parameters["rope_theta"]
• old style: config.rope_theta / config.rope_scaling

A compatibility fallback in qwen3_moe.py would fix this and avoid startup failures across different Qwen3 checkpoint formats.

I will submit a PR to address this issue. #20931

Reproduction

python3 -m sglang.launch_server
--model-path /gfs/space/chatrl/public/models/Qwen3-30B-A3B-Instruct-2507
--trust-remote-code
--host 0.0.0.0
--port 30000
--ep-size 2
--tp-size 2 \

Environment

root@d68dddca53dc1ca316399de49c20639c-taskrole1-0:/gfs/platform/public/infra/lxr/sglang# python3 -m sglang.check_env
Python: 3.12.3 (main, Nov 6 2025, 13:44:16) [GCC 13.3.0]
CUDA available: True
GPU 0,1: NVIDIA H800 80GB
GPU 0,1 Compute Capability: 9.0
CUDA_HOME: /usr/local/cuda
NVCC: Cuda compilation tools, release 12.9, V12.9.86
CUDA Driver Version: 580.105.08
PyTorch: 2.9.1+cu129
sglang: 0.0.0.dev1+gdce8b0606
sgl_kernel: 0.3.21
flashinfer_python: 0.5.3
flashinfer_cubin: 0.5.3
flashinfer_jit_cache: 0.5.3+cu129
triton: 3.5.1
transformers: 4.57.1
torchao: 0.9.0
numpy: 1.26.4
aiohttp: 3.13.3
fastapi: 0.128.0
hf_transfer: 0.1.9
huggingface_hub: 0.36.0
interegular: 0.3.3
modelscope: 1.33.0
orjson: 3.11.5
outlines: 0.1.11
packaging: 25.0
psutil: 7.2.1
pydantic: 2.12.5
python-multipart: 0.0.21
pyzmq: 27.1.0
uvicorn: 0.40.0
uvloop: 0.22.1
vllm: Module Not Found
xgrammar: 0.1.27
openai: 2.6.1
tiktoken: 0.12.0
anthropic: 0.75.0
litellm: Module Not Found
decord2: 3.0.0
NVIDIA Topology:
GPU0 GPU1 NIC0 NIC1 NIC2 NIC3 NIC4 NIC5 NIC6 NIC7 NIC8 NIC9 CPU Affinity NUMA Affinity GPU NUMA ID
GPU0 X NV18 SYS PXB PXB PIX PXB SYS SYS SYS SYS PXB 0-45,92-137 0 N/A
GPU1 NV18 X SYS PXB PXB PXB PIX SYS SYS SYS SYS PXB 0-45,92-137 0 N/A
NIC0 SYS SYS X SYS SYS SYS SYS SYS SYS SYS SYS SYS
NIC1 PXB PXB SYS X PXB PXB PXB SYS SYS SYS SYS PXB
NIC2 PXB PXB SYS PXB X PXB PXB SYS SYS SYS SYS PXB
NIC3 PIX PXB SYS PXB PXB X PXB SYS SYS SYS SYS PXB
NIC4 PXB PIX SYS PXB PXB PXB X SYS SYS SYS SYS PXB
NIC5 SYS SYS SYS SYS SYS SYS SYS X PXB PXB PXB SYS
NIC6 SYS SYS SYS SYS SYS SYS SYS PXB X PXB PXB SYS
NIC7 SYS SYS SYS SYS SYS SYS SYS PXB PXB X PXB SYS
NIC8 SYS SYS SYS SYS SYS SYS SYS PXB PXB PXB X SYS
NIC9 PXB PXB SYS PXB PXB PXB PXB SYS SYS SYS SYS X

Legend:

X = Self
SYS = Connection traversing PCIe as well as the SMP interconnect between NUMA nodes (e.g., QPI/UPI)
NODE = Connection traversing PCIe as well as the interconnect between PCIe Host Bridges within a NUMA node
PHB = Connection traversing PCIe as well as a PCIe Host Bridge (typically the CPU)
PXB = Connection traversing multiple PCIe bridges (without traversing the PCIe Host Bridge)
PIX = Connection traversing at most a single PCIe bridge
NV# = Connection traversing a bonded set of # NVLinks

NIC Legend:

NIC0: mlx5_0
NIC1: mlx5_gdr_0
NIC2: mlx5_gdr_1
NIC3: mlx5_gdr_2
NIC4: mlx5_gdr_3
NIC5: mlx5_gdr_4
NIC6: mlx5_gdr_5
NIC7: mlx5_gdr_6
NIC8: mlx5_gdr_7
NIC9: mlx5_stor_0

ulimit soft: 1048576