Checklist
Describe the bug
When running sglang.launch_server with a ModelOpt FP8 model (e.g., nvidia/Llama-3.1-8B-Instruct-FP8) and CUDA graph enabled, the server fails with:
RuntimeError: query and key must have the same dtype
This problem of mismatching qk data type would probably persist across multiple quantized models.
Reproduction
root@56f67d238243:/workspace/sglang# python3 -m sglang.launch_server \
--model-path nvidia/Llama-3.1-8B-Instruct-FP8 \
--quantization modelopt_fp8 \
--port 30000
Environment
Python: 3.12.3 (main, Aug 14 2025, 17:47:21) [GCC 13.3.0]
CUDA available: True
GPU 0: NVIDIA H100 80GB HBM3
GPU 0 Compute Capability: 9.0
CUDA_HOME: /usr/local/cuda
NVCC: Cuda compilation tools, release 12.8, V12.8.93
CUDA Driver Version: 580.126.09
PyTorch: 2.9.1+cu128
sglang: 0.5.8
sgl_kernel: 0.3.21
flashinfer_python: 0.6.2
flashinfer_cubin: 0.6.2
flashinfer_jit_cache: Module Not Found
triton: 3.5.1
transformers: 4.57.1
torchao: 0.9.0
numpy: 2.1.2
aiohttp: 3.13.3
fastapi: 0.128.1
hf_transfer: 0.1.9
huggingface_hub: 0.36.1
interegular: 0.3.3
modelscope: 1.34.0
orjson: 3.11.7
outlines: 0.1.11
packaging: 25.0
psutil: 7.1.0
pydantic: 2.12.5
python-multipart: 0.0.22
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: 0.77.1
litellm: Module Not Found
decord2: 3.0.0
NVIDIA Topology:
GPU0 NIC0 NIC1 NIC2 NIC3 NIC4 NIC5 NIC6 NIC7 NIC8 CPU Affinity NUMA Affinity GPU NUMA ID
GPU0 X PXB PIX NODE NODE SYS SYS SYS SYS SYS 0-15 0 N/A
NIC0 PXB X PXB NODE NODE SYS SYS SYS SYS SYS
NIC1 PIX PXB X NODE NODE SYS SYS SYS SYS SYS
NIC2 NODE NODE NODE X PXB SYS SYS SYS SYS SYS
NIC3 NODE NODE NODE PXB X SYS SYS SYS SYS SYS
NIC4 SYS SYS SYS SYS SYS X PXB NODE NODE NODE
NIC5 SYS SYS SYS SYS SYS PXB X NODE NODE NODE
NIC6 SYS SYS SYS SYS SYS NODE NODE X PXB NODE
NIC7 SYS SYS SYS SYS SYS NODE NODE PXB X NODE
NIC8 SYS SYS SYS SYS SYS NODE NODE NODE NODE 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_1
NIC2: mlx5_2
NIC3: mlx5_3
NIC4: mlx5_4
NIC5: mlx5_5
NIC6: mlx5_6
NIC7: mlx5_7
NIC8: mlx5_8
ulimit soft: 4096
Checklist
Describe the bug
When running sglang.launch_server with a ModelOpt FP8 model (e.g., nvidia/Llama-3.1-8B-Instruct-FP8) and CUDA graph enabled, the server fails with:
This problem of mismatching qk data type would probably persist across multiple quantized models.
Reproduction
Environment