Checklist

• 1. I have searched related issues but cannot get the expected help.
• 2. The bug has not been fixed in the latest version.
• 3. Please note that if the bug-related issue you submitted lacks corresponding environment info and a minimal reproducible demo, it will be challenging for us to reproduce and resolve the issue, reducing the likelihood of receiving feedback.
• 4. If the issue you raised is not a bug but a question, please raise a discussion at https://github.com/sgl-project/sglang/discussions/new/choose Otherwise, it will be closed.
• 5. Please use English, otherwise it will be closed.

Describe the bug

When setting BatchPrefillWithRaggedKVCacheWrapper backend to "cutlass" in flashinfer backend, the test result for Llama-3.1-8B-Instruct is low:

Accuracy: 0.018
Invalid: 0.110
Latency: 45.625 s
Output throughput: 12136.862 token/s

Triton backend result with the same test:

Accuracy: 0.788
Invalid: 0.001
Latency: 16.626 s
Output throughput: 8002.240 token/s

According to @yzh119 , inserting a synchronization before run will resolve the issue. Since the overhead should be bypassed, further modification is needed.

Reproduction

Flashinfer built from source on latest main.
Set BatchPrefillWithRaggedKVCacheWrapper backend to "cutlass" in flashinfer_backend.py.

        self.prefill_wrapper_ragged = BatchPrefillWithRaggedKVCacheWrapper(
            self.workspace_buffer, "NHD", backend="cutlass"
        )

Run server with:
python3 -m sglang.launch_server --model meta-llama/Llama-3.1-8B-Instruct --trust-remote --attention-backend flashinfer
Run test with:
python3 benchmark/gsm8k/bench_sglang.py --num-shots 8 --num-questions 1319 --parallel 1319

Environment

Python: 3.12.3 (main, Feb 4 2025, 14:48:35) [GCC 13.3.0]
CUDA available: True
GPU 0,1,2,3,4,5,6,7: NVIDIA B200
GPU 0,1,2,3,4,5,6,7 Compute Capability: 10.0
CUDA_HOME: /usr/local/cuda
NVCC: Cuda compilation tools, release 12.9, V12.9.41
CUDA Driver Version: 570.148.08
PyTorch: 2.7.0+cu128
sglang: 0.4.6.post5
sgl_kernel: 0.1.5
flashinfer_python: 0.2.5
triton: 3.3.0
transformers: 4.52.3
torchao: 0.9.0
numpy: 2.1.2
aiohttp: 3.12.6
fastapi: 0.115.12
hf_transfer: 0.1.9
huggingface_hub: 0.32.3
interegular: 0.3.3
modelscope: 1.26.0
orjson: 3.10.18
outlines: 0.1.11
packaging: 25.0
psutil: 7.0.0
pydantic: 2.11.5
python-multipart: 0.0.20
pyzmq: 26.4.0
uvicorn: 0.34.3
uvloop: 0.21.0
vllm: Module Not Found
xgrammar: 0.1.19
openai: 1.83.0
tiktoken: 0.9.0
anthropic: Module Not Found
litellm: Module Not Found
decord: Module Not Found