Installation guide for CUDA 11 with latest version.

[for-just-we]

The installation method: pip install "sglang[all]>=0.4.6.post2" is for CUDA 12. When I try to install for CUDA 11 with command:

pip install torch==2.6.0 torchvision==0.21.0 torchaudio==2.6.0 --index-url https://download.pytorch.org/whl/cu118
pip install cuda-bindings==11.8.6
pip install cuda-python==11.8.5
pip install cupy-cuda11x
pip install flashinfer-python==0.2.5 --index-url https://flashinfer.ai/whl/cu118/torch2.6
pip install sgl-kernel==0.1.1 -i https://docs.sglang.ai/whl/cu118
pip install "sglang[all]>=0.4.6.post2"

Then run command: python -m sglang.launch_server. And have following error trace.

Traceback (most recent call last):
  File "<frozen runpy>", line 198, in _run_module_as_main
  File "<frozen runpy>", line 88, in _run_code
  File "/home/cbj/programming/python/anaconda3/envs/sglang/lib/python3.11/site-packages/sglang/launch_server.py", line 6, in <module>
    from sglang.srt.entrypoints.http_server import launch_server
  File "/home/cbj/programming/python/anaconda3/envs/sglang/lib/python3.11/site-packages/sglang/srt/entrypoints/http_server.py", line 46, in <module>
    from sglang.srt.entrypoints.engine import _launch_subprocesses
  File "/home/cbj/programming/python/anaconda3/envs/sglang/lib/python3.11/site-packages/sglang/srt/entrypoints/engine.py", line 42, in <module>
    from sglang.srt.managers.data_parallel_controller import (
  File "/home/cbj/programming/python/anaconda3/envs/sglang/lib/python3.11/site-packages/sglang/srt/managers/data_parallel_controller.py", line 28, in <module>
    from sglang.srt.managers.io_struct import (
  File "/home/cbj/programming/python/anaconda3/envs/sglang/lib/python3.11/site-packages/sglang/srt/managers/io_struct.py", line 31, in <module>
    from sglang.srt.managers.schedule_batch import BaseFinishReason
  File "/home/cbj/programming/python/anaconda3/envs/sglang/lib/python3.11/site-packages/sglang/srt/managers/schedule_batch.py", line 47, in <module>
    from sglang.srt.configs.model_config import ModelConfig
  File "/home/cbj/programming/python/anaconda3/envs/sglang/lib/python3.11/site-packages/sglang/srt/configs/model_config.py", line 26, in <module>
    from sglang.srt.layers.quantization import QUANTIZATION_METHODS
  File "/home/cbj/programming/python/anaconda3/envs/sglang/lib/python3.11/site-packages/sglang/srt/layers/quantization/__init__.py", line 54, in <module>
    from sglang.srt.layers.quantization.awq import AWQConfig
  File "/home/cbj/programming/python/anaconda3/envs/sglang/lib/python3.11/site-packages/sglang/srt/layers/quantization/awq.py", line 18, in <module>
    from sgl_kernel import awq_dequantize
  File "/home/cbj/programming/python/anaconda3/envs/sglang/lib/python3.11/site-packages/sgl_kernel/__init__.py", line 13, in <module>
    from sgl_kernel import common_ops
ImportError: /home/cbj/programming/python/anaconda3/envs/sglang/lib/python3.11/site-packages/sgl_kernel/common_ops.abi3.so: undefined symbol: _Z24ApplyTokenBitmaskInplaceN2at6TensorES0_St8optionalIS0_E

It seems the common_ops.abi3.so in sgl_kernel-0.1.1-cp39-abi3-manylinux2014_x86_64.whl does not contain definition of function: ApplyTokenBitmaskInplace .

So may I ask if:

• 1. There could be guidance for CUDA-11 with the latest version, because it's not easy to update CUDA on a company or university server.
• 2. Could the problem of the issue in common_ops.abi3.so be fixed?