[Feature] How to support PyTorch 2.9.1? NVIDIA NVIDIA_DGX_Spark NVIDIA Tegra NVIDIA GB10

[srchen1987]

Checklist

• 1. If the issue you raised is not a feature but a question, please raise a discussion at https://github.com/sgl-project/sglang/discussions/new/choose Otherwise, it will be closed.
• 2. Please use English, otherwise it will be closed.

Motivation

I was used DGX Spark to deploy sglang and run qwen3-32B. But I find some error.

My envs:
`
python3 -m sglang.check_env
/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/torch/cuda/init.py:283: UserWarning:
Found GPU0 NVIDIA GB10 which is of cuda capability 12.1.
Minimum and Maximum cuda capability supported by this version of PyTorch is
(8.0) - (12.0)

warnings.warn(
Python: 3.12.12 | packaged by Anaconda, Inc. | (main, Oct 21 2025, 20:05:09) [GCC 11.2.0]
CUDA available: True
GPU 0: NVIDIA GB10
GPU 0 Compute Capability: 12.1
CUDA_HOME: /usr/local/cuda
NVCC: Cuda compilation tools, release 13.0, V13.0.88
CUDA Driver Version: 580.95.05
PyTorch: 2.9.1+cu130
sglang: 0.5.5.post1
sgl_kernel: 0.3.17
flashinfer_python: 0.5.0
flashinfer_cubin: 0.5.0
flashinfer_jit_cache: Module Not Found
triton: 3.5.1
transformers: 4.57.1
torchao: 0.9.0
numpy: 2.3.4
aiohttp: 3.13.2
fastapi: 0.121.1
hf_transfer: 0.1.9
huggingface_hub: 0.36.0
interegular: 0.3.3
modelscope: 1.31.0
orjson: 3.11.4
outlines: 0.1.11
packaging: 25.0
psutil: 7.1.3
pydantic: 2.12.4
python-multipart: 0.0.20
pyzmq: 27.1.0
uvicorn: 0.38.0
uvloop: 0.21.0
vllm: Module Not Found
xgrammar: 0.1.25
openai: 2.6.1
tiktoken: 0.12.0
anthropic: 0.72.0
litellm: Module Not Found
decord2: 2.0.0
NVIDIA Topology:
GPU0 NIC0 NIC1 NIC2 NIC3 CPU Affinity NUMA Affinity GPU NUMA ID
GPU0 X NODE NODE NODE NODE 0-19 0 N/A
NIC0 NODE X PIX NODE NODE
NIC1 NODE PIX X NODE NODE
NIC2 NODE NODE NODE X PIX
NIC3 NODE NODE NODE PIX 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: rocep1s0f0
NIC1: rocep1s0f1
NIC2: roceP2p1s0f0
NIC3: roceP2p1s0f1

ulimit soft: 1024

`

run command:
python -m sglang.launch_server --model-path models/Qwen3-32B --host 0.0.0.0 --port 30000 --tp-size 1

error:

/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/torch/cuda/init.py:283: UserWarning:
Found GPU0 NVIDIA GB10 which is of cuda capability 12.1.
Minimum and Maximum cuda capability supported by this version of PyTorch is
(8.0) - (12.0)

warnings.warn(
[2025-11-13 17:45:45] WARNING common.py:1538: Failed to get GPU memory capacity from nvidia-smi, falling back to torch.cuda.mem_get_info().
Traceback (most recent call last):
File "", line 198, in _run_module_as_main
File "", line 88, in _run_code
File "/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sglang/launch_server.py", line 24, in
server_args = prepare_server_args(sys.argv[1:])
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sglang/srt/server_args.py", line 4008, in prepare_server_args
return ServerArgs.from_cli_args(raw_args)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sglang/srt/server_args.py", line 3616, in from_cli_args
return cls(**{attr: getattr(args, attr) for attr in attrs})
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "", line 275, in init
File "/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sglang/srt/server_args.py", line 588, in post_init
self._handle_gpu_memory_settings(gpu_mem)
File "/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sglang/srt/server_args.py", line 834, in _handle_gpu_memory_settings
model_config = self.get_model_config()
^^^^^^^^^^^^^^^^^^^^^^^
File "/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sglang/srt/server_args.py", line 3637, in get_model_config
from sglang.srt.configs.model_config import ModelConfig
File "/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sglang/srt/configs/model_config.py", line 26, in
from sglang.srt.layers.quantization import QUANTIZATION_METHODS
File "/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sglang/srt/layers/quantization/init.py", line 19, in
from sglang.srt.layers.quantization.auto_round import AutoRoundConfig
File "/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sglang/srt/layers/quantization/auto_round.py", line 12, in
from sglang.srt.layers.quantization.utils import get_scalar_types
File "/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sglang/srt/layers/quantization/utils.py", line 13, in
from sglang.srt.layers.quantization.fp8_kernel import scaled_fp8_quant
File "/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sglang/srt/layers/quantization/fp8_kernel.py", line 46, in
from sgl_kernel import sgl_per_tensor_quant_fp8, sgl_per_token_quant_fp8
File "/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sgl_kernel/init.py", line 5, in
common_ops = _load_architecture_specific_ops()
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sgl_kernel/load_utils.py", line 188, in _load_architecture_specific_ops
raise ImportError(error_msg)
ImportError:
[sgl_kernel] CRITICAL: Could not load any common_ops library!

Attempted locations:

1. Architecture-specific pattern: /home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sgl_kernel/sm100/common_ops.* - found files: ['/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sgl_kernel/sm100/common_ops.abi3.so']
2. Fallback pattern: /home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sgl_kernel/common_ops.* - found files: []
3. Standard Python import: common_ops - failed

GPU Info:

• Compute capability: 121
• Expected variant: SM121 (precise math for compatibility)

Please ensure sgl_kernel is properly installed with:
pip install --upgrade sgl_kernel

Error details from previous import attempts:

• ImportError: libnvrtc.so.12: cannot open shared object file: No such file or directory
• ModuleNotFoundError: No module named 'common_ops'

Related resources

I found https://pytorch.org/get-started/locally support cuda13.0.

nvidia user:

[Fridge003]

Working on this:
#12969

[johnnynunez]

Checklist

• 1. If the issue you raised is not a feature but a question, please raise a discussion at https://github.com/sgl-project/sglang/discussions/new/choose Otherwise, it will be closed.[x] 2. Please use English, otherwise it will be closed.

Motivation

I was used DGX Spark to deploy sglang and run qwen3-32B. But I find some error.

My envs: ` python3 -m sglang.check_env /home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/torch/cuda/init.py:283: UserWarning: Found GPU0 NVIDIA GB10 which is of cuda capability 12.1. Minimum and Maximum cuda capability supported by this version of PyTorch is (8.0) - (12.0)

warnings.warn( Python: 3.12.12 | packaged by Anaconda, Inc. | (main, Oct 21 2025, 20:05:09) [GCC 11.2.0] CUDA available: True GPU 0: NVIDIA GB10 GPU 0 Compute Capability: 12.1 CUDA_HOME: /usr/local/cuda NVCC: Cuda compilation tools, release 13.0, V13.0.88 CUDA Driver Version: 580.95.05 PyTorch: 2.9.1+cu130 sglang: 0.5.5.post1 sgl_kernel: 0.3.17 flashinfer_python: 0.5.0 flashinfer_cubin: 0.5.0 flashinfer_jit_cache: Module Not Found triton: 3.5.1 transformers: 4.57.1 torchao: 0.9.0 numpy: 2.3.4 aiohttp: 3.13.2 fastapi: 0.121.1 hf_transfer: 0.1.9 huggingface_hub: 0.36.0 interegular: 0.3.3 modelscope: 1.31.0 orjson: 3.11.4 outlines: 0.1.11 packaging: 25.0 psutil: 7.1.3 pydantic: 2.12.4 python-multipart: 0.0.20 pyzmq: 27.1.0 uvicorn: 0.38.0 uvloop: 0.21.0 vllm: Module Not Found xgrammar: 0.1.25 openai: 2.6.1 tiktoken: 0.12.0 anthropic: 0.72.0 litellm: Module Not Found decord2: 2.0.0 NVIDIA Topology: GPU0 NIC0 NIC1 NIC2 NIC3 CPU Affinity NUMA Affinity GPU NUMA ID GPU0 X NODE NODE NODE NODE 0-19 0 N/A NIC0 NODE X PIX NODE NODE NIC1 NODE PIX X NODE NODE NIC2 NODE NODE NODE X PIX NIC3 NODE NODE NODE PIX 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: rocep1s0f0 NIC1: rocep1s0f1 NIC2: roceP2p1s0f0 NIC3: roceP2p1s0f1

ulimit soft: 1024

`

run command: python -m sglang.launch_server --model-path models/Qwen3-32B --host 0.0.0.0 --port 30000 --tp-size 1

error:

/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/torch/cuda/init.py:283: UserWarning: Found GPU0 NVIDIA GB10 which is of cuda capability 12.1. Minimum and Maximum cuda capability supported by this version of PyTorch is (8.0) - (12.0)

warnings.warn( [2025-11-13 17:45:45] WARNING common.py:1538: Failed to get GPU memory capacity from nvidia-smi, falling back to torch.cuda.mem_get_info(). Traceback (most recent call last): File "", line 198, in _run_module_as_main File "", line 88, in _run_code File "/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sglang/launch_server.py", line 24, in server_args = prepare_server_args(sys.argv[1:]) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sglang/srt/server_args.py", line 4008, in prepare_server_args return ServerArgs.from_cli_args(raw_args) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sglang/srt/server_args.py", line 3616, in from_cli_args return cls(**{attr: getattr(args, attr) for attr in attrs}) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "", line 275, in init File "/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sglang/srt/server_args.py", line 588, in post_init self._handle_gpu_memory_settings(gpu_mem) File "/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sglang/srt/server_args.py", line 834, in _handle_gpu_memory_settings model_config = self.get_model_config() ^^^^^^^^^^^^^^^^^^^^^^^ File "/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sglang/srt/server_args.py", line 3637, in get_model_config from sglang.srt.configs.model_config import ModelConfig File "/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sglang/srt/configs/model_config.py", line 26, in from sglang.srt.layers.quantization import QUANTIZATION_METHODS File "/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sglang/srt/layers/quantization/init.py", line 19, in from sglang.srt.layers.quantization.auto_round import AutoRoundConfig File "/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sglang/srt/layers/quantization/auto_round.py", line 12, in from sglang.srt.layers.quantization.utils import get_scalar_types File "/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sglang/srt/layers/quantization/utils.py", line 13, in from sglang.srt.layers.quantization.fp8_kernel import scaled_fp8_quant File "/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sglang/srt/layers/quantization/fp8_kernel.py", line 46, in from sgl_kernel import sgl_per_tensor_quant_fp8, sgl_per_token_quant_fp8 File "/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sgl_kernel/init.py", line 5, in common_ops = _load_architecture_specific_ops() ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sgl_kernel/load_utils.py", line 188, in _load_architecture_specific_ops raise ImportError(error_msg) ImportError: [sgl_kernel] CRITICAL: Could not load any common_ops library!

Attempted locations:

1. Architecture-specific pattern: /home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sgl_kernel/sm100/common_ops.* - found files: ['/home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sgl_kernel/sm100/common_ops.abi3.so']
2. Fallback pattern: /home/srchen/.conda/envs/sglang_env/lib/python3.12/site-packages/sgl_kernel/common_ops.* - found files: []
3. Standard Python import: common_ops - failed

GPU Info:

• Compute capability: 121
• Expected variant: SM121 (precise math for compatibility)

Please ensure sgl_kernel is properly installed with: pip install --upgrade sgl_kernel

Error details from previous import attempts:

• ImportError: libnvrtc.so.12: cannot open shared object file: No such file or directory
• ModuleNotFoundError: No module named 'common_ops'

Related resources

I found https://pytorch.org/get-started/locally support cuda13.0.

nvidia user:

This is only a warning that it is deleted on next upcoming pytorch version.
Anyways, @Fridge003 how is it going the PR?

[srchen1987]

@johnnynunez Can you provide an environment that runs in a non Docker environment?

Another issue is that my dgx spark was upgraded and damaged. I have restored the system dgx-spark-recopy-image-1.91.51-1. tar.gz according to the official documentation. However, this system is much older than my version and lacks many software I wrote a crawler that captured resources greater than version 1.91.51-1 and returned 404 queries, which were executed tens of thousands of times Did you not update and restore the system's image?

[johnnynunez]

@johnnynunez Can you provide an environment that runs in a non Docker environment?

Another issue is that my dgx spark was upgraded and damaged. I have restored the system dgx-spark-recopy-image-1.91.51-1. tar.gz according to the official documentation. However, this system is much older than my version and lacks many software I wrote a crawler that captured resources greater than version 1.91.51-1 and returned 404 queries, which were executed tens of thousands of times Did you not update and restore the system's image?

in theory, we launched recently new update. is it your problem?

Run SGLang Thor & Spark

1. Install uv

curl -LsSf https://astral.sh/uv/install.sh | sh

2. Create environment

uv venv .sglang --python 3.12
source .sglang/bin/activate

3. Install Pytorch

uv pip install torch torchvision torchaudio triton --index-url https://download.pytorch.org/whl/cu130

2. Export variables

export TORCH_CUDA_ARCH_LIST=12.1a
export TRITON_PTXAS_PATH=/usr/local/cuda/bin/ptxas
export PATH=/usr/local/cuda/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH

3. Install SGLang

pip install --upgrade pip
uv pip install sglang
uv pip install sgl-kernel --prerelease=allow --index-url https://docs.sglang.ai/whl/cu130/sgl-kernel/

5. Clean memory

sudo sysctl -w vm.drop_caches=3

6. Run

python3 -m sglang.launch_server --model-path <output_path> --tp 1

[srchen1987]

@johnnynunez Thank you for your answer！ I really couldn't find the new version of the recovery image... the last version is 1.91.51-1! It‘s very old！

[johnnynunez]

@johnnynunez Thank you for your answer！ I really couldn't find the new version of the recovery image... the last version is 1.91.51-1! It‘s very old！

no worries, recovery it and the do

sudo apt update & sudo apt full-upgrade

[srchen1987]

@johnnynunez
I executed according to the document you wrote.
python3 -m sglang.launch_server --model-path models/Qwen3-32B --tp 1 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/srchen/llm/.sglang/lib/python3.12/site-packages/sglang/launch_server.py", line 24, in <module> server_args = prepare_server_args(sys.argv[1:]) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/srchen/llm/.sglang/lib/python3.12/site-packages/sglang/srt/server_args.py", line 4106, in prepare_server_args return ServerArgs.from_cli_args(raw_args) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/srchen/llm/.sglang/lib/python3.12/site-packages/sglang/srt/server_args.py", line 3707, in from_cli_args return cls(**{attr: getattr(args, attr) for attr in attrs}) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "<string>", line 281, in __init__ File "/home/srchen/llm/.sglang/lib/python3.12/site-packages/sglang/srt/server_args.py", line 590, in __post_init__ self._handle_missing_default_values() File "/home/srchen/llm/.sglang/lib/python3.12/site-packages/sglang/srt/server_args.py", line 675, in _handle_missing_default_values self.device = get_device() ^^^^^^^^^^^^ File "/home/srchen/llm/.sglang/lib/python3.12/site-packages/sglang/srt/utils/common.py", line 1819, in get_device raise RuntimeError("No accelerator (CUDA, XPU, HPU) is available.") RuntimeError: No accelerator (CUDA, XPU, HPU) is available.

I run this python code:

python3 test_pytorch.py

import torch print(torch.cuda.is_available()) print(torch.__version__)
result :

False
2.8.0+cpu

So, I uninstall pytorch and reinstall pytorch(gpu)

pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu130

run python3 test_pytorch.py again

result:

True
2.10.0.dev20251118+cu130

But run python3 -m sglang.launch_server --model-path models/Qwen3-32B --tp 1

I have returned to the starting point again, If you have dgx-spark you can test it.

error message

/home/srchen/llm/.sglang/lib/python3.12/site-packages/torch/cuda/init.py:435: UserWarning:
Found GPU0 NVIDIA GB10 which is of cuda capability 12.1.
Minimum and Maximum cuda capability supported by this version of PyTorch is
(8.0) - (12.0)

queued_call()
[2025-11-20 19:10:05] WARNING common.py:1568: Failed to get GPU memory capacity from nvidia-smi, falling back to torch.cuda.mem_get_info().
Traceback (most recent call last):
File "", line 198, in _run_module_as_main
File "", line 88, in _run_code
File "/home/srchen/llm/.sglang/lib/python3.12/site-packages/sglang/launch_server.py", line 24, in
server_args = prepare_server_args(sys.argv[1:])
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/srchen/llm/.sglang/lib/python3.12/site-packages/sglang/srt/server_args.py", line 4106, in prepare_server_args
return ServerArgs.from_cli_args(raw_args)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/srchen/llm/.sglang/lib/python3.12/site-packages/sglang/srt/server_args.py", line 3707, in from_cli_args
return cls(**{attr: getattr(args, attr) for attr in attrs})
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "", line 281, in init
File "/home/srchen/llm/.sglang/lib/python3.12/site-packages/sglang/srt/server_args.py", line 596, in post_init
self._handle_gpu_memory_settings(gpu_mem)
File "/home/srchen/llm/.sglang/lib/python3.12/site-packages/sglang/srt/server_args.py", line 845, in _handle_gpu_memory_settings
model_config = self.get_model_config()
^^^^^^^^^^^^^^^^^^^^^^^
File "/home/srchen/llm/.sglang/lib/python3.12/site-packages/sglang/srt/server_args.py", line 3728, in get_model_config
from sglang.srt.configs.model_config import ModelConfig
File "/home/srchen/llm/.sglang/lib/python3.12/site-packages/sglang/srt/configs/model_config.py", line 26, in
from sglang.srt.layers.quantization import QUANTIZATION_METHODS
File "/home/srchen/llm/.sglang/lib/python3.12/site-packages/sglang/srt/layers/quantization/init.py", line 19, in
from sglang.srt.layers.quantization.auto_round import AutoRoundConfig
File "/home/srchen/llm/.sglang/lib/python3.12/site-packages/sglang/srt/layers/quantization/auto_round.py", line 12, in
from sglang.srt.layers.quantization.utils import get_scalar_types
File "/home/srchen/llm/.sglang/lib/python3.12/site-packages/sglang/srt/layers/quantization/utils.py", line 13, in
from sglang.srt.layers.quantization.fp8_kernel import scaled_fp8_quant
File "/home/srchen/llm/.sglang/lib/python3.12/site-packages/sglang/srt/layers/quantization/fp8_kernel.py", line 46, in
from sgl_kernel import sgl_per_tensor_quant_fp8, sgl_per_token_quant_fp8
File "/home/srchen/llm/.sglang/lib/python3.12/site-packages/sgl_kernel/init.py", line 5, in
common_ops = _load_architecture_specific_ops()
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/srchen/llm/.sglang/lib/python3.12/site-packages/sgl_kernel/load_utils.py", line 188, in _load_architecture_specific_ops
raise ImportError(error_msg)
ImportError:
[sgl_kernel] CRITICAL: Could not load any common_ops library!

Attempted locations:

1. Architecture-specific pattern: /home/srchen/llm/.sglang/lib/python3.12/site-packages/sgl_kernel/sm100/common_ops.* - found files: ['/home/srchen/llm/.sglang/lib/python3.12/site-packages/sgl_kernel/sm100/common_ops.abi3.so']
2. Fallback pattern: /home/srchen/llm/.sglang/lib/python3.12/site-packages/sgl_kernel/common_ops.* - found files: []
3. Standard Python import: common_ops - failed

GPU Info:

• Compute capability: 121
• Expected variant: SM121 (precise math for compatibility)

Please ensure sgl_kernel is properly installed with:
pip install --upgrade sgl_kernel

Error details from previous import attempts:

• ImportError: libnvrtc.so.12: cannot open shared object file: No such file or directory
• ModuleNotFoundError: No module named 'common_ops'

[github-actions]

This issue has been automatically closed due to inactivity. Please feel free to reopen it if needed.