RuntimeError with Assertion failed: driver_result == cudaDriverEntryPointSuccess.

Describe the bug

when running examples/llama/train_llama3_8b_fp8.sh and building transformer layer of GPTModel:

RuntimeError: /TransformerEngine/transformer_engine/common/util/cuda_driver.cpp:42 in function get_symbol: Assertion failed: driver_result == cudaDriverEntryPointSuccess. Could not find CUDA driver entry point for cuCtxGetCurrent when instantiating TERowParallelLinear when instantiating SelfAttentio

Steps/Code to reproduce bug

1. conda create -n megatronlm python=3.12
2. pip install torch==2.6.0
3. pip install megatron-core
4. pip install --no-build-isolation transformer-engine[pytorch]
5. Pip install regex six PyYAML psutil pybind11

then just run examples/llama/train_llama3_8b_fp8.sh

Environment overview (please complete the following information)
ubuntu 22.04
python 3.12
torch 2.6.0+cu124
cuda 12.4(by nvcc)
cudnn 9.1.0
tranformer-engine 2.8.0

Device details
H100

Additional context

Add any other context about the problem here.

Start with examples/llama/train_llama3_8b_fp8.sh and inspect transformer_engine/common/util/cuda_driver.cpp at line 42, especially get_symbol and the cuCtxGetCurrent lookup. Reproduce the failure with the listed PyTorch, CUDA, H100, and Transformer Engine versions, then trace the TERowParallelLinear and SelfAttention instantiation path. Done means the example proceeds without this assertion under the reported environment.