RuntimeError when running the "Hello World" example

[le-monde-bleu]

I am trying to run the "Hello World" example with the command

neutrino -p dmat python 06-fused-attention.py // [\!code highlight]

and got a RuntimeError:

Traceback (most recent call last):
  File "/home/wxy/06-fused-attention.py", line 762, in <module>
    bench_flash_attention.run(save_path=".", print_data=True)
  File "/home/wxy/anaconda3/envs/cuda12/lib/python3.11/site-packages/triton/testing.py", line 392, in run
    result_dfs.append(self._run(bench, save_path, show_plots, print_data, **kwargs))
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/wxy/anaconda3/envs/cuda12/lib/python3.11/site-packages/triton/testing.py", line 339, in _run
    ret = self.fn(**x_args, **{bench.line_arg: y}, **bench.args, **kwrags)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/wxy/06-fused-attention.py", line 726, in bench_flash_attention
    q = torch.randn((BATCH, H, N_CTX, HEAD_DIM), dtype=dtype, device=device, requires_grad=True)
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: CUDA error: API call is not supported in the installed CUDA driver
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1
Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.

Strangely, when I directly ran the workload:

python 06-fused-attention.py

I got no error. I wonder what could be the problem?

Besides, I noticed that CUDA 11.7 is incompatible with PyTorch 2.3.1, which is required by the triton workload. So I managed to install CUDA 12.1 toolkits and successfully ran 06-fused-attention.py. But neutrino -p dmat python 06-fused-attention.py // [\!code highlight] failed me.