there is a conflict between cuda and megengine

We have met the problem when we train FINet, do you have any idea? Thanks.

Traceback (most recent call last):
File "/home/hongchang/.conda/envs/finet/lib/python3.8/multiprocessing/process.py", line 315, in _bootstrap
self.run()
File "/home/hongchang/.conda/envs/finet/lib/python3.8/multiprocessing/process.py", line 108, in run
self._target(*self._args, **self._kwargs)
File "/home/hongchang/.conda/envs/finet/lib/python3.8/site-packages/megengine/distributed/launcher.py", line 43, in _run_wrapped
_check_device_initialized(device_type, dev)
File "/home/hongchang/.conda/envs/finet/lib/python3.8/site-packages/megengine/distributed/helper.py", line 195, in _check_device_initialized
raise RuntimeError(errmsg)
RuntimeError: The cuda env is set before the forked thread starts. Please do not use any cuda function or variable before forking.
Process Process-2:

Start by reproducing FINet training with the reported Python 3.8 environment and the traceback. Inspect megengine/distributed/launcher.py and megengine/distributed/helper.py at the reported checks; done means the CUDA initialization error is explained and a verified resolution or required environment details are documented.

Written by the indexing model from the issue text.