[Bug] Windows CLI: 2.3.0 crashes when working in GPU mode with RTX3060 laptop 6GB VRAM

- [Y] I didn't find a similar issue already open.
- [Y] I read the documentation (README AND Wiki)
- [Y] I have installed FFMpeg
- [Y] My problem is related to Spleeter only, not a derivative product (such as Webapplication, or GUI provided by others)

## Description

I use spleeter 2.3.0 in
windows CLI mode working on GPU then it crashes.(CPU 5900HS; GPU RTX3060 laptop 6GB VRAM),
and I discoverrd that the occupation of VRAM by the python process increased continuously up to 5.6GB before crash happened

## Step to reproduce

1. Installed cuda(11.1 11.2 11.4) and proper version of cudnn, pip install spleeter
2. Run in Windows CLI window:
D:\python>spleeter separate -p spleeter:2stems -o output spleeter/audio_example.mp3
3. Got error below:
2021-09-26 15:03:29.181055: F tensorflow/stream_executor/cuda/cuda_fft.cc:439] failed to initialize batched cufft plan with customized allocator: Failed to make cuFFT batched plan.
## Output
2021-09-26 15:03:29.181055: F tensorflow/stream_executor/cuda/cuda_fft.cc:439] failed to initialize batched cufft plan with customized allocator: Failed to make cuFFT batched plan.

```bash
D:\python>spleeter separate -p spleeter:2stems -o output spleeter/audio_example.mp3
2021-09-26 15:03:29.181055: F tensorflow/stream_executor/cuda/cuda_fft.cc:439] failed to initialize batched cufft plan with customized allocator: Failed to make cuFFT batched plan.
```

## Environment

| | |
| ----------------- | ------------------------------- |
| OS | Windows 10 |
| Installation type | pip |
| RAM available | ram 32GB vram 6GB|
| Hardware spec | CPU 5900HS with 32GB memory; GPU RTX3060 laptop with 6GB VRAM |

## Additional context

NONE