Out of memory on 6GB Nvidia 2060

[Chris-R-R]

I have been able to train and infer on this computer previously but I am running out of memory inferring a particular wav file (I have successfully processed longer wav files previously). Are the duration and sample rate of the wav file affecting memory usage?
I am using default parameters.

C:\Mangio-RVC-v23.7.0>runtime\python.exe infer-web.py --pycmd runtime\python.exe --port 7897
Found GPU NVIDIA GeForce RTX 2060
Set fp16_run to true in 32k.json
Set fp16_run to true in 40k.json
Set fp16_run to true in 48k.json
Use Language: en_US
Running on local URL: http://0.0.0.0:7897

To create a public link, set share=True in launch().
loading weights/Benny.pth
gin_channels: 256 self.spk_embed_dim: 109

loading rmvpe model
Traceback (most recent call last):
File "C:\Mangio-RVC-v23.7.0\infer-web.py", line 316, in vc_single
audio_opt = vc.pipeline(
File "C:\Mangio-RVC-v23.7.0\vc_infer_pipeline.py", line 564, in pipeline
self.vc(
File "C:\Mangio-RVC-v23.7.0\vc_infer_pipeline.py", line 450, in vc
(net_g.infer(feats, p_len, pitch, pitchf, sid)[0][0, 0])
File "C:\Mangio-RVC-v23.7.0\lib\infer_pack\models.py", line 752, in infer
m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
File "C:\Mangio-RVC-v23.7.0\runtime\lib\site-packages\torch\nn\modules\module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "C:\Mangio-RVC-v23.7.0\lib\infer_pack\models.py", line 104, in forward
x = self.encoder(x * x_mask, x_mask)
File "C:\Mangio-RVC-v23.7.0\runtime\lib\site-packages\torch\nn\modules\module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "C:\Mangio-RVC-v23.7.0\lib\infer_pack\attentions.py", line 65, in forward
y = self.attn_layers[i](x, x, attn_mask)
File "C:\Mangio-RVC-v23.7.0\runtime\lib\site-packages\torch\nn\modules\module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "C:\Mangio-RVC-v23.7.0\lib\infer_pack\attentions.py", line 221, in forward
x, self.attn = self.attention(q, k, v, mask=attn_mask)
File "C:\Mangio-RVC-v23.7.0\lib\infer_pack\attentions.py", line 265, in attention
relative_weights = self._absolute_position_to_relative_position(p_attn)
File "C:\Mangio-RVC-v23.7.0\lib\infer_pack\attentions.py", line 346, in _absolute_position_to_relative_position
x_flat = F.pad(x_flat, commons.convert_pad_shape([[0, 0], [0, 0], [length, 0]]))
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 528.00 MiB (GPU 0; 6.00 GiB total capacity; 3.75 GiB already allocated; 0 bytes free; 4.81 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

Traceback (most recent call last):
File "C:\Mangio-RVC-v23.7.0\runtime\lib\site-packages\gradio\routes.py", line 437, in run_predict
output = await app.get_blocks().process_api(
File "C:\Mangio-RVC-v23.7.0\runtime\lib\site-packages\gradio\blocks.py", line 1349, in process_api
data = self.postprocess_data(fn_index, result["prediction"], state)
File "C:\Mangio-RVC-v23.7.0\runtime\lib\site-packages\gradio\blocks.py", line 1283, in postprocess_data
prediction_value = block.postprocess(prediction_value)
File "C:\Mangio-RVC-v23.7.0\runtime\lib\site-packages\gradio\components.py", line 2586, in postprocess
file_path = self.audio_to_temp_file(
File "C:\Mangio-RVC-v23.7.0\runtime\lib\site-packages\gradio\components.py", line 360, in audio_to_temp_file
temp_dir = Path(dir) / self.hash_bytes(data.tobytes())

[Chris-R-R]

After reading some of the other bug reports here I was able to get it working by setting the following parameters in config.py:
x_pad = 1 x_query = 5 x_center = 30 x_max = 32

Are these parameters explained anywhere?
The calculation of GPU memory confused me a little, why is 0.4 added to self.gpu_mem?
Maybe it would be better to subtract something here instead to take into account gpu memory used by other programs and have a safety margin?