RuntimeError: CUDA out of memory. Tried to allocate 58.00 MiB (GPU 0; 6.00 GiB total capacity; 1.83 GiB already allocated; 2.49 GiB free; 2.02 GiB reserved in total by PyTorch)

{| Epoch: 1/1 (121/15311) | Loss: 1.719 | 0.73 steps/s | Step: 0k | }Traceback (most recent call last):
出现错误
RuntimeError: CUDA out of memory. Tried to allocate 58.00 MiB (GPU 0; 6.00 GiB total capacity; 1.83 GiB already allocated; 2.49 GiB free; 2.02 GiB reserved in total by PyTorch)
显示显存不足
已经修改### Tacotron Training
tts_schedule = [(2, 1e-3, 10_000, 8), # Progressive training schedule
(2, 5e-4, 15_000, 8), # (r, lr, step, batch_size)
(2, 2e-4, 20_000, 8), # (r, lr, step, batch_size)
(2, 1e-4, 30_000, 8), #
(2, 5e-5, 40_000, 8), #
(2, 1e-5, 60_000, 8), #
(2, 5e-6, 160_000, 8), # r = reduction factor (# of mel frames
(2, 3e-6, 320_000, 8), # synthesized for each decoder iteration)
(2, 1e-6, 640_000, 8)], # lr = learning rate

请问我要怎么调整才能继续训练合成器？感谢各位大佬

Start by reproducing the Tacotron Training run that reports the CUDA out-of-memory error, using the tts_schedule shown in the issue. Read the training configuration around that schedule and record which adjustment allows the run to continue without the same allocation failure.

Written by the indexing model from the issue text.