GPT-SoVITS
GPT-SoVITS copied to clipboard
RVC-Boss
SoVITS训练报错ZeroDivisionError
GPT训练正常。但SoVITS训练报错：
"python" GPT_SoVITS/s2_train.py --config "TEMP/tmp_s2.json"
INFO:meigui:{'train': {'log_interval': 100, 'eval_interval': 500, 'seed': 1234, 'epochs': 10, 'learning_rate': 0.0001, 'betas': [0.8, 0.99], 'eps': 1e-09, 'batch_size': 1, 'fp16_run': True, 'lr_decay': 0.999875, 'segment_size': 20480, 'init_lr_ratio': 1, 'warmup_epochs': 0, 'c_mel': 45, 'c_kl': 1.0, 'text_low_lr_rate': 0.4, 'pretrained_s2G': 'GPT_SoVITS/pretrained_models/s2G488k.pth', 'pretrained_s2D': 'GPT_SoVITS/pretrained_models/s2D488k.pth', 'if_save_latest': True, 'if_save_every_weights': True, 'save_every_epoch': 5, 'gpu_numbers': '0'}, 'data': {'max_wav_value': 32768.0, 'sampling_rate': 32000, 'filter_length': 2048, 'hop_length': 640, 'win_length': 2048, 'n_mel_channels': 128, 'mel_fmin': 0.0, 'mel_fmax': None, 'add_blank': True, 'n_speakers': 300, 'cleaned_text': True, 'exp_dir': 'logs/meigui'}, 'model': {'inter_channels': 192, 'hidden_channels': 192, 'filter_channels': 768, 'n_heads': 2, 'n_layers': 6, 'kernel_size': 3, 'p_dropout': 0.1, 'resblock': '1', 'resblock_kernel_sizes': [3, 7, 11], 'resblock_dilation_sizes': [[1, 3, 5], [1, 3, 5], [1, 3, 5]], 'upsample_rates': [10, 8, 2, 2, 2], 'upsample_initial_channel': 512, 'upsample_kernel_sizes': [16, 16, 8, 2, 2], 'n_layers_q': 3, 'use_spectral_norm': False, 'gin_channels': 512, 'semantic_frame_rate': '25hz', 'freeze_quantizer': True}, 's2_ckpt_dir': 'logs/meigui', 'content_module': 'cnhubert', 'save_weight_dir': 'SoVITS_weights', 'name': 'meigui', 'pretrain': None, 'resume_step': None}
INFO:torch.distributed.distributed_c10d:Added key: store_based_barrier_key:1 to store for rank: 0
INFO:torch.distributed.distributed_c10d:Rank 0: Completed store-based barrier for key:store_based_barrier_key:1 with 1 nodes.
phoneme_data_len: 3
wav_data_len: 99
100%|████████████████████████████████████████| 99/99 [00:00<00:00, 24775.42it/s]
skipped_phone: 0 , skipped_dur: 0
total left: 99
ssl_proj.weight not requires_grad
ssl_proj.bias not requires_grad
INFO:meigui:loaded pretrained GPT_SoVITS/pretrained_models/s2G488k.pth
<All keys matched successfully>
INFO:meigui:loaded pretrained GPT_SoVITS/pretrained_models/s2D488k.pth
<All keys matched successfully>
/root/miniconda3/lib/python3.10/site-packages/torch/optim/lr_scheduler.py:139: UserWarning: Detected call of `lr_scheduler.step()` before `optimizer.step()`. In PyTorch 1.1.0 and later, you should call them in the opposite order: `optimizer.step()` before `lr_scheduler.step()`. Failure to do this will result in PyTorch skipping the first value of the learning rate schedule. See more details at https://pytorch.org/docs/stable/optim.html#how-to-adjust-learning-rate
warnings.warn("Detected call of `lr_scheduler.step()` before `optimizer.step()`. "
Exception ignored in: <function _MultiProcessingDataLoaderIter.__del__ at 0x7f5c840ca710>
Traceback (most recent call last):
File "/root/miniconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 1478, in __del__
self._shutdown_workers()
File "/root/miniconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 1409, in _shutdown_workers
if not self._shutdown:
AttributeError: '_MultiProcessingDataLoaderIter' object has no attribute '_shutdown'
Traceback (most recent call last):
File "/root/autodl-tmp/workdir/GPT-SoVITS/GPT_SoVITS/s2_train.py", line 402, in <module>
main()
File "/root/autodl-tmp/workdir/GPT-SoVITS/GPT_SoVITS/s2_train.py", line 53, in main
mp.spawn(run, nprocs=n_gpus, args=(n_gpus, hps,))
File "/root/miniconda3/lib/python3.10/site-packages/torch/multiprocessing/spawn.py", line 239, in spawn
return start_processes(fn, args, nprocs, join, daemon, start_method='spawn')
File "/root/miniconda3/lib/python3.10/site-packages/torch/multiprocessing/spawn.py", line 197, in start_processes
while not context.join():
File "/root/miniconda3/lib/python3.10/site-packages/torch/multiprocessing/spawn.py", line 160, in join
raise ProcessRaisedException(msg, error_index, failed_process.pid)
torch.multiprocessing.spawn.ProcessRaisedException:
-- Process 0 terminated with the following error:
Traceback (most recent call last):
File "/root/miniconda3/lib/python3.10/site-packages/torch/multiprocessing/spawn.py", line 69, in _wrap
fn(i, *args)
File "/root/autodl-tmp/workdir/GPT-SoVITS/GPT_SoVITS/s2_train.py", line 172, in run
train_and_evaluate(rank, epoch, hps, [net_g, net_d], [optim_g, optim_d], [scheduler_g, scheduler_d], scaler,
File "/root/autodl-tmp/workdir/GPT-SoVITS/GPT_SoVITS/s2_train.py", line 195, in train_and_evaluate
for batch_idx, (ssl, ssl_lengths, spec, spec_lengths, y, y_lengths, text, text_lengths) in tqdm(enumerate(train_loader)):
File "/root/miniconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 436, in __iter__
self._iterator = self._get_iterator()
File "/root/miniconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 388, in _get_iterator
return _MultiProcessingDataLoaderIter(self)
File "/root/miniconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 994, in __init__
super().__init__(loader)
File "/root/miniconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 603, in __init__
self._sampler_iter = iter(self._index_sampler)
File "/root/autodl-tmp/workdir/GPT-SoVITS/GPT_SoVITS/module/data_utils.py", line 293, in __iter__
ids_bucket = ids_bucket + ids_bucket * (rem // len_bucket) + ids_bucket[:(rem % len_bucket)]
ZeroDivisionError: integer division or modulo by zero
这个bug是 6-name2semantic.tsv 没有正确生成，应该是多进程同步问题，等待作者更新。一个手动解决方法是，在”微调训练“tab页上，手动3个步骤，再一键三连，多试试几次。
我是SoVits训练正常，GPT训练报错ZeroDivisionError：
Traceback (most recent call last):
File "H:\speech\GPT-SoVITS\GPT_SoVITS\s1_train.py", line 171, in
同样的问题开启sovits训练和gpt训练就报错。
File "F:\GPT-SoVITS_\GPT-SoVITS\GPT_SoVITS\s2_train.py", line 402, in
-- Process 0 terminated with the following error: Traceback (most recent call last): File "F:\GPT-SoVITS_\GPT-SoVITS\runtime\lib\site-packages\torch\multiprocessing\spawn.py", line 69, in wrap fn(i, *args) File "F:\GPT-SoVITS\GPT-SoVITS\GPT_SoVITS\s2_train.py", line 69, in run train_dataset = TextAudioSpeakerLoader(hps.data)######## File "F:\GPT-SoVITS_\GPT-SoVITS\GPT_SoVITS\module\data_utils.py", line 55, in init for _ in range(max(2, int(min_num / leng))): ZeroDivisionError: division by zero
Fixed | 通过在计算前添加零持续时间的修改检查来修复该问题。
导航：C:\GPT-SoVITS-beta\GPT-SoVITS\GPT_SoVITS\module 编辑：data_utils Replace All Code / 替换所有代码：
import time
import logging
import os
import random
import traceback
import numpy as np
import torch
import torch.utils.data
from tqdm import tqdm
from module import commons
from module.mel_processing import spectrogram_torch
from text import cleaned_text_to_sequence
from utils import load_wav_to_torch, load_filepaths_and_text
import torch.nn.functional as F
from functools import lru_cache
import requests
from scipy.io import wavfile
from io import BytesIO
from my_utils import load_audio
class TextAudioSpeakerLoader(torch.utils.data.Dataset):
"""
1) loads audio, speaker_id, text pairs
2) normalizes text and converts them to sequences of integers
3) computes spectrograms from audio files.
"""
def __init__(self, hparams, val=False):
exp_dir = hparams.exp_dir
self.path2 = "%s/2-name2text.txt" % exp_dir
self.path4 = "%s/4-cnhubert" % exp_dir
self.path5 = "%s/5-wav32k" % exp_dir
assert os.path.exists(self.path2)
assert os.path.exists(self.path4)
assert os.path.exists(self.path5)
names4 = set([name[:-3] for name in list(os.listdir(self.path4))]) # 去除.pt后缀
names5 = set(os.listdir(self.path5))
self.phoneme_data = {}
with open(self.path2, "r", encoding="utf8") as f:
lines = f.read().strip("\n").split("\n")
for line in lines:
tmp = line.split("\t")
if (len(tmp) != 4):
continue
self.phoneme_data[tmp[0]] = [tmp[1]]
self.audiopaths_sid_text = list(set(self.phoneme_data) & names4 & names5)
tmp = self.audiopaths_sid_text
leng = len(tmp)
min_num = 100
if (leng < min_num):
self.audiopaths_sid_text = []
for _ in range(max(2, int(min_num / leng))):
self.audiopaths_sid_text += tmp
self.max_wav_value = hparams.max_wav_value
self.sampling_rate = hparams.sampling_rate
self.filter_length = hparams.filter_length
self.hop_length = hparams.hop_length
self.win_length = hparams.win_length
self.sampling_rate = hparams.sampling_rate
self.val = val
random.seed(1234)
random.shuffle(self.audiopaths_sid_text)
print("phoneme_data_len:", len(self.phoneme_data.keys()))
print("wav_data_len:", len(self.audiopaths_sid_text))
audiopaths_sid_text_new = []
lengths = []
skipped_phone = 0
skipped_dur = 0
for audiopath in tqdm(self.audiopaths_sid_text):
try:
phoneme = self.phoneme_data[audiopath][0]
phoneme = phoneme.split(' ')
phoneme_ids = cleaned_text_to_sequence(phoneme)
except Exception:
print(f"{audiopath} not in self.phoneme_data !")
skipped_phone += 1
continue
size = os.path.getsize("%s/%s" % (self.path5, audiopath))
duration = size / self.sampling_rate / 2
if duration == 0:
print(f"Zero duration for {audiopath}, skipping...")
skipped_dur += 1
continue
if 54 > duration > 0.6 or self.val:
audiopaths_sid_text_new.append([audiopath, phoneme_ids])
lengths.append(size // (2 * self.hop_length))
else:
skipped_dur += 1
continue
print("skipped_phone: ", skipped_phone, ", skipped_dur: ", skipped_dur)
print("total left: ", len(audiopaths_sid_text_new))
assert len(audiopaths_sid_text_new) > 1 # 至少能凑够batch size，这里todo
self.audiopaths_sid_text = audiopaths_sid_text_new
self.lengths = lengths
def get_audio_text_speaker_pair(self, audiopath_sid_text):
audiopath, phoneme_ids = audiopath_sid_text
text = torch.FloatTensor(phoneme_ids)
try:
spec, wav = self.get_audio("%s/%s" % (self.path5, audiopath))
with torch.no_grad():
ssl = torch.load("%s/%s.pt" % (self.path4, audiopath), map_location="cpu")
if (ssl.shape[-1] != spec.shape[-1]):
typee = ssl.dtype
ssl = F.pad(ssl.float(), (0, 1), mode="replicate").to(typee)
ssl.requires_grad = False
except:
traceback.print_exc()
spec = torch.zeros(1025, 100)
wav = torch.zeros(1, 100 * self.hop_length)
ssl = torch.zeros(1, 768, 100)
text = text[-1:]
print("load audio or ssl error!!!!!!", audiopath)
return (ssl, spec, wav, text)
def get_audio(self, filename):
audio_array = load_audio(filename, self.sampling_rate) # load_audio的方法是已经归一化到-1~1之间的，不用再/32768
audio = torch.FloatTensor(audio_array) # /32768
audio_norm = audio
audio_norm = audio_norm.unsqueeze(0)
spec = spectrogram_torch(audio_norm, self.filter_length, self.sampling_rate, self.hop_length, self.win_length,
center=False)
spec = torch.squeeze(spec, 0)
return spec, audio_norm
def get_sid(self, sid):
sid = torch.LongTensor([int(sid)])
return sid
def __getitem__(self, index):
# with torch.no_grad():
return self.get_audio_text_speaker_pair(self.audiopaths_sid_text[index])
def __len__(self):
return len(self.audiopaths_sid_text)
def random_slice(self, ssl, wav, mel):
assert abs(ssl.shape[-1] - wav.shape[-1] // self.hop_length) < 3, (
"first", ssl.shape, wav.shape)
len_mel = mel.shape[1]
if self.val:
reference_mel = mel[:, :len_mel // 3]
return reference_mel, ssl, wav, mel
dir = random.randint(0, 1)
sep_point = random.randint(int(len_mel // 3), int(len_mel // 3 * 2))
if dir == 0:
reference_mel = mel[:, :sep_point]
ssl = ssl[:, :, sep_point:]
wav2 = wav[:, sep_point * self.hop_length:]
mel = mel[:, sep_point:]
else:
reference_mel = mel[:, sep_point:]
ssl = ssl[:, :, :sep_point]
wav2 = wav[:, :sep_point * self.hop_length]
mel = mel[:, :sep_point]
assert abs(ssl.shape[-1] - wav2.shape[-1] // self.hop_length) < 3, (
ssl.shape, wav.shape, wav2.shape, mel.shape, sep_point, self.hop_length, sep_point * self.hop_length, dir)
return reference_mel, ssl, wav2, mel
class TextAudioSpeakerCollate():
""" Zero-pads model inputs and targets
"""
def __init__(self, return_ids=False):
self.return_ids = return_ids
def __call__(self, batch):
"""Collate's training batch from normalized text, audio and speaker identities
PARAMS
------
batch: [text_normalized, spec_normalized, wav_normalized, sid]
"""
# Right zero-pad all one-hot text sequences to max input length
_, ids_sorted_decreasing = torch.sort(
torch.LongTensor([x[1].size(1) for x in batch]),
dim=0, descending=True)
max_ssl_len = max([x[0].size(2) for x in batch])
max_ssl_len = int(2 * ((max_ssl_len // 2) + 1))
max_spec_len = max([x[1].size(1) for x in batch])
max_spec_len = int(2 * ((max_spec_len // 2) + 1))
max_wav_len = max([x[2].size(1) for x in batch])
max_text_len = max([x[3].size(0) for x in batch])
ssl_lengths = torch.LongTensor(len(batch))
spec_lengths = torch.LongTensor(len(batch))
wav_lengths = torch.LongTensor(len(batch))
text_lengths = torch.LongTensor(len(batch))
spec_padded = torch.FloatTensor(len(batch), batch[0][1].size(0), max_spec_len)
wav_padded = torch.FloatTensor(len(batch), 1, max_wav_len)
ssl_padded = torch.FloatTensor(len(batch), batch[0][0].size(1), max_ssl_len)
text_padded = torch.LongTensor(len(batch), max_text_len)
spec_padded.zero_()
wav_padded.zero_()
ssl_padded.zero_()
text_padded.zero_()
for i in range(len(ids_sorted_decreasing)):
row = batch[ids_sorted_decreasing[i]]
ssl = row[0]
ssl_padded[i, :, :ssl.size(2)] = ssl[0, :, :]
ssl_lengths[i] = ssl.size(2)
spec = row[1]
spec_padded[i, :, :spec.size(1)] = spec
spec_lengths[i] = spec.size(1)
wav = row[2]
wav_padded[i, :, :wav.size(1)] = wav
wav_lengths[i] = wav.size(1)
text = row[3]
text_padded[i, :text.size(0)] = text
text_lengths[i] = text.size(0)
return ssl_padded, ssl_lengths, spec_padded, spec_lengths, wav_padded, wav_lengths, text_padded, text_lengths
class DistributedBucketSampler(torch.utils.data.distributed.DistributedSampler):
"""
Maintain similar input lengths in a batch.
Length groups are specified by boundaries.
Ex) boundaries = [b1, b2, b3] -> any batch is included either {x | b1 < length(x) <=b2} or {x | b2 < length(x) <= b3}.
It removes samples which are not included in the boundaries.
Ex) boundaries = [b1, b2, b3] -> any x s.t. length(x) <= b1 or length(x) > b3 are discarded.
"""
def __init__(self, dataset, batch_size, boundaries, num_replicas=None, rank=None, shuffle=True):
super().__init__(dataset, num_replicas=num_replicas, rank=rank, shuffle=shuffle)
self.lengths = dataset.lengths
self.batch_size = batch_size
self.boundaries = boundaries
self.buckets, self.num_samples_per_bucket = self._create_buckets()
self.total_size = sum(self.num_samples_per_bucket)
self.num_samples = self.total_size // self.num_replicas
def _create_buckets(self):
buckets = [[] for _ in range(len(self.boundaries) - 1)]
for i in range(len(self.lengths)):
length = self.lengths[i]
idx_bucket = self._bisect(length)
if idx_bucket != -1:
buckets[idx_bucket].append(i)
i = len(buckets) - 1
while i >= 0:
if len(buckets[i]) == 0:
buckets.pop(i)
self.boundaries.pop(i + 1)
i -= 1
num_samples_per_bucket = []
for i in range(len(buckets)):
len_bucket = len(buckets[i])
total_batch_size = self.num_replicas * self.batch_size
rem = (total_batch_size - (len_bucket % total_batch_size)) % total_batch_size
num_samples_per_bucket.append(len_bucket + rem)
return buckets, num_samples_per_bucket
def __iter__(self):
g = torch.Generator()
g.manual_seed(self.epoch)
indices = []
if self.shuffle:
for bucket in self.buckets:
indices.append(torch.randperm(len(bucket), generator=g).tolist())
else:
for bucket in self.buckets:
indices.append(list(range(len(bucket))))
batches = []
for i in range(len(self.buckets)):
bucket = self.buckets[i]
len_bucket = len(bucket)
ids_bucket = indices[i]
num_samples_bucket = self.num_samples_per_bucket[i]
rem = num_samples_bucket - len_bucket
ids_bucket = ids_bucket + ids_bucket * (rem // len_bucket) + ids_bucket[:(rem % len_bucket)]
ids_bucket = ids_bucket[self.rank::self.num_replicas]
for j in range(len(ids_bucket) // self.batch_size):
batch = [bucket[idx] for idx in ids_bucket[j * self.batch_size:(j + 1) * self.batch_size]]
batches.append(batch)
if self.shuffle:
batch_ids = torch.randperm(len(batches), generator=g).tolist()
batches = [batches[i] for i in batch_ids]
self.batches = batches
assert len(self.batches) * self.batch_size == self.num_samples
return iter(self.batches)
def _bisect(self, x, lo=0, hi=None):
if hi is None:
hi = len(self.boundaries) - 1
if hi > lo:
mid = (hi + lo) // 2
if self.boundaries[mid] < x and x <= self.boundaries[mid + 1]:
return mid
elif x <= self.boundaries[mid]:
return self._bisect(x, lo, mid)
else:
return self._bisect(x, mid + 1, hi)
else:
return -1
def __len__(self):
return self.num_samples // self.batch_size
def _bisect(self, x, lo=0, hi=None):
if hi is None:
hi = len(self.boundaries) - 1
if hi > lo:
mid = (hi + lo) // 2
if self.boundaries[mid] < x and x <= self.boundaries[mid + 1]:
return mid
elif x <= self.boundaries[mid]:
return self._bisect(x, lo, mid)
else:
return self._bisect(x, mid + 1, hi)
else:
return -1
def __len__(self):
return self.num_samples // self.batch_size
i = len(buckets) - 1 while i >= 0: if len(buckets[i]) == 0: buckets.pop(i) self.boundaries.pop(i + 1) i -= 1 这段内容是什么意思？也复制到 data_utils 吗？？求解答
i = len(buckets) - 1 while i >= 0: if len(buckets[i]) == 0: buckets.pop(i) self.boundaries.pop(i + 1) i -= 1 这段内容是什么意思？也复制到 data_utils 吗？？求解答
对不起，我只是指出了添加的代码。当 len_bucket 为零时，尝试计算余数或商时会发生错误。我修改了 DistributedBucketSampler 类中的 _create_buckets 方法，确保循环运行直到 i 变得小于 0，从而防止除以零的可能性。
只需粘贴并替换data_utils.py模块下的所有现有代码即可进行更新。
GPT训练正常。但SoVITS训练报错：
"python" GPT_SoVITS/s2_train.py --config "TEMP/tmp_s2.json" INFO:meigui:{'train': {'log_interval': 100, 'eval_interval': 500, 'seed': 1234, 'epochs': 10, 'learning_rate': 0.0001, 'betas': [0.8, 0.99], 'eps': 1e-09, 'batch_size': 1, 'fp16_run': True, 'lr_decay': 0.999875, 'segment_size': 20480, 'init_lr_ratio': 1, 'warmup_epochs': 0, 'c_mel': 45, 'c_kl': 1.0, 'text_low_lr_rate': 0.4, 'pretrained_s2G': 'GPT_SoVITS/pretrained_models/s2G488k.pth', 'pretrained_s2D': 'GPT_SoVITS/pretrained_models/s2D488k.pth', 'if_save_latest': True, 'if_save_every_weights': True, 'save_every_epoch': 5, 'gpu_numbers': '0'}, 'data': {'max_wav_value': 32768.0, 'sampling_rate': 32000, 'filter_length': 2048, 'hop_length': 640, 'win_length': 2048, 'n_mel_channels': 128, 'mel_fmin': 0.0, 'mel_fmax': None, 'add_blank': True, 'n_speakers': 300, 'cleaned_text': True, 'exp_dir': 'logs/meigui'}, 'model': {'inter_channels': 192, 'hidden_channels': 192, 'filter_channels': 768, 'n_heads': 2, 'n_layers': 6, 'kernel_size': 3, 'p_dropout': 0.1, 'resblock': '1', 'resblock_kernel_sizes': [3, 7, 11], 'resblock_dilation_sizes': [[1, 3, 5], [1, 3, 5], [1, 3, 5]], 'upsample_rates': [10, 8, 2, 2, 2], 'upsample_initial_channel': 512, 'upsample_kernel_sizes': [16, 16, 8, 2, 2], 'n_layers_q': 3, 'use_spectral_norm': False, 'gin_channels': 512, 'semantic_frame_rate': '25hz', 'freeze_quantizer': True}, 's2_ckpt_dir': 'logs/meigui', 'content_module': 'cnhubert', 'save_weight_dir': 'SoVITS_weights', 'name': 'meigui', 'pretrain': None, 'resume_step': None} INFO:torch.distributed.distributed_c10d:Added key: store_based_barrier_key:1 to store for rank: 0 INFO:torch.distributed.distributed_c10d:Rank 0: Completed store-based barrier for key:store_based_barrier_key:1 with 1 nodes. phoneme_data_len: 3 wav_data_len: 99 100%|████████████████████████████████████████| 99/99 [00:00<00:00, 24775.42it/s] skipped_phone: 0 , skipped_dur: 0 total left: 99 ssl_proj.weight not requires_grad ssl_proj.bias not requires_grad INFO:meigui:loaded pretrained GPT_SoVITS/pretrained_models/s2G488k.pth <All keys matched successfully> INFO:meigui:loaded pretrained GPT_SoVITS/pretrained_models/s2D488k.pth <All keys matched successfully> /root/miniconda3/lib/python3.10/site-packages/torch/optim/lr_scheduler.py:139: UserWarning: Detected call of `lr_scheduler.step()` before `optimizer.step()`. In PyTorch 1.1.0 and later, you should call them in the opposite order: `optimizer.step()` before `lr_scheduler.step()`. Failure to do this will result in PyTorch skipping the first value of the learning rate schedule. See more details at https://pytorch.org/docs/stable/optim.html#how-to-adjust-learning-rate warnings.warn("Detected call of `lr_scheduler.step()` before `optimizer.step()`. " Exception ignored in: <function _MultiProcessingDataLoaderIter.__del__ at 0x7f5c840ca710> Traceback (most recent call last): File "/root/miniconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 1478, in __del__ self._shutdown_workers() File "/root/miniconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 1409, in _shutdown_workers if not self._shutdown: AttributeError: '_MultiProcessingDataLoaderIter' object has no attribute '_shutdown' Traceback (most recent call last): File "/root/autodl-tmp/workdir/GPT-SoVITS/GPT_SoVITS/s2_train.py", line 402, in <module> main() File "/root/autodl-tmp/workdir/GPT-SoVITS/GPT_SoVITS/s2_train.py", line 53, in main mp.spawn(run, nprocs=n_gpus, args=(n_gpus, hps,)) File "/root/miniconda3/lib/python3.10/site-packages/torch/multiprocessing/spawn.py", line 239, in spawn return start_processes(fn, args, nprocs, join, daemon, start_method='spawn') File "/root/miniconda3/lib/python3.10/site-packages/torch/multiprocessing/spawn.py", line 197, in start_processes while not context.join(): File "/root/miniconda3/lib/python3.10/site-packages/torch/multiprocessing/spawn.py", line 160, in join raise ProcessRaisedException(msg, error_index, failed_process.pid) torch.multiprocessing.spawn.ProcessRaisedException: -- Process 0 terminated with the following error: Traceback (most recent call last): File "/root/miniconda3/lib/python3.10/site-packages/torch/multiprocessing/spawn.py", line 69, in _wrap fn(i, *args) File "/root/autodl-tmp/workdir/GPT-SoVITS/GPT_SoVITS/s2_train.py", line 172, in run train_and_evaluate(rank, epoch, hps, [net_g, net_d], [optim_g, optim_d], [scheduler_g, scheduler_d], scaler, File "/root/autodl-tmp/workdir/GPT-SoVITS/GPT_SoVITS/s2_train.py", line 195, in train_and_evaluate for batch_idx, (ssl, ssl_lengths, spec, spec_lengths, y, y_lengths, text, text_lengths) in tqdm(enumerate(train_loader)): File "/root/miniconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 436, in __iter__ self._iterator = self._get_iterator() File "/root/miniconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 388, in _get_iterator return _MultiProcessingDataLoaderIter(self) File "/root/miniconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 994, in __init__ super().__init__(loader) File "/root/miniconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 603, in __init__ self._sampler_iter = iter(self._index_sampler) File "/root/autodl-tmp/workdir/GPT-SoVITS/GPT_SoVITS/module/data_utils.py", line 293, in __iter__ ids_bucket = ids_bucket + ids_bucket * (rem // len_bucket) + ids_bucket[:(rem % len_bucket)] ZeroDivisionError: integer division or modulo by zero
+1
Fixed | 通过在计算前添加零持续时间的修改检查来修复该问题。
导航：C:\GPT-SoVITS-beta\GPT-SoVITS\GPT_SoVITS\module 编辑：data_utils Replace All Code / 替换所有代码：
import time import logging import os import random import traceback import numpy as np import torch import torch.utils.data from tqdm import tqdm from module import commons from module.mel_processing import spectrogram_torch from text import cleaned_text_to_sequence from utils import load_wav_to_torch, load_filepaths_and_text import torch.nn.functional as F from functools import lru_cache import requests from scipy.io import wavfile from io import BytesIO from my_utils import load_audio class TextAudioSpeakerLoader(torch.utils.data.Dataset): """ 1) loads audio, speaker_id, text pairs 2) normalizes text and converts them to sequences of integers 3) computes spectrograms from audio files. """ def __init__(self, hparams, val=False): exp_dir = hparams.exp_dir self.path2 = "%s/2-name2text.txt" % exp_dir self.path4 = "%s/4-cnhubert" % exp_dir self.path5 = "%s/5-wav32k" % exp_dir assert os.path.exists(self.path2) assert os.path.exists(self.path4) assert os.path.exists(self.path5) names4 = set([name[:-3] for name in list(os.listdir(self.path4))]) # 去除.pt后缀 names5 = set(os.listdir(self.path5)) self.phoneme_data = {} with open(self.path2, "r", encoding="utf8") as f: lines = f.read().strip("\n").split("\n") for line in lines: tmp = line.split("\t") if (len(tmp) != 4): continue self.phoneme_data[tmp[0]] = [tmp[1]] self.audiopaths_sid_text = list(set(self.phoneme_data) & names4 & names5) tmp = self.audiopaths_sid_text leng = len(tmp) min_num = 100 if (leng < min_num): self.audiopaths_sid_text = [] for _ in range(max(2, int(min_num / leng))): self.audiopaths_sid_text += tmp self.max_wav_value = hparams.max_wav_value self.sampling_rate = hparams.sampling_rate self.filter_length = hparams.filter_length self.hop_length = hparams.hop_length self.win_length = hparams.win_length self.sampling_rate = hparams.sampling_rate self.val = val random.seed(1234) random.shuffle(self.audiopaths_sid_text) print("phoneme_data_len:", len(self.phoneme_data.keys())) print("wav_data_len:", len(self.audiopaths_sid_text)) audiopaths_sid_text_new = [] lengths = [] skipped_phone = 0 skipped_dur = 0 for audiopath in tqdm(self.audiopaths_sid_text): try: phoneme = self.phoneme_data[audiopath][0] phoneme = phoneme.split(' ') phoneme_ids = cleaned_text_to_sequence(phoneme) except Exception: print(f"{audiopath} not in self.phoneme_data !") skipped_phone += 1 continue size = os.path.getsize("%s/%s" % (self.path5, audiopath)) duration = size / self.sampling_rate / 2 if duration == 0: print(f"Zero duration for {audiopath}, skipping...") skipped_dur += 1 continue if 54 > duration > 0.6 or self.val: audiopaths_sid_text_new.append([audiopath, phoneme_ids]) lengths.append(size // (2 * self.hop_length)) else: skipped_dur += 1 continue print("skipped_phone: ", skipped_phone, ", skipped_dur: ", skipped_dur) print("total left: ", len(audiopaths_sid_text_new)) assert len(audiopaths_sid_text_new) > 1 # 至少能凑够batch size，这里todo self.audiopaths_sid_text = audiopaths_sid_text_new self.lengths = lengths def get_audio_text_speaker_pair(self, audiopath_sid_text): audiopath, phoneme_ids = audiopath_sid_text text = torch.FloatTensor(phoneme_ids) try: spec, wav = self.get_audio("%s/%s" % (self.path5, audiopath)) with torch.no_grad(): ssl = torch.load("%s/%s.pt" % (self.path4, audiopath), map_location="cpu") if (ssl.shape[-1] != spec.shape[-1]): typee = ssl.dtype ssl = F.pad(ssl.float(), (0, 1), mode="replicate").to(typee) ssl.requires_grad = False except: traceback.print_exc() spec = torch.zeros(1025, 100) wav = torch.zeros(1, 100 * self.hop_length) ssl = torch.zeros(1, 768, 100) text = text[-1:] print("load audio or ssl error!!!!!!", audiopath) return (ssl, spec, wav, text) def get_audio(self, filename): audio_array = load_audio(filename, self.sampling_rate) # load_audio的方法是已经归一化到-1~1之间的，不用再/32768 audio = torch.FloatTensor(audio_array) # /32768 audio_norm = audio audio_norm = audio_norm.unsqueeze(0) spec = spectrogram_torch(audio_norm, self.filter_length, self.sampling_rate, self.hop_length, self.win_length, center=False) spec = torch.squeeze(spec, 0) return spec, audio_norm def get_sid(self, sid): sid = torch.LongTensor([int(sid)]) return sid def __getitem__(self, index): # with torch.no_grad(): return self.get_audio_text_speaker_pair(self.audiopaths_sid_text[index]) def __len__(self): return len(self.audiopaths_sid_text) def random_slice(self, ssl, wav, mel): assert abs(ssl.shape[-1] - wav.shape[-1] // self.hop_length) < 3, ( "first", ssl.shape, wav.shape) len_mel = mel.shape[1] if self.val: reference_mel = mel[:, :len_mel // 3] return reference_mel, ssl, wav, mel dir = random.randint(0, 1) sep_point = random.randint(int(len_mel // 3), int(len_mel // 3 * 2)) if dir == 0: reference_mel = mel[:, :sep_point] ssl = ssl[:, :, sep_point:] wav2 = wav[:, sep_point * self.hop_length:] mel = mel[:, sep_point:] else: reference_mel = mel[:, sep_point:] ssl = ssl[:, :, :sep_point] wav2 = wav[:, :sep_point * self.hop_length] mel = mel[:, :sep_point] assert abs(ssl.shape[-1] - wav2.shape[-1] // self.hop_length) < 3, ( ssl.shape, wav.shape, wav2.shape, mel.shape, sep_point, self.hop_length, sep_point * self.hop_length, dir) return reference_mel, ssl, wav2, mel class TextAudioSpeakerCollate(): """ Zero-pads model inputs and targets """ def __init__(self, return_ids=False): self.return_ids = return_ids def __call__(self, batch): """Collate's training batch from normalized text, audio and speaker identities PARAMS ------ batch: [text_normalized, spec_normalized, wav_normalized, sid] """ # Right zero-pad all one-hot text sequences to max input length _, ids_sorted_decreasing = torch.sort( torch.LongTensor([x[1].size(1) for x in batch]), dim=0, descending=True) max_ssl_len = max([x[0].size(2) for x in batch]) max_ssl_len = int(2 * ((max_ssl_len // 2) + 1)) max_spec_len = max([x[1].size(1) for x in batch]) max_spec_len = int(2 * ((max_spec_len // 2) + 1)) max_wav_len = max([x[2].size(1) for x in batch]) max_text_len = max([x[3].size(0) for x in batch]) ssl_lengths = torch.LongTensor(len(batch)) spec_lengths = torch.LongTensor(len(batch)) wav_lengths = torch.LongTensor(len(batch)) text_lengths = torch.LongTensor(len(batch)) spec_padded = torch.FloatTensor(len(batch), batch[0][1].size(0), max_spec_len) wav_padded = torch.FloatTensor(len(batch), 1, max_wav_len) ssl_padded = torch.FloatTensor(len(batch), batch[0][0].size(1), max_ssl_len) text_padded = torch.LongTensor(len(batch), max_text_len) spec_padded.zero_() wav_padded.zero_() ssl_padded.zero_() text_padded.zero_() for i in range(len(ids_sorted_decreasing)): row = batch[ids_sorted_decreasing[i]] ssl = row[0] ssl_padded[i, :, :ssl.size(2)] = ssl[0, :, :] ssl_lengths[i] = ssl.size(2) spec = row[1] spec_padded[i, :, :spec.size(1)] = spec spec_lengths[i] = spec.size(1) wav = row[2] wav_padded[i, :, :wav.size(1)] = wav wav_lengths[i] = wav.size(1) text = row[3] text_padded[i, :text.size(0)] = text text_lengths[i] = text.size(0) return ssl_padded, ssl_lengths, spec_padded, spec_lengths, wav_padded, wav_lengths, text_padded, text_lengths class DistributedBucketSampler(torch.utils.data.distributed.DistributedSampler): """ Maintain similar input lengths in a batch. Length groups are specified by boundaries. Ex) boundaries = [b1, b2, b3] -> any batch is included either {x | b1 < length(x) <=b2} or {x | b2 < length(x) <= b3}. It removes samples which are not included in the boundaries. Ex) boundaries = [b1, b2, b3] -> any x s.t. length(x) <= b1 or length(x) > b3 are discarded. """ def __init__(self, dataset, batch_size, boundaries, num_replicas=None, rank=None, shuffle=True): super().__init__(dataset, num_replicas=num_replicas, rank=rank, shuffle=shuffle) self.lengths = dataset.lengths self.batch_size = batch_size self.boundaries = boundaries self.buckets, self.num_samples_per_bucket = self._create_buckets() self.total_size = sum(self.num_samples_per_bucket) self.num_samples = self.total_size // self.num_replicas def _create_buckets(self): buckets = [[] for _ in range(len(self.boundaries) - 1)] for i in range(len(self.lengths)): length = self.lengths[i] idx_bucket = self._bisect(length) if idx_bucket != -1: buckets[idx_bucket].append(i) i = len(buckets) - 1 while i >= 0: if len(buckets[i]) == 0: buckets.pop(i) self.boundaries.pop(i + 1) i -= 1 num_samples_per_bucket = [] for i in range(len(buckets)): len_bucket = len(buckets[i]) total_batch_size = self.num_replicas * self.batch_size rem = (total_batch_size - (len_bucket % total_batch_size)) % total_batch_size num_samples_per_bucket.append(len_bucket + rem) return buckets, num_samples_per_bucket def __iter__(self): g = torch.Generator() g.manual_seed(self.epoch) indices = [] if self.shuffle: for bucket in self.buckets: indices.append(torch.randperm(len(bucket), generator=g).tolist()) else: for bucket in self.buckets: indices.append(list(range(len(bucket)))) batches = [] for i in range(len(self.buckets)): bucket = self.buckets[i] len_bucket = len(bucket) ids_bucket = indices[i] num_samples_bucket = self.num_samples_per_bucket[i] rem = num_samples_bucket - len_bucket ids_bucket = ids_bucket + ids_bucket * (rem // len_bucket) + ids_bucket[:(rem % len_bucket)] ids_bucket = ids_bucket[self.rank::self.num_replicas] for j in range(len(ids_bucket) // self.batch_size): batch = [bucket[idx] for idx in ids_bucket[j * self.batch_size:(j + 1) * self.batch_size]] batches.append(batch) if self.shuffle: batch_ids = torch.randperm(len(batches), generator=g).tolist() batches = [batches[i] for i in batch_ids] self.batches = batches assert len(self.batches) * self.batch_size == self.num_samples return iter(self.batches) def _bisect(self, x, lo=0, hi=None): if hi is None: hi = len(self.boundaries) - 1 if hi > lo: mid = (hi + lo) // 2 if self.boundaries[mid] < x and x <= self.boundaries[mid + 1]: return mid elif x <= self.boundaries[mid]: return self._bisect(x, lo, mid) else: return self._bisect(x, mid + 1, hi) else: return -1 def __len__(self): return self.num_samples // self.batch_size
感谢大佬，但是把文件代码改完后报如下的错
Traceback (most recent call last):
File "D:\GPT-SOVITS\GPT_SoVITS\s2_train.py", line 600, in
-- Process 0 terminated with the following error: Traceback (most recent call last): File "C:\Users***\anaconda3\envs\ame\lib\site-packages\torch\multiprocessing\spawn.py", line 68, in _wrap fn(i, *args) File "D:\GPT-SOVITS\GPT_SoVITS\s2_train.py", line 85, in run train_dataset = TextAudioSpeakerLoader(hps.data) ######## File "D:\GPT-SOVITS\GPT_SoVITS\module\data_utils.py", line 34, in init assert os.path.exists(self.path2) AssertionError
GPT训练正常。但SoVITS训练报错：
"python" GPT_SoVITS/s2_train.py --config "TEMP/tmp_s2.json" INFO:meigui:{'train': {'log_interval': 100, 'eval_interval': 500, 'seed': 1234, 'epochs': 10, 'learning_rate': 0.0001, 'betas': [0.8, 0.99], 'eps': 1e-09, 'batch_size': 1, 'fp16_run': True, 'lr_decay': 0.999875, 'segment_size': 20480, 'init_lr_ratio': 1, 'warmup_epochs': 0, 'c_mel': 45, 'c_kl': 1.0, 'text_low_lr_rate': 0.4, 'pretrained_s2G': 'GPT_SoVITS/pretrained_models/s2G488k.pth', 'pretrained_s2D': 'GPT_SoVITS/pretrained_models/s2D488k.pth', 'if_save_latest': True, 'if_save_every_weights': True, 'save_every_epoch': 5, 'gpu_numbers': '0'}, 'data': {'max_wav_value': 32768.0, 'sampling_rate': 32000, 'filter_length': 2048, 'hop_length': 640, 'win_length': 2048, 'n_mel_channels': 128, 'mel_fmin': 0.0, 'mel_fmax': None, 'add_blank': True, 'n_speakers': 300, 'cleaned_text': True, 'exp_dir': 'logs/meigui'}, 'model': {'inter_channels': 192, 'hidden_channels': 192, 'filter_channels': 768, 'n_heads': 2, 'n_layers': 6, 'kernel_size': 3, 'p_dropout': 0.1, 'resblock': '1', 'resblock_kernel_sizes': [3, 7, 11], 'resblock_dilation_sizes': [[1, 3, 5], [1, 3, 5], [1, 3, 5]], 'upsample_rates': [10, 8, 2, 2, 2], 'upsample_initial_channel': 512, 'upsample_kernel_sizes': [16, 16, 8, 2, 2], 'n_layers_q': 3, 'use_spectral_norm': False, 'gin_channels': 512, 'semantic_frame_rate': '25hz', 'freeze_quantizer': True}, 's2_ckpt_dir': 'logs/meigui', 'content_module': 'cnhubert', 'save_weight_dir': 'SoVITS_weights', 'name': 'meigui', 'pretrain': None, 'resume_step': None} INFO:torch.distributed.distributed_c10d:Added key: store_based_barrier_key:1 to store for rank: 0 INFO:torch.distributed.distributed_c10d:Rank 0: Completed store-based barrier for key:store_based_barrier_key:1 with 1 nodes. phoneme_data_len: 3 wav_data_len: 99 100%|████████████████████████████████████████| 99/99 [00:00<00:00, 24775.42it/s] skipped_phone: 0 , skipped_dur: 0 total left: 99 ssl_proj.weight not requires_grad ssl_proj.bias not requires_grad INFO:meigui:loaded pretrained GPT_SoVITS/pretrained_models/s2G488k.pth <All keys matched successfully> INFO:meigui:loaded pretrained GPT_SoVITS/pretrained_models/s2D488k.pth <All keys matched successfully> /root/miniconda3/lib/python3.10/site-packages/torch/optim/lr_scheduler.py:139: UserWarning: Detected call of `lr_scheduler.step()` before `optimizer.step()`. In PyTorch 1.1.0 and later, you should call them in the opposite order: `optimizer.step()` before `lr_scheduler.step()`. Failure to do this will result in PyTorch skipping the first value of the learning rate schedule. See more details at https://pytorch.org/docs/stable/optim.html#how-to-adjust-learning-rate warnings.warn("Detected call of `lr_scheduler.step()` before `optimizer.step()`. " Exception ignored in: <function _MultiProcessingDataLoaderIter.__del__ at 0x7f5c840ca710> Traceback (most recent call last): File "/root/miniconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 1478, in __del__ self._shutdown_workers() File "/root/miniconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 1409, in _shutdown_workers if not self._shutdown: AttributeError: '_MultiProcessingDataLoaderIter' object has no attribute '_shutdown' Traceback (most recent call last): File "/root/autodl-tmp/workdir/GPT-SoVITS/GPT_SoVITS/s2_train.py", line 402, in <module> main() File "/root/autodl-tmp/workdir/GPT-SoVITS/GPT_SoVITS/s2_train.py", line 53, in main mp.spawn(run, nprocs=n_gpus, args=(n_gpus, hps,)) File "/root/miniconda3/lib/python3.10/site-packages/torch/multiprocessing/spawn.py", line 239, in spawn return start_processes(fn, args, nprocs, join, daemon, start_method='spawn') File "/root/miniconda3/lib/python3.10/site-packages/torch/multiprocessing/spawn.py", line 197, in start_processes while not context.join(): File "/root/miniconda3/lib/python3.10/site-packages/torch/multiprocessing/spawn.py", line 160, in join raise ProcessRaisedException(msg, error_index, failed_process.pid) torch.multiprocessing.spawn.ProcessRaisedException: -- Process 0 terminated with the following error: Traceback (most recent call last): File "/root/miniconda3/lib/python3.10/site-packages/torch/multiprocessing/spawn.py", line 69, in _wrap fn(i, *args) File "/root/autodl-tmp/workdir/GPT-SoVITS/GPT_SoVITS/s2_train.py", line 172, in run train_and_evaluate(rank, epoch, hps, [net_g, net_d], [optim_g, optim_d], [scheduler_g, scheduler_d], scaler, File "/root/autodl-tmp/workdir/GPT-SoVITS/GPT_SoVITS/s2_train.py", line 195, in train_and_evaluate for batch_idx, (ssl, ssl_lengths, spec, spec_lengths, y, y_lengths, text, text_lengths) in tqdm(enumerate(train_loader)): File "/root/miniconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 436, in __iter__ self._iterator = self._get_iterator() File "/root/miniconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 388, in _get_iterator return _MultiProcessingDataLoaderIter(self) File "/root/miniconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 994, in __init__ super().__init__(loader) File "/root/miniconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 603, in __init__ self._sampler_iter = iter(self._index_sampler) File "/root/autodl-tmp/workdir/GPT-SoVITS/GPT_SoVITS/module/data_utils.py", line 293, in __iter__ ids_bucket = ids_bucket + ids_bucket * (rem // len_bucket) + ids_bucket[:(rem % len_bucket)] ZeroDivisionError: integer division or modulo by zero
请问解决了吗？
解决了，是我当时还没用明白软件的原因。
| | Zack | | @.*** | ---- 回复的原邮件 ---- | 发件人 | @.> | | 发送日期 | 2024年3月13日 17:54 | | 收件人 | @.> | | 抄送人 | @.>, @.> | | 主题 | Re: [RVC-Boss/GPT-SoVITS] SoVITS训练报错ZeroDivisionError (Issue #79) |
GPT训练正常。但SoVITS训练报错：
"python" GPT_SoVITS/s2_train.py --config "TEMP/tmp_s2.json"
INFO:meigui:{'train': {'log_interval': 100, 'eval_interval': 500, 'seed': 1234, 'epochs': 10, 'learning_rate': 0.0001, 'betas': [0.8, 0.99], 'eps': 1e-09, 'batch_size': 1, 'fp16_run': True, 'lr_decay': 0.999875, 'segment_size': 20480, 'init_lr_ratio': 1, 'warmup_epochs': 0, 'c_mel': 45, 'c_kl': 1.0, 'text_low_lr_rate': 0.4, 'pretrained_s2G': 'GPT_SoVITS/pretrained_models/s2G488k.pth', 'pretrained_s2D': 'GPT_SoVITS/pretrained_models/s2D488k.pth', 'if_save_latest': True, 'if_save_every_weights': True, 'save_every_epoch': 5, 'gpu_numbers': '0'}, 'data': {'max_wav_value': 32768.0, 'sampling_rate': 32000, 'filter_length': 2048, 'hop_length': 640, 'win_length': 2048, 'n_mel_channels': 128, 'mel_fmin': 0.0, 'mel_fmax': None, 'add_blank': True, 'n_speakers': 300, 'cleaned_text': True, 'exp_dir': 'logs/meigui'}, 'model': {'inter_channels': 192, 'hidden_channels': 192, 'filter_channels': 768, 'n_heads': 2, 'n_layers': 6, 'kernel_size': 3, 'p_dropout': 0.1, 'resblock': '1', 'resblock_kernel_sizes': [3, 7, 11], 'resblock_dilation_sizes': [[1, 3, 5], [1, 3, 5], [1, 3, 5]], 'upsample_rates': [10, 8, 2, 2, 2], 'upsample_initial_channel': 512, 'upsample_kernel_sizes': [16, 16, 8, 2, 2], 'n_layers_q': 3, 'use_spectral_norm': False, 'gin_channels': 512, 'semantic_frame_rate': '25hz', 'freeze_quantizer': True}, 's2_ckpt_dir': 'logs/meigui', 'content_module': 'cnhubert', 'save_weight_dir': 'SoVITS_weights', 'name': 'meigui', 'pretrain': None, 'resume_step': None}
INFO:torch.distributed.distributed_c10d:Added key: store_based_barrier_key:1 to store for rank: 0
INFO:torch.distributed.distributed_c10d:Rank 0: Completed store-based barrier for key:store_based_barrier_key:1 with 1 nodes.
phoneme_data_len: 3
wav_data_len: 99
100%|████████████████████████████████████████| 99/99 [00:00<00:00, 24775.42it/s]
skipped_phone: 0 , skipped_dur: 0
total left: 99
ssl_proj.weight not requires_grad
ssl_proj.bias not requires_grad
INFO:meigui:loaded pretrained GPT_SoVITS/pretrained_models/s2G488k.pth
<All keys matched successfully>
INFO:meigui:loaded pretrained GPT_SoVITS/pretrained_models/s2D488k.pth
<All keys matched successfully>
/root/miniconda3/lib/python3.10/site-packages/torch/optim/lr_scheduler.py:139: UserWarning: Detected call of lr_scheduler.step() before optimizer.step(). In PyTorch 1.1.0 and later, you should call them in the opposite order: optimizer.step() before lr_scheduler.step(). Failure to do this will result in PyTorch skipping the first value of the learning rate schedule. See more details at https://pytorch.org/docs/stable/optim.html#how-to-adjust-learning-rate
warnings.warn("Detected call of lr_scheduler.step() before optimizer.step(). "
Exception ignored in: <function _MultiProcessingDataLoaderIter.del at 0x7f5c840ca710>
Traceback (most recent call last):
File "/root/miniconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 1478, in del
self._shutdown_workers()
File "/root/miniconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 1409, in _shutdown_workers
if not self._shutdown:
AttributeError: '_MultiProcessingDataLoaderIter' object has no attribute '_shutdown'
Traceback (most recent call last):
File "/root/autodl-tmp/workdir/GPT-SoVITS/GPT_SoVITS/s2_train.py", line 402, in
-- Process 0 terminated with the following error: Traceback (most recent call last): File "/root/miniconda3/lib/python3.10/site-packages/torch/multiprocessing/spawn.py", line 69, in _wrap fn(i, *args) File "/root/autodl-tmp/workdir/GPT-SoVITS/GPT_SoVITS/s2_train.py", line 172, in run train_and_evaluate(rank, epoch, hps, [net_g, net_d], [optim_g, optim_d], [scheduler_g, scheduler_d], scaler, File "/root/autodl-tmp/workdir/GPT-SoVITS/GPT_SoVITS/s2_train.py", line 195, in train_and_evaluate for batch_idx, (ssl, ssl_lengths, spec, spec_lengths, y, y_lengths, text, text_lengths) in tqdm(enumerate(train_loader)): File "/root/miniconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 436, in iter self._iterator = self._get_iterator() File "/root/miniconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 388, in _get_iterator return _MultiProcessingDataLoaderIter(self) File "/root/miniconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 994, in init super().init(loader) File "/root/miniconda3/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 603, in init self._sampler_iter = iter(self._index_sampler) File "/root/autodl-tmp/workdir/GPT-SoVITS/GPT_SoVITS/module/data_utils.py", line 293, in iter ids_bucket = ids_bucket + ids_bucket * (rem // len_bucket) + ids_bucket[:(rem % len_bucket)] ZeroDivisionError: integer division or modulo by zero
请问解决了吗？
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>