尝试将二郎神1.3B换为 3.9B过程中遇到的问题

[pengpengzi]

下载3.9B模型放入相应的模型文件夹，
修改text/chinese_bert中

import sys
import torch
from config import config
from transformers import MegatronBertModel, BertTokenizer

LOCAL_PATH = "./bert/Erlangshen-MegatronBert-3.9B-Chinese"
# LOCAL_PATH = "./bert/Erlangshen-MegatronBert-1.3B-Chinese"
tokenizer = BertTokenizer.from_pretrained(LOCAL_PATH)

然后运行 python bert_gen.py python clap_gen.py都顺利运行
但是运行train_ma.py报错张量不匹配

(py38_vits) [www@localhost Bert-VITS2-Medlive]$ python bert_gen.py
  0%|                                                                                                                                                                         | 0/676 [00:00<?, ?it/s]/home/www/.conda/envs/py38_vits/lib/python3.8/site-packages/torch/_utils.py:776: UserWarning: TypedStorage is deprecated. It will be removed in the future and UntypedStorage will be the only storage class. This should only matter to you if you are using storages directly.  To access UntypedStorage directly, use tensor.untyped_storage() instead of tensor.storage()
  return self.fget.__get__(instance, owner)()
100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 676/676 [00:42<00:00, 15.79it/s]
bert生成完毕!, 共有676个bert.pt生成!
(py38_vits) [www@localhost Bert-VITS2-Medlive]$ python clap_gen.py
100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 676/676 [00:00<00:00, 59611.25it/s]
clap生成完毕!, 共有676个emo.pt生成!
(py38_vits) [www@localhost Bert-VITS2-Medlive]$ python train_ms.py
加载config中的配置localhost
加载config中的配置10086
加载config中的配置1
加载config中的配置0
加载config中的配置0
加载环境变量
MASTER_ADDR: localhost,
MASTER_PORT: 10086,
WORLD_SIZE: 1,
RANK: 0,
LOCAL_RANK: 0
03-01 01:37:26 INFO     | data_utils.py:64 | Init dataset...
100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 672/672 [00:00<00:00, 74341.20it/s]
03-01 01:37:26 INFO     | data_utils.py:79 | skipped: 1, total: 672
03-01 01:37:26 INFO     | data_utils.py:64 | Init dataset...
100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 4/4 [00:00<00:00, 52924.97it/s]
03-01 01:37:26 INFO     | data_utils.py:79 | skipped: 0, total: 4
Using noise scaled MAS for VITS2
Using duration discriminator for VITS2
Use existed model, skip downloading.
INFO:models:Loaded checkpoint 'Data/hxj/models/DUR_0.pth' (iteration 1)
INFO:models:Loaded checkpoint 'Data/hxj/models/WD_0.pth' (iteration 1)
INFO:models:Loaded checkpoint 'Data/hxj/models/G_0.pth' (iteration 1)
INFO:models:Loaded checkpoint 'Data/hxj/models/D_0.pth' (iteration 1)
******************检测到模型存在，epoch为 1，gloabl step为 0*********************
  0%|                                                                                                                                                                          | 0/46 [00:00<?, ?it/s]
Traceback (most recent call last):
  File "train_ms.py", line 840, in <module>
    run()
  File "train_ms.py", line 373, in run
    train_and_evaluate(
  File "train_ms.py", line 437, in train_and_evaluate
    for batch_idx, (
  File "/home/www/.conda/envs/py38_vits/lib/python3.8/site-packages/tqdm/std.py", line 1182, in __iter__
    for obj in iterable:
  File "/home/www/.conda/envs/py38_vits/lib/python3.8/site-packages/torch/utils/data/dataloader.py", line 633, in __next__
    data = self._next_data()
  File "/home/www/.conda/envs/py38_vits/lib/python3.8/site-packages/torch/utils/data/dataloader.py", line 1345, in _next_data
    return self._process_data(data)
  File "/home/www/.conda/envs/py38_vits/lib/python3.8/site-packages/torch/utils/data/dataloader.py", line 1371, in _process_data
    data.reraise()
  File "/home/www/.conda/envs/py38_vits/lib/python3.8/site-packages/torch/_utils.py", line 644, in reraise
    raise exception
RuntimeError: Caught RuntimeError in DataLoader worker process 0.
Original Traceback (most recent call last):
  File "/home/www/.conda/envs/py38_vits/lib/python3.8/site-packages/torch/utils/data/_utils/worker.py", line 308, in _worker_loop
    data = fetcher.fetch(index)
  File "/home/www/.conda/envs/py38_vits/lib/python3.8/site-packages/torch/utils/data/_utils/fetch.py", line 54, in fetch
    return self.collate_fn(data)
  File "/home/www/Bert-VITS2-Medlive/data_utils.py", line 242, in __call__
    bert_padded[i, :, : bert.size(1)] = bert
RuntimeError: The expanded size of the tensor (2048) must match the existing size (2560) at non-singleton dimension 0.  Target sizes: [2048, 43].  Tensor sizes: [2560, 43]

有人知道这种问题应该如何解决吗

[Stardust-minus]

3.9B输出的hidden size为2560，非2048