when bert_src is bert-base-multilingual-cased and bert_tgt is bert-base-chinese，The system cannot be trained

[zhaofuchen]

My configuration information is as follows:
preprocessing.yml

train_src: work-en-zh-1/train.tok.srcmt
train_tgt: work-en-zh-1/train.tok.pe
valid_src: work-en-zh-1/dev.tok.srcmt
valid_tgt: work-en-zh-1/dev.tok.pe
save_data: prep-data
src_vocab_size: 200000
tgt_vocab_size: 200000
shard_size: 100000
bert_src: bert-base-multilingual-cased
bert_tgt: bert-base-chinese
src_seq_length: 200
tgt_seq_length: 100

train-config.yml

save_model: ape-model
data: prep-data
train_steps: 50000
start_decay_steps: 50000
valid_steps: 1000
save_checkpoint_steps: 10000
keep_checkpoint: 1
rnn_size: 768
word_vec_size: 768
transformer_ff: 3072
heads: 12
layers: 12
position_encoding: 'true'
share_embeddings: 'true'
share_decoder_embeddings: 'true'
encoder_type: bert
enc_bert_type: bert-base-multilingual-cased
decoder_type: bert
dec_bert_type: bert-base-chinese
bert_decoder_token_type: B
bert_decoder_init_context: 'true'
share_self_attn: 'true'
dropout: 0.1
label_smoothing: 0.1
optim: bertadam
learning_rate: 0.00005
warmup_steps: 5000
batch_type: tokens
normalization: tokens
accum_count: 2
batch_size: 512
max_grad_norm: 0
param_init: 0
param_init_glorot: 'true'
valid_batch_size: 8
average_decay: 0.0001
seed: 42
world_size: 1
gpu_ranks: 0

The system fails to train and prompts the following error

[2020-07-02 18:07:13,048 INFO] Building & saving training data...
[2020-07-02 18:07:13,049 INFO] Reading source and target files: work-en-zh-1/train.tok.srcmt work-en-zh-1/train.tok.pe.
[2020-07-02 18:07:13,054 INFO] Building shard 0.
[2020-07-02 18:07:17,401 INFO] * saving 0th train data shard to prep-data.train.0.pt.
[2020-07-02 18:07:18,624 INFO] Building & saving validation data...
[2020-07-02 18:07:18,625 INFO] Reading source and target files: work-en-zh-1/dev.tok.srcmt work-en-zh-1/dev.tok.pe.
[2020-07-02 18:07:18,625 INFO] Building shard 0.
[2020-07-02 18:07:19,202 INFO] * saving 0th valid data shard to prep-data.valid.0.pt.
[2020-07-02 18:07:19,400 INFO] Building & saving vocabulary...
[2020-07-02 18:07:19,557 INFO] * reloading prep-data.train.0.pt.
[2020-07-02 18:07:19,844 INFO] BERT vocab has 21128 tokens.
[2020-07-02 18:07:19,930 INFO] * tgt vocab size: 21128.
[2020-07-02 18:07:19,944 INFO] BERT vocab has 119547 tokens.

[2020-07-02 19:01:55,260 INFO] encoder: 177262848
[2020-07-02 19:01:55,260 INFO] decoder: 85095560
[2020-07-02 19:01:55,260 INFO] * number of parameters: 262358408
[2020-07-02 19:01:55,271 INFO] Starting training on GPU: [0]
[2020-07-02 19:01:55,271 INFO] Start training loop and validate every 1000 steps...
[2020-07-02 19:01:55,517 INFO] Loading dataset from prep-data.train.0.pt, number of examples: 6997
Traceback (most recent call last):
File "./OpenNMT-APE-master/train.py", line 109, in
main(opt)
File "./OpenNMT-APE-master/train.py", line 39, in main
single_main(opt, 0)
File "/data/nextcloud/dbc2017/files/ape/OpenNMT-APE-master/onmt/train_single.py", line 116, in main
valid_steps=opt.valid_steps)
File "/data/nextcloud/dbc2017/files/ape/OpenNMT-APE-master/onmt/trainer.py", line 209, in train
report_stats)
File "/data/nextcloud/dbc2017/files/ape/OpenNMT-APE-master/onmt/trainer.py", line 330, in _gradient_accumulation
trunc_size=trunc_size)
File "/data/nextcloud/dbc2017/files/ape/OpenNMT-APE-master/onmt/utils/loss.py", line 158, in call
loss, stats = self._compute_loss(batch, **shard)
File "/data/nextcloud/dbc2017/files/ape/OpenNMT-APE-master/onmt/utils/loss.py", line 233, in _compute_loss
scores = self.generator(bottled_output)
File "/root/anaconda3/envs/python367/lib/python3.6/site-packages/torch/nn/modules/module.py", line 550, in call
result = self.forward(*input, **kwargs)
File "/root/anaconda3/envs/python367/lib/python3.6/site-packages/torch/nn/modules/container.py", line 100, in forward
input = module(input)
File "/root/anaconda3/envs/python367/lib/python3.6/site-packages/torch/nn/modules/module.py", line 550, in call
result = self.forward(*input, **kwargs)
File "/root/anaconda3/envs/python367/lib/python3.6/site-packages/torch/nn/modules/linear.py", line 87, in forward
return F.linear(input, self.weight, self.bias)
File "/root/anaconda3/envs/python367/lib/python3.6/site-packages/torch/nn/functional.py", line 1610, in linear
ret = torch.addmm(bias, input, weight.t())
RuntimeError: The expanded size of the tensor (119547) must match the existing size (21128) at non-singleton dimension 1. Target sizes: [297, 119547]. Tensor sizes: [21128]

Through analysis, we can know that it is because src vocab size and tgt vocab size is not same，As follows:

[2020-07-02 19:26:43,820 INFO] * src vocab size = 119547
[2020-07-02 19:26:43,820 INFO] * tgt vocab size = 21128

Why does this happen？
Has anyone encountered this problem, please help me, thank you very much!