Add rnn_search_model #729
Add rnn_search_model #729
Conversation
peterzhang2029
force-pushed
the
add_rnn_search
branch
from
ce13bd5 to
3fd720b
Compare
|
@peterzhang2029 I hope to know whether the multi-threaded/multi-card training can correctly work for RNN search model? Thank you. |
|
@lcy-seso 目前在多线程上面测试是可以work的，在多卡上面现在用的机器环境会有显存的问题，需要再验证一下。 |
|
好的~ 感谢。 |
| @@ -0,0 +1,25 @@ | |||
| source_dict_dim = 10000 | |||
| target_dict_dim = 10000 | |||
|
|
|||
There was a problem hiding this comment.
move line 1 ~ 2 into ModelConfig.
| target_dict_dim = model_conf.target_dict_dim | ||
| is_generating = model_conf.is_generating | ||
| beam_size = model_conf.beam_size | ||
| max_length = model_conf.max_length |
There was a problem hiding this comment.
line 34 ~ 41, pass these configurations as function parameters.
| paddle.reader.shuffle( | ||
| paddle.dataset.wmt16.train(train_conf.source_dict_dim, | ||
| train_conf.target_dict_dim), | ||
| buf_size=1000), |
There was a problem hiding this comment.
please make buf_size larger, for example, 100000.
| paddle.reader.shuffle( | ||
| paddle.dataset.wmt16.test(train_conf.source_dict_dim, | ||
| train_conf.target_dict_dim), | ||
| buf_size=1000), |
There was a problem hiding this comment.
please make buf_size larger, for example, 100000.
|
|
resolve #728