dcslin commented on pull request #733:
URL: https://github.com/apache/singa/pull/733#issuecomment-646635254
Currently it is working:
```
# PYTHONPATH=./build/python python3 examples/rnn/imdb_train.py -h
usage: imdb_train.py [-h] [-m MAX_EPOCH] [-b BS] [-l LR] [-i DEVICE_ID]
[--mode MODE] [--return-sequences RETURN_SEQUENCES]
[--bidirectional BIDIRECTIONAL] [--num-layers
NUM_LAYERS]
optional arguments:
-h, --help show this help message and exit
-m MAX_EPOCH, --max-epoch MAX_EPOCH
maximum epochs
-b BS, --batch-size BS
batch size
-l LR, --learning-rate LR
initial learning rate
-i DEVICE_ID, --device-id DEVICE_ID
which GPU to use
--mode MODE relu, tanh, lstm, gru
--return-sequences RETURN_SEQUENCES
return sequences
--bidirectional BIDIRECTIONAL
bidirectional lstm
--num-layers NUM_LAYERS
num layers
# PYTHONPATH=./build/python python3 examples/rnn/imdb_train.py
epoch 0 loss [0.71537286]; acc 0.546
epoch 1 loss [0.65785825]; acc 0.604
epoch 2 loss [0.6347779]; acc 0.636
epoch 3 loss [0.61257607]; acc 0.661
epoch 4 loss [0.5874256]; acc 0.685
eval acc 0.594
```
Implemented the "Ex" versions of api, which is better with supporting on
masking and row major(no need transpose). But encountering
CUDNN_STATUS_EXECUTION_FAILED error
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
