chrishkchris opened a new issue #680: Propose to upgrade RNN operation buffering for Graph URL: https://github.com/apache/singa/issues/680 In singa/examples/rnn/train.py, When we run it with Graph using Sequential, i.e. line 200: model.graph(True, True), the training can complete without error. However, when we run it with Graph using node input dependencies, i.e. line 200: model.graph(True, False), it will display the error as below:  Summarized from some discussion with @XJDKC, the error may be due to the following: 1. There may be some operations we have used but did not specify read_blocks and write_blocks well 2. Hence, some operations are performed in advance because the operation is not buffer as required Therefore, to support computation of graph using node input dependencies, we may need to update the buffering of some RNN operation.
---------------------------------------------------------------- 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] With regards, Apache Git Services
