This repository has been archived by the owner on Oct 15, 2019. It is now read-only.
[WIP] fixing memory issue in RNN #112
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
jermainewang
force-pushed
the
rnn-memory
branch
from
bd8b3b9
to
d07eb78
Compare
1. Tape autograd logic. 2. Use generated id rather than builtin id function for tape. 3. Fix bug in tape: gradient += is also recorded in tape.
Conflicts: minpy/primitive.py minpy/tape.py
|
Everything on minpy side should be fixed in this PR right now. @lryta @ZihengJiang @hotpxl please have a look. @HrWangChengdu would you mind help me rerun all the examples to see whether they are still working? |
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Main reason of RNN memory issue:
Changes:
tape.get_gradientnow takes a tuple of arrays as input and return the tuple as results.core.gradfunction is self-contained (all the inputs and outputs are specified through the interface). Therefore thetape.get_gradientcould take no arguments since all arguments required to get gradient must be the root node of the tape.tape.get_gradientfunction should implement a reverse order from the last node to the root node. The forward path could be saved since it is already done when generating grad_records.@hotpxl