-
Notifications
You must be signed in to change notification settings - Fork 5.5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[DNNL] Fix accuracy in INT8 FC #22404
Merged
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
test=develop
wojtuss
reviewed
Jan 30, 2020
c519621
to
4444af7
Compare
wojtuss
approved these changes
Jan 31, 2020
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
Please cherry-pick into release/1.7 |
Sand3r-
added a commit
to Sand3r-/Paddle
that referenced
this pull request
Feb 3, 2020
test=release/1.7 * Enable quantize to reorder to nchw as well * Correct FC MKL-DNN input dim requirements to accept 3D * Improve DNNL FC format, error and 3D input handling * Improve error checking in FC * Improve PADDLE_ENFORCE messages in fc-related files * Remove data layout attribute from obligatory pass args * Fix message in fc_mkldnn_pass to be logically correct
luotao1
pushed a commit
that referenced
this pull request
Feb 4, 2020
test=release/1.7 * Enable quantize to reorder to nchw as well * Correct FC MKL-DNN input dim requirements to accept 3D * Improve DNNL FC format, error and 3D input handling * Improve error checking in FC * Improve PADDLE_ENFORCE messages in fc-related files * Remove data layout attribute from obligatory pass args * Fix message in fc_mkldnn_pass to be logically correct
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
PR introduces changes that enable int8 FC to correctly perform computations in ernie topology.
Terminology used in description:
nhwc
-like ornchw
-like means any format that is descendant of these formats, such as ncw or nwc for 3-dim data blobs, nhwc, nchw for 4-dims and so on.Description:
Background:
The
nhwc
data layout was formerly used in all int8 computations so far, and quantize op had it hardcoded to reorder to nhwc-like format during quantizaiton.Problem:
Unfortunately, due to how this format is laid out in memory, it was not possible to correctly use this format when fully-connected op used
in_num_col_dims
parameter to merge some of the first dimensions, since it assumes an nchw-like format.Solution:
To counter this, the PR makes it possible to set to which format (or in fact layout) should quantize reorder to during quantizaiton. Possible options being either
NCHW
orNCHW
by setting itsoutput_format
parameter.To further utilise this new feature,
cpu_quantize_pass
now accepts an optional attributedata_layout
which allows to set desired data format for all quantizes that will be added to the graph.