Fix quantize model deploy bugs when using MKLDNN #45920
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
|
你的PR提交成功，感谢你对开源项目的贡献! |
wozna
reviewed
Sep 9, 2022
yeliang2258
changed the title
yeliang2258
changed the title
yeliang2258
changed the title
wozna
reviewed
Sep 30, 2022
paddle/fluid/framework/ir/mkldnn/params_quantization_mkldnn_pass.cc
Outdated
Show resolved
Hide resolved
paddle/fluid/framework/ir/mkldnn/params_quantization_mkldnn_pass.cc
Outdated
Show resolved
Hide resolved
jiangjiajun
requested changes
Oct 13, 2022
jiangjiajun
approved these changes
Oct 13, 2022
ZeyuChen
pushed a commit
that referenced
this pull request
Oct 20, 2022
* Fix quantize model deploy bugs when using MKLDNN (#45920) * fix immutable op quantize bugs * fix * fix build bug * fix test * notest,test=inference * fix ppyoloe acc drop bugs * fix test * fix test * add test * fix * fix * fix test * fix refined name bug * fix test * bias fix * fix matmul weight dequant bug * re-ci * fix tester * fix test * fix tester * update weight dequantize func * update code * update test for converage * update test * update cmake * update cmakelist * update code * rerun ci * remove useless code * re-ci * update code * update code * fix header * update code for log
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Outdated suggestions cannot be applied.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 types
Bug fixes
PR changes
Others
Describe
Fix ppyoloe、MobileNetV3_large、Ernie 3.0-Medium and bert acc drop bugs.
Avoid the problem that the weights are quantize and dequantized multiple times when multiple conv2d share the same weight. Weight names in the picodet model: conv2d_160.w_0.quantized.dequantized
Fix immutable op quantize bugs in MKLDNN.
Similar to the following combination, the output of shape is int32 type, if the slice is quantized, an error will be reported.
Therefore, I add a judgment, If the input of the OP to be quantized is not float32 type, the OP will not be quantized.
The model is: https://bj.bcebos.com/v1/paddle-slim-models/act/ppyoloe_crn_l_300e_coco_quant.tar