This repository has been archived by the owner on Nov 17, 2023. It is now read-only.
[FEATURE] Add query_keys transformer version without split #21115
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
|
Hey @agrabows , Thanks for submitting the PR
CI supported jobs: [unix-cpu, unix-gpu, website, windows-cpu, centos-gpu, edge, sanity, centos-cpu, windows-gpu, miscellaneous, clang] Note: |
mseth10
added
pr-awaiting-testing
agrabows
force-pushed
the
qk_fuse_no_split
branch
from
5c11296
to
2495ae9
Compare
mseth10
added
pr-awaiting-testing
bartekkuncer
suggested changes
Aug 9, 2022
mseth10
added
pr-awaiting-testing
mseth10
added
pr-awaiting-testing
mseth10
added
pr-awaiting-testing
|
|
mseth10
added
pr-work-in-progress
mseth10
added
pr-awaiting-testing
bartekkuncer
suggested changes
Aug 12, 2022
bgawrych
reviewed
Aug 12, 2022
mseth10
added
pr-awaiting-testing
mseth10
added
pr-awaiting-testing
bgawrych
reviewed
Aug 22, 2022
| input_names.emplace_back("min_q"); | ||
| input_names.emplace_back("max_q"); | ||
| } | ||
| input_names.emplace_back("keys"); |
There was a problem hiding this comment.
"key" input name should be before "min_q" and "max_q"
Comment on lines
377
to
385
| float* output_min_0 = outputs[2].data().dptr<float>(); | ||
| float* output_max_0 = outputs[3].data().dptr<float>(); | ||
| float* output_min_1 = outputs[4].data().dptr<float>(); | ||
| float* output_max_1 = outputs[5].data().dptr<float>(); | ||
| *output_min_0 = min_output_; | ||
| *output_max_0 = max_output_; | ||
| *output_min_1 = min_output_; | ||
| *output_max_1 = max_output_; | ||
| } |
There was a problem hiding this comment.
There are only 3 output tensors. Doesn't it throw exception?
There was a problem hiding this comment.
it didn't but you are right, this code was incorrect
| @@ -29,54 +29,78 @@ | |||
| #include "operator/subgraph/common.h" | |||
| #include "dnnl_transformer-inl.h" | |||
|
|
|||
| // 3 tensors within one (queries key values) = | |||
| // 3 tensors within one (queries key values) | |||
There was a problem hiding this comment.
Suggested change
| // 3 tensors within one (queries key values) | |
| // 3 tensors within one (queries keys values) |
| *status = kFirstSwapAx; | ||
| matched_list->push_back(&input_node); | ||
| } | ||
| return true; |
There was a problem hiding this comment.
Shouldn't this return true be in 'if' statement below matched_list->push_back(&input_node); ?
mseth10
added
pr-awaiting-testing
|
@mxnet-bot run ci [unix-gpu] |
|
Jenkins CI successfully triggered : [unix-gpu] |
mseth10
added
pr-awaiting-testing
bgawrych
approved these changes
Aug 23, 2022
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.
Description
MXNet is fusing split, reshape, swapaxis and batch_dot operators for performance purpose. In gpt-2 model this fuse could be done as well if we exclude split.
->
Checklist
Essentials