Search graph for quantization parameters #6690
Merged
Conversation
This file contains hidden or 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
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/6690
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 9ea1837 with merge base 793f17e ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
Erik-Lundell
added
partner: arm
facebook-github-bot
added
the
CLA Signed
Erik-Lundell
force-pushed
the
search-graph
branch
from
b912390 to
feb2354
Compare
|
Rebase pls |
Erik-Lundell
force-pushed
the
search-graph
branch
from
feb2354 to
cfa9093
Compare
Generalizes the search for quantization parameters.
The idea is to make a graph like this a valid quantized graph:
dq -> view -> transpose -> some_op
^
/
dq ------> expand -------/
For a subset of operations 'passable_op' it is is allowed to "pass
through" the op when searching for qparams. If multiple qparams
are encounterd in one search, they are asserted to be equal.
Signed-off-by: Erik Lundell <[email protected]>
Change-Id: I6dbb82fb39164c246ea74b9642d907dba22ab2c3
Signed-off-by: Erik Lundell <[email protected]> Change-Id: I4feb4b5c6d269d7c0ff4312c17fec52da413fa5a
Signed-off-by: Erik Lundell <[email protected]> Change-Id: If415a5ee3185c3401069652906a0633491bca1b0
Erik-Lundell
force-pushed
the
search-graph
branch
from
f7c6491 to
9ea1837
Compare
freddan80
approved these changes
Nov 11, 2024
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Outdated suggestions cannot be applied.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.
(This change was merged but had to be reverted due to type checking errors, see #6452)
Generalizes the search for quantization parameters. The idea is to make a graph like this (part of) a valid quantized graph:
dq -> view -> transpose -> some_op
dq ------> expand -----------/^
For a subset of operations 'passable_op' that don't modify the values of the tensors it is is allowed to "pass through" the op when searching for qparams. If multiple qparams are encounterd in one search, they are asserted to be equal.
The reason for this is to unify what a quantized graph looks like, removing the need for some ad hoc exceptions (for example checking if a view is before an addmm etc.) that were in place before and in general cleaning up the handling of quantization in the ArmBackend. In particular, some decompositions in the to_edge step seem to not insert q/ dq ops which breaks the pattern of dq -> op -> q.
With this change, Arm internal passes can also in some cases skip inserting quantize ops, simplifying the passes.