Run decompositions before the quantizer #7111
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/7111
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit a11afab with merge base 2d499b3 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
|
This pull request was exported from Phabricator. Differential Revision: D66461406 |
zonglinpeng
approved these changes
Dec 2, 2024
|
|
||
| if model_gm_has_SDPA(model_gm): # pyre-fixme[6] | ||
| decomp_table = torch.export.default_decompositions() | ||
| ops_to_keep = [ |
There was a problem hiding this comment.
would be nice to leave the same comment inline
| torch.ops.aten.linear.default, | ||
| torch.ops.aten.matmul.default, | ||
| ] | ||
| # pyre-fixme[6]: For 1st argument expected `Dict[typing.Callable[..., typing.Any |
There was a problem hiding this comment.
pyre are disabled in ET but still used internally. We need to sort it out but not in this PR
facebook-github-bot
pushed a commit
that referenced
this pull request
Dec 2, 2024
Summary: In the current flow, decompositions run in `to_edge()`, long after the quantization process is done. This creates a lot of issues, since we cannot quantize any operations contained in the large operators that the graph tracer can give (e.g. aten.scaled_dot_product_attention, aten.rnn_<tanh, relu>.input, and a few others). Any models using those will see many fp32 operators in the final graph. Running the decomps earlier solves the problem, but we need to retain a couple operators that we do rely on in the quantizer, like `aten.linear`, `aten.conv1d` and `aten.conv2d`. Reviewed By: zonglinpeng Differential Revision: D66461406
facebook-github-bot
force-pushed
the
export-D66461406
branch
from
cc66aa4 to
81c7522
Compare
|
This pull request was exported from Phabricator. Differential Revision: D66461406 |
Summary: In the current flow, decompositions run in `to_edge()`, long after the quantization process is done. This creates a lot of issues, since we cannot quantize any operations contained in the large operators that the graph tracer can give (e.g. aten.scaled_dot_product_attention, aten.rnn_<tanh, relu>.input, and a few others). Any models using those will see many fp32 operators in the final graph. Running the decomps earlier solves the problem, but we need to retain a couple operators that we do rely on in the quantizer, like `aten.linear`, `aten.conv1d` and `aten.conv2d`. Reviewed By: zonglinpeng Differential Revision: D66461406
mcremon-meta
force-pushed
the
export-D66461406
branch
from
81c7522 to
a11afab
Compare
|
This pull request was exported from Phabricator. Differential Revision: D66461406 |
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.
Summary:
In the current flow, decompositions run in
to_edge(), long after the quantization process is done. This creates a lot of issues, since we cannot quantize any operations contained in the large operators that the graph tracer can give (e.g. aten.scaled_dot_product_attention, aten.rnn_<tanh, relu>.input, and a few others).Any models using those will see many fp32 operators in the final graph. Running the decomps earlier solves the problem, but we need to retain a couple operators that we do rely on in the quantizer, namely
aten.linear,aten.conv1dandaten.conv2d.Differential Revision: D66461406