[Torch][PTQ] Examples are updated for the new PTQ TORCH backend #2246
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
daniil-lyakhov
requested review from
alexsu52,
nikita-malininn and
AlexanderDokuchaev
daniil-lyakhov
force-pushed
the
dl/torch/fix_splitted_inference_graph
branch
from
7ffdcf7 to
f921950
Compare
Codecov Report
Additional details and impacted files@@ Coverage Diff @@
## develop #2246 +/- ##
===========================================
- Coverage 90.71% 85.92% -4.80%
===========================================
Files 485 485
Lines 43632 43679 +47
===========================================
- Hits 39582 37531 -2051
- Misses 4050 6148 +2098
Flags with carried forward coverage won't be shown. Click here to find out more.
|
alexsu52
reviewed
Nov 6, 2023
daniil-lyakhov
changed the title
Ticket 124397 |
AlexanderDokuchaev
approved these changes
Nov 6, 2023
examples/post_training_quantization/torch/ssd300_vgg16/requirements.txt
Outdated
Show resolved
Hide resolved
nikita-savelyevv
approved these changes
Nov 7, 2023
daniil-lyakhov
force-pushed
the
dl/torch/fix_splitted_inference_graph
branch
from
a6bf44f to
5c0ab0b
Compare
…ted_inference_graph
| @@ -540,6 +540,7 @@ def _get_quantization_target_points( | |||
| self._backend_entity.shapeof_metatypes, | |||
| self._backend_entity.dropout_metatypes, | |||
| self._backend_entity.read_variable_metatypes, | |||
| nncf_graph_contains_constants=backend != BackendType.TORCH, | |||
There was a problem hiding this comment.
I'd like to note that we introduce backend-specific logic into a common part of the PTQ, which does not fit with our approach. Do you agree with that @alexsu52, @AlexanderDokuchaev, @nikita-savelyevv?
There was a problem hiding this comment.
Yes, I agree. @daniil-lyakhov Could you update filter_constant_nodes pass for discontinuous graphs?
There was a problem hiding this comment.
Done, please check
There was a problem hiding this comment.
I tried implementation we discussed db0ad8d, but found a corner case that is breaking the pass:
Graph
Before pass:
Expected after pass:
Actual after pass:
That's because convolution node should be an input node to survive filtering, but we have no prior information yet that can help us to mark this convolution as an input node
nikita-malininn
approved these changes
Nov 8, 2023
github-actions
bot
added
NNCF OpenVINO
daniil-lyakhov
force-pushed
the
dl/torch/fix_splitted_inference_graph
branch
from
87c8d95 to
db0ad8d
Compare
github-actions
bot
added
NNCF Common
This reverts commit db0ad8d.
|
Torch precommit is failing due to problems in e00f6b7, ticket 124679 |
nikita-malininn
approved these changes
Nov 10, 2023
daniil-lyakhov
added a commit
to daniil-lyakhov/nncf
that referenced
this pull request
Nov 10, 2023
…vinotoolkit#2246) - Do not filter constant nodes for torch backend in the inference graph - Fix version in requarements.txt for examples of post_training_quantization - for ssd300_vgg16 is not available to use torch 2.1.0 (failed on export to onnx Unsupported: ONNX export of operator get_pool_ceil_padding, tracing is not supporting too) - Update metrics - Add to PTEngine convert inputs to model's device to sync behavior with `create_compress_model` - Mobilenet_v2 example converting PyTorch model to IR by tracing (without onnx). - nncf.quantize for PyTorch works with copy of the target model To make PTQ work properly with disconnected graphs (like in [example](https://github.com/openvinotoolkit/nncf/blob/develop/examples/post_training_quantization/torch/ssd300_vgg16/main.py)) 124417 test_examples build 128 --------- Co-authored-by: Alexander Dokuchaev <alexander.dokuchaev@intel.com>
alexsu52
pushed a commit
that referenced
this pull request
Mar 21, 2024
### Changes Update accuracy drop reference value. ### Reason for changes In #2246 reference accuracy values for original and quantized models were updated, but the corresponding accuracy drop reference value was not updated. ### Related tickets 136123
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.
Changes
create_compress_modelReason for changes
To make PTQ work properly with disconnected graphs (like in example)
Related tickets
124417
Tests
test_examples build 128