Dl/quantization/passes for splitted graphs #25
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
…_for_splitted_graphs
### Changes Fixed output text in the ptq onnx example ### Reason for changes Incorrect output text in the ptq onnx example ### Related tickets N/A ### Tests test_examples
daniil-lyakhov
force-pushed
the
dl/quantization/passes_for_splitted_graphs
branch
6 times, most recently
from
27899b0
to
81ad932
Compare
daniil-lyakhov
force-pushed
the
dl/quantization/passes_for_splitted_graphs
branch
from
c3efc6a
to
4a244f3
Compare
daniil-lyakhov
force-pushed
the
dl/quantization/passes_for_splitted_graphs
branch
2 times, most recently
from
d27a7f3
to
03e3faa
Compare
daniil-lyakhov
force-pushed
the
dl/quantization/passes_for_splitted_graphs
branch
4 times, most recently
from
07becc6
to
f3da495
Compare
…2289) ### Changes as stated in the title
### Changes According to https://docs.openvino.ai/2023.1/openvino_docs_OV_UG_lpt.html Placement FQ nodes like in OV backend. ### Related tickets 124839
### Changes - As stated in the title. ### Related tickets - 125654
### Changes Strip traced tensors when exit from the tracing context ### Reason for changes `nncf_model_input` patched `torch.Tensor` with `TracedTensor` which does not support deepcopy. ### Related tickets 125357 ### Tests test_wrap_model_with_example_input e2e 506 test_examples 154
### Changes - Migrated to openvino.convert_model(...) - Upgraded OpenVINO version to 2023.2 in PTQ TensorFlow example ### Reason for changes Aligned examples with the OpenVINO 2023.2 ### Related tickets 125257 ### Tests test_examples 157
### Changes - Add models_hub_tests to check `wrap_model`, like https://github.com/openvinotoolkit/openvino/tree/master/tests/model_hub_tests/torch_tests - Add `Model Hub` github action to manual run ### Related tickets 124852 ### Tests tests/torch/models_hub_test
### Changes Fix `do_copy` option for `strip` function ### Reason for changes Does not make a copy of the quantized model by PTQ in `strip` function.
### Changes - Changed references for the LSTM model because of openvinotoolkit/openvino#20368. - Changed references for the `yolo-v4-tiny-tf` model due to layers name changing. - Fixed bug with the in-place layer insertion. ### Reason for changes - Updated initialization for ReadValue layers (for now not only zeroes can be set as the initial value). - Probably MO frontend (TF) changed that led to layer name changes in models. ### Related tickets - 125066 ### Tests - openvino recommit with nightly - 24 --------- Co-authored-by: Daniil Lyakhov <[email protected]>
…penvinotoolkit#2290) ### Changes * Post hooks inputs are being unwrapped from `torch.return_type` types to `torch.tensor` type. * Pre hooks outputs are being wrapped from `torch.tensor` to `torch.return_type` in case torch input were wrapped in the first place. ### Reason for changes To enable post hook insertion after torch operations which return `torch.tensor_type` values instead of `torch.tensor` values. ### Related tickets ### Tests * tests/torch/test_nncf_network.py is updated * tests/torch/test_return_types.py is introduced
github-actions
bot
added
documentation
…sses_for_splitted_graphs
### Changes - Update torch examples, added argument `--export-model-path` instead of `--to-onnx` that select output type by suffix '.xml' or '.onnx'. - Update tests to use Command to run examples. - Remove tests of q_dq export path. - Remove code to make html report, now metrics dumps in `results.csv`. - Added "build url" in `results.csv`. - Actualize reference metrics. - Move normalization of input to accuracy cheker configs. - Use `target_ov` and `target_pt` to define reference metrics for backends. - Skip test `unet_mapillary_int8` and `unet_mapillary_magnitude_sparsity_int8` models by 123448. - Skip `resnet18_imagenet_binarization_dorefa` model by 22543. Updates in CI: - Update script to report `final e2e_result.html` for e2e tests. - Now if any test falls trigger_job will be fails. After merge requires update ci-pipelines. ### Related tickets 117885 ### Tests tests/torch/test_sota_checkpoints.py ### TODO - update metrics after merge openvinotoolkit#2227 - check metrics after openvinotoolkit#2211
…kit#2245) ### Changes Adjusted codecov.yml for a project coverage target of 90%, set up paths for per-component flags (i.e. TORCH, TENSORFLOW etc.) so that the per-component test coverage could be calculated correctly. Made codecov post their trademark "Reach" graph for visual cues to coverage differences. Codecov will now wait for all CI checks to pass before posting the first version of the comment. ### Reason for changes Should preserve the good coverage we have right now - the target checks are informational at the moment, may want to make them blocking in the future. Should evaluate the reach graph to see whether it is helpful in estimating the coverage impact of a given PR. ### Related tickets N/A ### Tests Existing precommit tests
…inotoolkit#2047) Added a warning when starting training object detection model without provided weights #### Tickets 116938
…oolkit#2082) ### Changes 1) Introduce Layout parameters for Convolutional / MM layers for OV backend 2) Refactor ChannelAlignment/SmoothQuant/MinMax/Wegihts compression algorithms to use layout parameters from layer attributes 3) Fix `get_matmul_channel_axes` from openvino node utils ### Reason for changes 1-2) To separate weights layout collection code in layer attributes instead of separate collection in each algorithm 3) To work properly with 1d constants in PTQ algorithms ### Related tickets 114328 114583 ### Tests * tests/openvino/native/test_layer_attributes.py are updated * tests/openvino/native/quantization/test_channel_alignment.py * tests/openvino/native/test_smooth_quant.py * tests/openvino/native/test_node_utils.py
### Changes - Patch missed magic function ``` ['__ixor__', '__itruediv__', '__rtruediv__', '__abs__', '__ror__', '__ior__', '__rdiv__', '__rxor__', '__iand__', '__neg__', '__ipow__', '__rpow__', '__invert__', '__rand__'] ``` - Update functions names for metatypes. - New metatype `PTNegativeMetatype` for `-tensor` and `tensor.neg()` operations. ### Related tickets 124852 ### Tests test_patch_magic_functions test_op_for_patch_magic_functions models_hub_test - Was `91 failed, 602 passed`, after `83 failed, 610 passed`
### Changes
Replace node name only if node id is string (it can also be `int` like
in patterns)
### Reason for changes
```python
for original_name in nx_graph.nodes():
> dot_name = original_name.replace(__CHARACTER_REPLACE_FROM, __CHARACTER_REPLACE_TO)
E AttributeError: 'int' object has no attribute 'replace'
nncf/common/utils/dot_file_rw.py:104: AttributeError
```
### Tests
tests/common/graph/test_graph_pattern.py::test_dump
daniil-lyakhov
force-pushed
the
dl/quantization/passes_for_splitted_graphs
branch
from
24cbc33
to
4266ed6
Compare
…#2288) ### Changes Added `INT8_SYM` compression mode ### Reason for changes `INT8_SYM` mode can provide a better performance and required for dynamic quantization ### Related tickets 124823 ### Tests Updated tests/openvino/native/quantization/test_weights_compression.py
### Changes Update reference of metrics according last weekly tests
### Changes - incorrect use of `map` function, as result `any(is_input_mask_empty_map)` condition always `False` - fix node type comparing with list
### Changes - As stated in the title.
daniil-lyakhov
force-pushed
the
dl/quantization/passes_for_splitted_graphs
branch
from
10e286e
to
127ae4e
Compare
…sses_for_splitted_graphs
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
Reason for changes
Related tickets
Tests