Bump pytorch 2.9.0 (#3690)

### Changes

- Updated `torch` to version 2.9.0 and `torchvision` to version 0.24.0
  - some examples still used 2.8.0 until update optimum
```
    from torch.onnx.symbolic_opset14 import (  # noqa: E402
ImportError: cannot import name '_attention_scale' from 'torch.onnx.symbolic_opset14' (/tmp/pytest-of-runner/pytest-0/test_examples_llm_compression_0/venv/lib/python3.10/site-packages/torch/onnx/symbolic_opset14.py)
```

- Set `dynamo=False` in all calls to `torch.onnx.export` throughout the
codebase and tests to ensure compatibility with PyTorch 2.9.0 and work
around issues with the `ultralytics` library.

- Applied a workaround in scripts using `ultralytics` to set the default
value for the `dynamo` parameter in `torch.onnx.export`, with a TODO to
remove after upstream fixes.

- Use `check_guards=False` for FX model to no add extra function to
model

- Update referenced FX graphs, in reason of fused `slice` and `select`
operations

### Tests

manual/job/post_training_quantization/738/
https://github.com/openvinotoolkit/nncf/actions/runs/18719524116
https://github.com/openvinotoolkit/nncf/actions/runs/18720059603
nightly/job/torch_nightly/687/

Author: AlexanderDokuchaev

constraints.txt

@@ -2,13 +2,14 @@
 openvino==2025.3.0
 
 # Pytorch
-torch==2.8.0
-torchvision==0.23.0
+torch==2.9.0
+torchvision==0.24.0
 
 # ONNX
 onnx==1.17.0; python_version < '3.13'
 onnx==1.18.0; python_version >= '3.13'
 onnxruntime==1.21.1
+onnxscript==0.5.4
 
 # TensorFlow
 tensorflow==2.15.1

docs/Installation.md

@@ -49,7 +49,7 @@ as well as the supported versions of Python:
 
 | NNCF      | OpenVINO   | PyTorch  | ONNX     | TensorFlow | Python |
 |-----------|------------|----------|----------|------------|--------|
-| `develop` | `2025.3.0` | `2.8.0`  | `1.17.0` | `2.15.1`   | `3.10` |
+| `develop` | `2025.3.0` | `2.9.0`  | `1.17.0` | `2.15.1`   | `3.10` |
 | `2.18.0`  | `2025.3.0` | `2.8.0`  | `1.17.0` | `2.15.1`   | `3.10` |
 | `2.17.0`  | `2025.2.0` | `2.7.1`  | `1.17.0` | `2.15.1`   | `3.10` |
 | `2.16.0`  | `2025.1.0` | `2.6.0`  | `1.17.0` | `2.15.1`   | `3.10` |
@@ -70,4 +70,4 @@ as well as the supported versions of Python:
 
 > (*) Python 3.9 or higher is required for TensorFlow 2.15.1
 
-This repository is tested on Python* 3.10.14, PyTorch* 2.8.0 (NVidia CUDA\* Toolkit 12.6) and TensorFlow* 2.15.1 (NVidia CUDA\* Toolkit 11.8).
+This repository is tested on Python* 3.10.14, PyTorch* 2.9.0 (NVidia CUDA\* Toolkit 12.6) and TensorFlow* 2.15.1 (NVidia CUDA\* Toolkit 11.8).

examples/llm_compression/openvino/tiny_llama_find_hyperparams/requirements.txt

@@ -4,4 +4,4 @@ openvino==2025.3.0
 optimum-intel==1.24.0
 transformers==4.52.1
 onnx==1.17.0
-torch==2.8.0
+torch==2.9.0

examples/llm_compression/torch_fx/tiny_llama/requirements.txt

@@ -2,5 +2,5 @@ transformers==4.52.1
 datasets==2.14.7
 openvino==2025.3.0
 optimum==1.24.0
-torch==2.8.0
-torchvision==0.23.0
+torch==2.9.0
+torchvision==0.24.0

examples/post_training_quantization/onnx/mobilenet_v2/requirements.txt

@@ -1,5 +1,5 @@
-torch==2.8.0
-torchvision==0.23.0
+torch==2.9.0
+torchvision==0.24.0
 fastdownload==0.0.7
 onnx==1.17.0
 onnxruntime==1.21.1

examples/post_training_quantization/onnx/yolov8_quantize_with_accuracy_control/main.py

@@ -34,6 +34,11 @@
 ROOT = Path(__file__).parent.resolve()
 
 
+# WA: to set default value for dynamo parameter to False
+# TODO(AlexanderDokuchaev): remove after update ultralytics
+torch.onnx.export = partial(torch.onnx.export, dynamo=False)
+
+
 def validate(
     model: onnx.ModelProto,
     data_loader: torch.utils.data.DataLoader,

examples/post_training_quantization/openvino/yolov8/main.py

@@ -10,6 +10,7 @@
 # limitations under the License.
 import re
 import subprocess
+from functools import partial
 from pathlib import Path
 from typing import Any
 
@@ -32,6 +33,11 @@
 ROOT = Path(__file__).parent.resolve()
 
 
+# WA: to set default value for dynamo parameter to False
+# TODO(AlexanderDokuchaev): remove after update ultralytics
+torch.onnx.export = partial(torch.onnx.export, dynamo=False)
+
+
 def validate(
     model: ov.Model, data_loader: torch.utils.data.DataLoader, validator: DetectionValidator, num_samples: int = None
 ) -> tuple[dict, int, int]:

examples/post_training_quantization/openvino/yolov8/requirements.txt

@@ -1,4 +1,4 @@
 ultralytics==8.3.22
 onnx==1.17.0
 openvino==2025.3.0
-torch==2.8.0
+torch==2.9.0

examples/post_training_quantization/openvino/yolov8_quantize_with_accuracy_control/main.py

@@ -33,6 +33,10 @@
 
 ROOT = Path(__file__).parent.resolve()
 
+# WA: to set default value for dynamo parameter to False
+# TODO(AlexanderDokuchaev): remove after update ultralytics
+torch.onnx.export = partial(torch.onnx.export, dynamo=False)
+
 
 def validate(
     model: ov.Model, data_loader: torch.utils.data.DataLoader, validator: SegmentationValidator, num_samples: int = None

examples/post_training_quantization/openvino/yolov8_quantize_with_accuracy_control/requirements.txt

@@ -1,4 +1,4 @@
 ultralytics==8.3.22
 onnx==1.17.0
 openvino==2025.3.0
-torch==2.8.0
+torch==2.9.0