huggingface/xpu: update transformers test instruction

Signed-off-by: Dmitry Rogozhkin <dmitry.v.rogozhkin@intel.com>

Author: dvrogozh

huggingface/accelerate-with-pytorch-xpu.md

@@ -6,13 +6,25 @@ Haggingface Accelerate supports PyTorch XPU backend starting from version `0.32.
 
 # Testing
 
-Accelerate library can be formally tested for XPU backend by the project included tests on the system with supported Intel GPU graphics:
+Accelerate library can be formally tested for XPU backend by the project included tests on the system with supported Intel GPU graphics. To run tests:
+
+* Install Accelerate library and test dependencies:
+
+```
+pip install -e .
+pip install -e ".[dev]"
+```
+
+* Execute tests:
 
 ```
 cd /path/to/accelerate/clone/copy
 python3 -m pytest tests
 ```
 
+> [!NOTE]
+> Accelerate depends on PyTorch `torch` package which gets installed during installation of `accelerate`. If you need to test against different version of PyTorch – reinstall it after installing Accelerate.
+
 # Using with PyTorch 2.5
 
 | Accelerate | Failed | Passed | Skipped |

huggingface/transformers-with-pytorch-xpu.md

@@ -6,7 +6,16 @@ Haggingface Transformers support PyTorch XPU backend starting from version `4.42
 
 # Testing
 
-Transformers library can be formally tested for XPU backend by the project included tests on the system with supported Intel GPU graphics. To run tests first create `spec.py` file with the following contents:
+Transformers library can be formally tested for XPU backend by the project included tests on the system with supported Intel GPU graphics. To run tests:
+
+* Install Transformers library and test dependencies:
+
+```
+pip install -e .
+pip install -e ".[dev]"
+```
+
+* Create `spec.py` file with the following contents:
 
 ```
 import torch
@@ -18,13 +27,16 @@ EMPTY_CACHE_FN = torch.xpu.empty_cache
 DEVICE_COUNT_FN = torch.xpu.device_count
 ```
 
-After that tests can be executed as follows:
+* After that tests can be executed as follows:
 
 ```
 cd /path/to/transformers/clone/copy
 TRANSFORMERS_TEST_DEVICE_SPEC=spec.py python3 -m pytest --pspec tests/
 ```
 
+> [!NOTE]
+> Some Transformers tests depend on `torchvision` built with `libPNG`. Install on Ubuntu with `apt-get install libpng-dev`, then rebuild `torchvision` and rerun affected Transformers tests.
+
 # Using with PyTorch 2.5
 
 XPU backend in PyTorch 2.5 greatly improved coverage of supported aten eager mode operators with just few missing for the executed Transformer tests. They are much easier to indentify now since [intel/torch-xpu-ops#318] is available in PyTorch 2.5 and adding `PYTORCH_ENABLE_XPU_FALLBACK=1` works fluently.