frontend/pytorch: add aten::cross_entropy_loss translator and tests (fixes #29691, relates #28584)

[Rahuldrabit]

1. Core Implementation

File: src/frontends/pytorch/src/op/cross_entropy_loss.cpp

Implements the translate_cross_entropy_loss function that converts PyTorch's cross_entropy_loss operation to OpenVINO operations. The implementation follows PyTorch's specification:

cross_entropy_loss = log_softmax + nll_loss

Key features:

• Applies LogSoftmax along dimension 1 (class dimension)
• Uses one-hot masking plus ReduceSum to select target class probabilities (NLL loss)
• Supports optional class weights via Gather + Multiply
• Supports three reduction modes:
  • none (0): No reduction, returns per-sample loss
  • mean (1): Mean of all losses (default)
  • sum (2): Sum of all losses
• Handles both 2D inputs (N, C) and higher-dimensional inputs (N, C, d1, d2, ...)

2. Operation Registration

File: src/frontends/pytorch/src/op_table.cpp

• Added forward declaration: OP_CONVERTER(translate_cross_entropy_loss);
• Registered operation: {"aten::cross_entropy_loss", op::translate_cross_entropy_loss},

3. Tests

File: tests/layer_tests/pytorch_tests/test_cross_entropy_loss.py

Comprehensive test suite covering:

• Basic functionality with different reduction modes (mean, sum, none)
• Class weights
• Different input shapes (2D and spatial dimensions)
• Edge cases (binary classification, single sample, many classes)

[github-actions]

This PR will be closed in a week because of 2 weeks of no activity.

[Copilot]

Pull request overview

This PR implements the PyTorch cross_entropy_loss operation translator for OpenVINO, addressing issue #29691. The implementation follows PyTorch's specification where cross_entropy_loss = log_softmax + nll_loss, providing a complete translation from PyTorch operations to OpenVINO's operation set.

Changes:

• Implements cross_entropy_loss translator with support for three reduction modes (none, mean, sum), optional class weights, and spatial dimensions
• Registers the operation in the op_table with proper alphabetical ordering
• Adds comprehensive tests covering basic functionality, reduction modes, class weights, and spatial inputs

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

File	Description
src/frontends/pytorch/src/op/cross_entropy_loss.cpp	Core translator implementation with log_softmax + one-hot masking approach for NLL loss, supporting weights and all reduction modes
src/frontends/pytorch/src/op_table.cpp	Adds forward declaration and registration of the new translator in alphabetical order
tests/layer_tests/pytorch_tests/test_cross_entropy_loss.py	Test suite covering reduction modes (mean/sum/none), class weights, and spatial dimensions (2D and 4D inputs)

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The copyright year should be 2018-2026 to be consistent with all other files in this directory.

[Rahuldrabit]

@copilot open a new pull request to apply changes based on this feedback

[Copilot]

The copyright year should be 2018-2026 to be consistent with all other test files in this directory.

[github-actions]

This PR will be closed in a week because of 2 weeks of no activity.

[github-actions]

This PR was closed because it has been stalled for 2 week with no activity.