Revert 'reduce_avg' attribute from 'dim_arg' to 'dim' and update op_mapping

[mstojkovicTT]

Problem description

The previous change in PR #1101 modified the named attribute of reduce_avg from dim to dim_arg to align with tt-mlir naming conventions. However, this change has been identified as problematic and needs to be reverted.

What's changed

This PR reverts the changes introduced in PR #1101, restoring the named attribute of reduce_avg back to dim. Additionally, the op_mapping has been updated from dim to dim_arg to ensure compatibility with ttir.

Checklist

• New/Existing tests provide coverage for changes

[github-actions]

	Tests	Passed ✅	Skipped ⚠️	Failed
TT-Forge-FE Tests	501 ran	436 passed	65 skipped	0 failed

Test	Result
No test annotations available

[github-actions]

	Tests	Passed ✅	Skipped ⚠️	Failed
TT-Forge-FE Tests	560 ran	484 passed	76 skipped	0 failed

Test	Result
No test annotations available

[nvukobratTT]

For clarify, can we update attribute names in the op definition as well? E.g.

forge/forge/op/reduce.py

[mstojkovicTT]

For clarify, can we update attribute names in the op definition as well? E.g.

Correct me if I am wrong, but ReduceAvg seems to have correct namings (dim was not changed from dim_arg to dim here)

[github-actions]

	Tests	Passed ✅	Skipped ⚠️	Failed
TT-Forge-FE Tests	555 ran	474 passed	81 skipped	0 failed

Test	Result
No test annotations available

[github-actions]

	Tests	Passed ✅	Skipped ⚠️	Failed
TT-Forge-FE Tests	496 ran	436 passed	60 skipped	0 failed

Test	Result
No test annotations available

[nvukobratTT]

For clarify, can we update attribute names in the op definition as well? E.g.

Correct me if I am wrong, but ReduceAvg seems to have correct namings (dim was not changed from dim_arg to dim here)

Did you spot this line: dim_arg=[dim] ?

def ReduceAvg(name: str, operandA: Tensor, dim: int, keep_dim: bool = True) -> Tensor:
    """
    Reduce by averaging along the given dimension

    Parameters
    ----------
    name: str
        Op name, unique to the module, or leave blank to autoset

    operandA: Tensor
        First operand

    dim: int
        Dimension along which to reduce. A positive number 0 - 3 or negative from -1 to -4.

    Returns
    -------
    Tensor
        Forge tensor
    """

    assert (dim >= -4) and (dim <= 3)
    # if dim < 0:
    #     dim += 4

    return op("reduce_avg", name, operandA, attrs=(dim, keep_dim), dim_arg=[dim], keep_dim=keep_dim).get_tensor()

[mstojkovicTT]

For clarify, can we update attribute names in the op definition as well? E.g.

Correct me if I am wrong, but ReduceAvg seems to have correct namings (dim was not changed from dim_arg to dim here)

Did you spot this line: dim_arg=[dim] ?

def ReduceAvg(name: str, operandA: Tensor, dim: int, keep_dim: bool = True) -> Tensor:
    """
    Reduce by averaging along the given dimension

    Parameters
    ----------
    name: str
        Op name, unique to the module, or leave blank to autoset

    operandA: Tensor
        First operand

    dim: int
        Dimension along which to reduce. A positive number 0 - 3 or negative from -1 to -4.

    Returns
    -------
    Tensor
        Forge tensor
    """

    assert (dim >= -4) and (dim <= 3)
    # if dim < 0:
    #     dim += 4

    return op("reduce_avg", name, operandA, attrs=(dim, keep_dim), dim_arg=[dim], keep_dim=keep_dim).get_tensor()

Right, i missed that, thanks!

[github-actions]

	Tests	Passed ✅	Skipped ⚠️	Failed
TT-Forge-FE Tests	496 ran	436 passed	60 skipped	0 failed

Test	Result
No test annotations available

[github-actions]

	Tests	Passed ✅	Skipped ⚠️	Failed
TT-Forge-FE Tests	555 ran	474 passed	81 skipped	0 failed

Test	Result
No test annotations available