[OV] Optimized compression to MXFP4 data type

[nikita-savelyevv]

Changes

Added optimized compression to MXFP4 data type for OpenVINO backend.

Model	Memory Before (MiB)	Memory After (MiB)	Time Before (sec)	Time After (sec)
llama-3.2-1b bf16	2778.55	2548.37 (-8.29%)	66.33	24.95 (-62.40%)
llama-3.2-1b fp16	3434.61	2963.03 (-13.73%)	62.12	24.85 (-59.98%)
llama-3.2-1b fp32	2041.79	1576.43 (-22.81%)	62.72	25.85 (-58.77%)
phi4-mini bf16	6384.81	5725.66 (-10.33%)	197.36	66.22 (-66.44%)
phi4-mini fp16	8863.53	8375.75 (-5.51%)	195.85	66.93 (-65.82%)
phi4-mini fp32	4406.82	3897.91 (-11.54%)	195.25	68.83 (-64.76%)
llama-3.1-8b bf16	7297.72	6096.06 (-16.46%)	413.86	135.25 (-67.32%)
llama-3.1-8b fp16	7946.93	8311.89 (+4.58%)	413.64	136.05 (-67.11%)
llama-3.1-8b fp32	7310.48	5043.89 (-30.03%)	411.45	140.66 (-65.81%)

Reason for changes

Improving user experience.

Related tickets

164717

Tests

Extended optimized compression tests.

[ljaljushkin]

what about adding small unit test for such rounding?
maybe also add some corner cases: -0.0, 0.0, more than max, less then min

@pytest.mark.parametrize(
        "center_idx,expected_idx",
        [
            (0, 1),
            (1, 1),
            (2, 3),
            (6, 7),
            (7, 7),
            (-1, -2),
        ],
    )
    def test_exact_quantile_center_values(center_idx, expected_idx, description):
        """Test that values exactly at quantile centers round to nearest even index."""
        center_val = CENTER_OF_MXFP4_QUANTILES[center_idx]
        expected_q = MXFP4_QUANTILES[expected_idx]
        
        norm_weight = Tensor(np.array([center_val], dtype=np.float32))
        result = _calculate_float_quantized_weight(norm_weight, CompressWeightsMode.MXFP4)
        
        # Verify the result matches expected quantile
        assert result.data[0] == expected_q, (
            f"{description}: Expected {expected_q}, got {result.data[0]} "
            f"for center value {center_val}"
        )

[nikita-savelyevv]

Added. Also, tests/openvino/optimized_functions/test_compression_functions.py::test_quantization_alignment fails without the highlighted if logic.

[ljaljushkin]

Suggested change

	For NF4 quantization quantizes the weights to 16 levels on [-1, 1] interval.
	TODO(nikita-savelyevv): add support for MXFP4 and MXFP8_E4M3 once ticket 164851 is resolved
	For MXFP4 quantization quantizes the weights to 16 levels on [-6, 6] interval.
	NF4 format uses 16 levels in [-1, 1] range, while MXFP4 uses 16 levels in [-6, 6].

[nikita-savelyevv]

Done

[ljaljushkin]

Suggested change

	): # native Tensorflow operaors do not guarantee to return a tensor on an initial device.
	): # native Tensorflow operators do not guarantee to return a tensor on an initial device.

[nikita-savelyevv]

Done