What does this PR do?

Type of change: new feature

Adds rotate.seed to TensorQuantizer rotate config. The default seed=None preserves regular normalized Hadamard rotation. Setting an integer seed enables deterministic Random Hadamard Transform signs and uses the matching inverse order for mode="rotate_back".

This is stacked on #1879.

Usage

quant_cfg = {
    "quant_cfg": [
        {
            "quantizer_name": "*weight_quantizer",
            "cfg": {
                "num_bits": 4,
                "block_sizes": {-1: 128},
                "fake_quant": True,
                "rotate": {
                    "enable": True,
                    "mode": "rotate_back",
                    "seed": 123,
                },
            },
        },
    ],
    "algorithm": "max",
}

Use "seed": None or omit seed for regular Hadamard rotation.

Testing

• pre-commit run --files CHANGELOG.rst modelopt/torch/quantization/config.py modelopt/torch/quantization/nn/functional.py modelopt/torch/quantization/nn/modules/tensor_quantizer.py tests/unit/torch/quantization/test_tensor_quant_cpu.py tests/gpu/torch/quantization/test_hadamard.py
• pytest_pwd tests/unit/torch/quantization/test_tensor_quant_cpu.py::TestQuantizerAttributeConfig::test_rotate_mode_serialization tests/unit/torch/quantization/test_tensor_quant_cpu.py::test_tensor_quantizer_rotate_back_rejects_real_quant tests/unit/torch/quantization/test_tensor_quant_cpu.py::test_tensor_quantizer_rotate_mode_can_rotate_back tests/unit/torch/quantization/test_tensor_quant_cpu.py::test_tensor_quantizer_rotate_mode_preserves_default_path -q
• ssh a8 'cd /home/scratch.akuriparambi_coreai/pr-review-branches/modelopt-tensorquantizer-random-hadamard && /home/akuriparambi/.codex/tools/pytest_pwd tests/gpu/torch/quantization/test_hadamard.py::test_hadamard_transform_seeded_inverse -q'

Before your PR is "Ready for review"

Make sure you read and follow Contributor guidelines and your commits are signed (git commit -s -S).

Make sure you read and follow the Security Best Practices (e.g. avoiding hardcoded trust_remote_code=True, torch.load(..., weights_only=False), pickle, etc.).

• Is this change backward compatible?: ✅
• If you copied code from any other sources or added a new PIP dependency, did you follow guidance in CONTRIBUTING.md: N/A
• Did you write any new necessary tests?: ✅
• Did you update Changelog?: ✅
• Did you get Claude approval on this PR?: N/A, draft PR

Additional Information

Stacked on #1879.