API to measure MSE for target quantizers (#940)

## What does this PR do? **Type of change:** new feature ?  **Overview:** add an API to measure MSE for target quantizers given a forward loop ## Usage  ```python # 1. Quantize the model as usual model = mtq.quantize(model, quant_cfg, forward_loop) # 2. Compute MSE for all quantizers mse = mtq.compute_quantization_mse(model, forward_loop) # 3. Print the top-5 noisiest quantizers for name, err in sorted(mse.items(), key=lambda x: -x[1])[:5]: print(f"{name}: {err:.4e}") ``` ## Testing  Unit test and test with HF PTQ ## Before your PR is "*Ready for review*"  - **Make sure you read and follow [Contributor guidelines](https://github.com/NVIDIA/Model-Optimizer/blob/main/CONTRIBUTING.md)** and your commits are signed. - **Is this change backward compatible?**: Yes/No  - **Did you write any new necessary tests?**: Yes/No - **Did you add or update any necessary documentation?**: Yes/No - **Did you update [Changelog](https://github.com/NVIDIA/Model-Optimizer/blob/main/CHANGELOG.rst)?**: Yes/No  ## Additional Information   ## Summary by CodeRabbit * **New Features** * Added an API to measure per-quantizer mean-squared error (MSE) between original and fake-quantized tensors; supports wildcard and callable filtering, skips disabled/non-fake-quant quantizers, and runs safely under no-grad. * **Tests** * Added comprehensive tests for MSE validity, pattern and callable filtering, union behavior, exclusion of disabled quantizers, preservation of model state, and forward-hook cleanup. * **Documentation** * Updated changelog to document the new MSE measurement API.  --------- Signed-off-by: weimingc <17592131+meenchen@users.noreply.github.com> Signed-off-by: Wei-Ming Chen <17592131+meenchen@users.noreply.github.com>
2026-10-02 03:14:52 +08:00 · 2026-03-06 22:02:49 +00:00
parent be6dfad920
commit 22423041f7
3 changed files with 219 additions and 0 deletions
@@ -19,6 +19,7 @@ NVIDIA Model Optimizer Changelog
 - Enable PTQ workflow for Qwen3.5 MoE models.
 - Add ``nvfp4_omlp_only`` quantization format for NVFP4 quantization. This is similar to ``nvfp4_mlp_only`` but also quantizes the output projection layer in attention.
 - ``pass_through_bwd`` in the quantization config is now default to True. Please set it to False if you want to use STE with zeroed outlier gradients for potentially better QAT accuracy.
+- Add :meth:`compute_quantization_mse <modelopt.torch.quantization.model_quant.compute_quantization_mse>` API to measure per-quantizer mean-squared quantization error, with flexible wildcard and callable filtering.

 **Misc**

@@ -41,6 +41,7 @@ from .nn import QuantModule, TensorQuantizer
 __all__ = [
    "auto_quantize",
    "calibrate",
+    "compute_quantization_mse",
    "disable_quantizer",
    "enable_quantizer",
    "fold_weight",
@@ -535,3 +536,79 @@ def fold_weight(model: nn.Module, keep_attrs: bool = False):
    for name, module in model.named_modules():
        if isinstance(module, QuantModule):
            module.fold_weight(keep_attrs)
+
+
+@torch.no_grad()
+def compute_quantization_mse(
+    model: nn.Module,
+    forward_loop: ForwardLoop,
+    wildcards: str | Callable | list[str | Callable] = "*",
+) -> dict[str, float]:
+    """Compute the mean-squared quantization error for selected quantizers.
+
+    Runs ``forward_loop`` through the model while recording, for every matching
+    :class:`TensorQuantizer`, the MSE between the original float tensor and
+    its fake-quantized (Q→DQ) counterpart. Values are averaged over all
+    calibration batches.
+
+    Args:
+        model: A quantized model (output of :func:`quantize`).
+        forward_loop: Callable that takes ``model`` and runs data through it.
+        wildcards: One or more fnmatch glob patterns (or callable filters)
+            matched against :class:`TensorQuantizer` module names in
+            ``model.named_modules()``.  Follows the same convention as
+            ``quant_cfg`` wildcard keys.  Defaults to ``"*"`` (all quantizers).
+
+    Returns:
+        A dict mapping each matched quantizer's fully-qualified name to its
+        mean MSE (float).  Quantizers that are disabled or not in fake-quant
+        mode are skipped and absent from the output.
+
+    Example::
+
+        mse = mtq.compute_quantization_mse(
+            model,
+            forward_loop,
+            wildcards=["*k_bmm_quantizer", "*v_bmm_quantizer"],
+        )
+        for name, err in sorted(mse.items()):
+            print(f"{name}: {err:.4e}")
+    """
+    if not isinstance(wildcards, list):
+        wildcards = [wildcards]
+
+    def _matches(name: str) -> bool:
+        return any(fnmatch.fnmatch(name, w) if isinstance(w, str) else w(name) for w in wildcards)
+
+    accumulators: dict[str, dict] = {}  # name -> {"sum": float, "count": int}
+    hooks = []
+
+    for name, module in model.named_modules():
+        if not isinstance(module, TensorQuantizer):
+            continue
+        if not _matches(name):
+            continue
+        if not (module._if_quant and module._fake_quant) or module._disabled:
+            continue
+        accumulators[name] = {"sum": 0.0, "count": 0}
+
+        def _make_hook(acc):
+            def hook(mod, inp, out):
+                original = inp[0].detach().float()
+                quantized = out.detach().float()
+                acc["sum"] += torch.mean((original - quantized) ** 2).item()
+                acc["count"] += 1
+
+            return hook
+
+        hooks.append(module.register_forward_hook(_make_hook(accumulators[name])))
+
+    try:
+        forward_loop(model)
+    finally:
+        for h in hooks:
+            h.remove()
+
+    return {
+        name: acc["sum"] / acc["count"] for name, acc in accumulators.items() if acc["count"] > 0
+    }
@@ -0,0 +1,141 @@
+# SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Unit tests for mtq.compute_quantization_mse()."""
+
+import torch
+from _test_utils.torch.quantization.models import SimpleLinear
+
+import modelopt.torch.quantization as mtq
+from modelopt.torch.quantization.nn import TensorQuantizer
+
+INT8_CFG = {
+    "quant_cfg": {
+        "*weight_quantizer": {"num_bits": 8, "axis": 0},
+        "*input_quantizer": {"num_bits": 8, "axis": None},
+    },
+    "algorithm": "max",
+}
+
+
+def _make_quantized_model():
+    model = SimpleLinear()
+    calib_data = [model.get_input() for _ in range(4)]
+
+    def forward_loop(m):
+        for batch in calib_data:
+            m(batch)
+
+    mtq.quantize(model, INT8_CFG, forward_loop)
+    return model, forward_loop
+
+
+class TestComputeQuantizationMse:
+    def test_returns_nonnegative_values(self):
+        """MSE values must be >= 0 for all quantizers."""
+        model, forward_loop = _make_quantized_model()
+        mse = mtq.compute_quantization_mse(model, forward_loop)
+        assert len(mse) > 0
+        assert all(v >= 0.0 for v in mse.values())
+
+    def test_wildcard_star_covers_all_enabled_fake_quant(self):
+        """Default wildcard '*' should return an entry for every enabled fake-quant quantizer."""
+        model, forward_loop = _make_quantized_model()
+        mse = mtq.compute_quantization_mse(model, forward_loop, wildcards="*")
+
+        expected_names = {
+            name
+            for name, module in model.named_modules()
+            if isinstance(module, TensorQuantizer)
+            and module._if_quant
+            and module._fake_quant
+            and not module._disabled
+        }
+        assert set(mse.keys()) == expected_names
+
+    def test_wildcard_filters_by_suffix(self):
+        """A suffix pattern should restrict results to matching quantizer names."""
+        model, forward_loop = _make_quantized_model()
+        mse = mtq.compute_quantization_mse(model, forward_loop, wildcards="*weight_quantizer")
+        assert len(mse) > 0
+        assert all("weight_quantizer" in k for k in mse)
+        # No input quantizers should appear
+        assert not any("input_quantizer" in k for k in mse)
+
+    def test_list_of_wildcards(self):
+        """A list of patterns should return the union of matched quantizers."""
+        model, forward_loop = _make_quantized_model()
+        mse_weight = mtq.compute_quantization_mse(
+            model, forward_loop, wildcards="*weight_quantizer"
+        )
+        mse_input = mtq.compute_quantization_mse(model, forward_loop, wildcards="*input_quantizer")
+        mse_both = mtq.compute_quantization_mse(
+            model, forward_loop, wildcards=["*weight_quantizer", "*input_quantizer"]
+        )
+        assert set(mse_both.keys()) == set(mse_weight.keys()) | set(mse_input.keys())
+
+    def test_callable_filter(self):
+        """A callable wildcard should select quantizers by arbitrary predicate."""
+        model, forward_loop = _make_quantized_model()
+        # Pick only quantizers belonging to the first linear layer (net.0)
+        mse = mtq.compute_quantization_mse(model, forward_loop, wildcards=lambda n: "net.0" in n)
+        assert len(mse) > 0
+        assert all("net.0" in k for k in mse)
+
+    def test_disabled_quantizer_absent_from_result(self):
+        """A quantizer disabled after calibration must not appear in the output."""
+        model, forward_loop = _make_quantized_model()
+
+        # Disable one quantizer and record its name
+        disabled_name = None
+        for name, module in model.named_modules():
+            if isinstance(module, TensorQuantizer) and module._if_quant and module._fake_quant:
+                module.disable()
+                disabled_name = name
+                break
+
+        assert disabled_name is not None, "No enabled quantizer found to disable"
+
+        mse = mtq.compute_quantization_mse(model, forward_loop)
+        assert disabled_name not in mse
+
+    def test_no_matching_wildcard_returns_empty_dict(self):
+        """A pattern that matches nothing should return an empty dict."""
+        model, forward_loop = _make_quantized_model()
+        mse = mtq.compute_quantization_mse(
+            model, forward_loop, wildcards="*nonexistent_quantizer_xyz*"
+        )
+        assert mse == {}
+
+    def test_does_not_modify_model_parameters(self):
+        """Running MSE measurement must leave model weights unchanged."""
+        model, forward_loop = _make_quantized_model()
+        params_before = {k: v.clone() for k, v in model.named_parameters()}
+        mtq.compute_quantization_mse(model, forward_loop)
+        for k, v in model.named_parameters():
+            assert torch.equal(v, params_before[k]), f"Parameter {k} was modified"
+
+    def test_hooks_removed_after_call(self):
+        """All forward hooks registered during the call must be cleaned up."""
+        model, forward_loop = _make_quantized_model()
+
+        hooks_before = sum(
+            len(m._forward_hooks) for m in model.modules() if isinstance(m, TensorQuantizer)
+        )
+        mtq.compute_quantization_mse(model, forward_loop)
+        hooks_after = sum(
+            len(m._forward_hooks) for m in model.modules() if isinstance(m, TensorQuantizer)
+        )
+        assert hooks_after == hooks_before