[NVBUG: 5373030] Disable the weight adjustment for int32 bias from onnxruntime (NVIDIA#510)

## What does this PR do?

**Type of change:** 
Bug Fix

**Overview:** 
- Disable the weight adjustment for int32 bias in onnxruntime by default


## Usage

```python
python -m modelopt.onnx.quantization --onnx_path=code031_gemm_batch.onnx --simplify  --calibration_eps trt --quantize_mode fp8 --disable_mha_qdq --high_precision_dtype fp16
```

## Testing
Able to quantize the code031_gemm_batch.onnx model

## Before your PR is "*Ready for review*"
<!-- If you haven't finished some of the above items you can still open
`Draft` PR. -->

- **Make sure you read and follow [Contributor
guidelines](https://github.com/NVIDIA/TensorRT-Model-Optimizer/blob/main/CONTRIBUTING.md)**
and your commits are signed.
- **Is this change backward compatible?**: Yes
- **Did you write any new necessary tests?**: No
- **Did you add or update any necessary documentation?**: Yes
- **Did you update
[Changelog](https://github.com/NVIDIA/TensorRT-Model-Optimizer/blob/main/CHANGELOG.rst)?**:
No

---------

Signed-off-by: ajrasane <[email protected]>

Author: ajrasane

modelopt/onnx/quantization/fp8.py

@@ -272,6 +272,10 @@ def quantize(
                 trt_guided_options["group_qdq_tensors"] = group_qdq_tensors
                 logger.debug(f"Grouping QDQ tensors for concat elimination: {group_qdq_tensors}")
 
+        # Add disable_int32_weight_adjustment flag to extra options
+        trt_guided_options["QDQDisableWeightAdjustForInt32Bias"] = True
+        logger.debug("Disabled weight adjustment for INT32 bias in QDQ quantization")
+
         # Create a temp file for intermediate model
         tmp_onnx_file, tmp_onnx_path = tempfile.mkstemp(suffix=".onnx")
         os.close(tmp_onnx_file)

modelopt/onnx/quantization/int8.py

@@ -237,6 +237,10 @@ def quantize(
                 trt_guided_options["group_qdq_tensors"] = group_qdq_tensors
                 logger.debug(f"Found {len(group_qdq_tensors)} tensor groups for concat elimination")
 
+        # Add disable_int32_weight_adjustment flag to extra options
+        trt_guided_options["QDQDisableWeightAdjustForInt32Bias"] = True
+        logger.debug("Disabled weight adjustment for INT32 bias in QDQ quantization")
+
         # Create a temp file for intermediate model
         tmp_onnx_file, tmp_onnx_path = tempfile.mkstemp(suffix=".onnx")
         os.close(tmp_onnx_file)

modelopt/onnx/quantization/ort_patching.py

@@ -1600,6 +1600,7 @@ def _quantize_static(
         ("TrtExtraPluginLibraryPaths", "trt_extra_plugin_lib_paths"),
         ("ExecutionProviders", "execution_providers"),
         ("group_qdq_tensors", "group_qdq_tensors"),
+        ("QDQDisableWeightAdjustForInt32Bias", "disable_int32_weight_adjustment"),
         # ==========================================================
     ]
     calib_extra_options = {