Fix the onnx checker to use model path when model size > 2gib (#502)

## What does this PR do?

**Type of change:** Bug Fix

**Overview:** There was a failure during quantization for fp32 whisper
large model.
`Error:
ValueError: This protobuf of onnx model is too large (>2GiB). Call
check_model with model path instead.`

Essentially the bug was that while using the checker, we can only give
model object as input if model size < 2gb, otherwise the path to model
needs to be given as input. So, I changed the input given to the checker
according to size of model in trt_utils.py.

## Testing
Tried quantizing after the fix. The quantization is working after
applying the fix with no errors.

Signed-off-by: Hrishith Thadicherla <[email protected]>

Author: hthadicherla

modelopt/onnx/trt_utils.py

@@ -335,7 +335,13 @@ def load_onnx_model(
             intermediate_generated_files.append(ir_version_onnx_path)
 
     # Check that the model is valid
-    onnx.checker.check_model(onnx_model)
+    if use_external_data_format:
+        # For large models, use the file path to avoid protobuf size limitation
+        model_path_to_check = ir_version_onnx_path or static_shaped_onnx_path or onnx_path
+        onnx.checker.check_model(model_path_to_check)
+    else:
+        # For smaller models, checking the model object is fine
+        onnx.checker.check_model(onnx_model)
 
     return (
         onnx_model,