[sparseml.transformers.export_onnx] add --opset flag (#1768)

src/sparseml/transformers/export.py

@@ -84,6 +84,7 @@
 from transformers.tokenization_utils_base import PaddingStrategy
 
 from sparseml.optim import parse_recipe_variables
+from sparseml.pytorch.opset import TORCH_DEFAULT_ONNX_OPSET
 from sparseml.pytorch.optim import ScheduledModifierManager
 from sparseml.pytorch.utils import export_onnx
 from sparseml.transformers.sparsification import Trainer
@@ -245,6 +246,7 @@ def export_transformer_to_onnx(
     trust_remote_code: bool = False,
     data_args: Optional[Union[Dict[str, Any], str]] = None,
     one_shot: Optional[str] = None,
+    opset: int = TORCH_DEFAULT_ONNX_OPSET,
 ) -> str:
     """
     Exports the saved transformers file to ONNX at batch size 1 using
@@ -266,6 +268,7 @@ def export_transformer_to_onnx(
     :param data_args: additional args to instantiate a `DataTrainingArguments`
         instance for exporting samples
     :param one_shot: one shot recipe to be applied before exporting model
+    :param opset: ONNX opset to export with
     :return: path to the exported ONNX file
     """
     task = task.replace("_", "-").replace(" ", "-")
@@ -398,6 +401,7 @@ def export_transformer_to_onnx(
         inputs,
         onnx_file_path,
         convert_qat=convert_qat,
+        opset=opset,
         **kwargs,
     )
     _LOGGER.info(f"ONNX exported to {onnx_file_path}")
@@ -562,6 +566,12 @@ def _parse_args() -> argparse.Namespace:
         action="store_true",
         help=("Set flag to allow custom models in HF-transformers"),
     )
+    parser.add_argument(
+        "--opset",
+        type=int,
+        default=TORCH_DEFAULT_ONNX_OPSET,
+        help=f"ONNX opset to export with, default: {TORCH_DEFAULT_ONNX_OPSET}",
+    )
 
     return parser.parse_args()
 
@@ -577,6 +587,7 @@ def export(
     trust_remote_code: bool = False,
     data_args: Optional[str] = None,
     one_shot: Optional[str] = None,
+    opset: int = TORCH_DEFAULT_ONNX_OPSET,
 ):
     if os.path.exists(model_path):
         # expand to absolute path to support downstream logic
@@ -592,6 +603,7 @@ def export(
         trust_remote_code=trust_remote_code,
         data_args=data_args,
         one_shot=one_shot,
+        opset=opset,
     )
 
     deployment_folder_dir = create_deployment_folder(
@@ -616,6 +628,7 @@ def main():
         trust_remote_code=args.trust_remote_code,
         data_args=args.data_args,
         one_shot=args.one_shot,
+        opset=args.opset,
     )