Update docstring

Co-authored-by: Gwena Cunha <4861122+gcunhase@users.noreply.github.com>
Signed-off-by: Gal Hubara-Agam <96368689+galagam@users.noreply.github.com>

Author: galagam

modelopt/onnx/autocast/convert.py

@@ -220,7 +220,7 @@ def convert_to_f16(
                                   infer_shapes. This is a workaround (WAR) when only type inference is
                                   needed without shape inference. Default: False.
         opset: Target ONNX opset version. If None, uses default minimum opset based on precision type
-               (22 for bf16, 13 for fp16) and Q/DQ node requirements. The opset may be automatically
+               (22 for bf16, 19 for fp16) and Q/DQ node requirements. The opset may be automatically
                increased if Q/DQ nodes in the model require a higher version (e.g., FP8 requires 19,
                INT4 requires 21, NVFP4 requires 23).
     """