quark.onnx.tools.remove_bf16_cast

Rmove bfloat16 cast ops for an onnx model.

param input_model_path:

the path of input bfloat16 quantized model with bfloat16 cast

param output_model_path:

the path of bfloat16 quantized model with no bfloat16 cast

Use the remove_bf16_cast.py to remove bfloat16 cast for a bfloat16 quantized model:

` python remove_bf16_cast.py --input $INPUT_MODEL_PATH --output $OUTPUT_MODEL_PATH `