quark.onnx.tools.convert_fp16_to_fp32

Convert tensor float16 type in the ONNX ModelProto input to tensor float.

param model:

ONNX ModelProto object

param disable_shape_infer:

Type/shape information is needed for conversion to work. Set to True only if the model already has type/shape information for all tensors.

return:

converted ONNX ModelProto object

Examples:

Example 1: Convert ONNX ModelProto object:
import float16
new_onnx_model = float16.convert_float16_to_float(onnx_model)

Example 2: Convert ONNX model binary file:
import onnx
import float16
onnx_model = onnx.load_model('model.onnx')
new_onnx_model = float16.convert_float16_to_float(onnx_model)
onnx.save_model(new_onnx_model, 'new_model.onnx')

Use the convert_float16_to_float.py to convert a float16 model to a float32 model:

` python convert_fp16_to_fp32.py --input $FLOAT_16_ONNX_MODEL_PATH --output $FLOAT_32_ONNX_MODEL_PATH `

The conversion from float16 models to float32 models may result in the generation of unnecessary operations such as casts in the model. It is recommended to use onnx-simplifier to remove these redundant nodes.