Only convert inputs to FP16 when FP16 stage is used

groq / mlagility

Machine Learning Agility (MLAgility) benchmark and benchmarking tools

MIT License

38 stars 11 forks source link

Only convert inputs to FP16 when FP16 stage is used #336

Closed danielholanda closed 1 year ago

danielholanda commented 1 year ago

Issue

src/onnxflow/common/tensor_helpers.py:30: save_inputs() converts inputs to FP16 even when the FP16 is not part of the user-provided custom sequence.

Task

Ensure that inputs are only converted to FP16 when the FP16 stage is used.