Add QDQ Translator - Githubissues

NVIDIA / Deep-Learning-Accelerator-SW

NVIDIA DLA-SW, the recipes and tools for running deep learning workloads on NVIDIA DLA cores for inference applications.

Other

178 stars 14 forks source link

Closed nvoliver closed 1 year ago

nvoliver commented 1 year ago

first public version of QDQ Translator: Translate ONNX models from QAT graphs – the result of Quantization-Aware Training, containing Quantize and DeQuantize (Q/DQ) nodes – to ONNX graphs without Q/DQ nodes, along with extracted tensor scales.
fix licenses to this repo's MIT license
replace np.bool with bool in operator/s (to support newer NumPy versions)