ONEforALL-S003 / ONE

On-device Neural Engine / q-implant
Other
1 stars 2 forks source link

[q-implant] Quantization Parameter sharing for operators which are ad… #17

Open ssafyWooJeong opened 1 year ago

ssafyWooJeong commented 1 year ago

…ded while model conversion.

Operators(such as transpose, reshape and etc.) can be added while model conversion. Those operators are not present at original model, so corresponding quantization parameter could not exist on qparam.json. For those operators, share predecessor's quantization parameter to make appropriate graph.

ONE-DCO-1.0-Signed-off-by: WooJeong Jeong wj6349@gmail.com