You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi. In relation to quantization, I'm wondering if there was ever any thought given to including activation statistics in the tensors in the ONNX format. I have looked at the quantization operators (although I've not found many converters that support them) but just having activation statistics would allow translators to target formats to make up their own minds on quantization. A minimum of tensor wide min/max/std/mean would be very useful. TFLite already includes this information in quantized graphs.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
-
Hi. In relation to quantization, I'm wondering if there was ever any thought given to including activation statistics in the tensors in the ONNX format. I have looked at the quantization operators (although I've not found many converters that support them) but just having activation statistics would allow translators to target formats to make up their own minds on quantization. A minimum of tensor wide min/max/std/mean would be very useful. TFLite already includes this information in quantized graphs.
Beta Was this translation helpful? Give feedback.
All reactions