Quantization.MXNet

Quantization.MXNet copied to clipboard

Hi everyone, when i'm trying to launch provided in readme python command, i'm catching error (mx_quant) root@226a37f8fc1a:/workspace/quantization/quant_mxnet/examples# python simulate_quantization.py --model=mobilenet1.0 ************************* Settings ************************* model : mobilenet1.0 print_model : False list_models...

L-ED

一些问题...

4

您好，我正在阅读并测试您的代码，我有一些疑惑和问题想要请教您。 我想复现white paper里mobileNetv2的per-channel量化，结果在imageNet val上面测试结果只有0.09%，说明per-channel的方案完全没有效果，请问您有什么头绪吗？

evi-Genius

How to avoid "self.current_input_max = F.max(F.abs(x), axis=(1, 2, 3)).mean().asscalar()" to be a node in graph ?

12

"self.current_input_max = F.max(F.abs(x), axis=(1, 2, 3)).mean().asscalar()" inside "def _conv2d_forward", this will create symbol node, and should cause error because asscalar() is not supported in symbol. How to avoid "self.current_input_max =...

yang2640