提交 ae2f00cc 编写于 作者: Z Zhen Wang 提交者: whs

fix the missing links. (#3223)

上级 346e1343
...@@ -74,11 +74,11 @@ $q = scale * r + b$ ...@@ -74,11 +74,11 @@ $q = scale * r + b$
对于通用矩阵乘法(`GEMM`),输入$X$和权重$W$的量化操作可被表述为如下过程: 对于通用矩阵乘法(`GEMM`),输入$X$和权重$W$的量化操作可被表述为如下过程:
$$ X_q = \left \lfloor \frac{X}{X_m} * (n - 1) \right \rceil $$ $$ W_q = \left \lfloor \frac{W}{W_m} * (n - 1) \right \rceil $$ $$ X_q = \left \lfloor \frac{X}{X_m} * (n - 1) \right \rceil $$ $$ W_q = \left \lfloor \frac{W}{W_m} * (n - 1) \right \rceil $$
执行通用矩阵乘法: 执行通用矩阵乘法:
$$ Y = X_q * W_q $$ $$ Y_q = X_q * W_q $$
反量化$Y$: 对量化乘积结果$Yq$进行反量化:
$$ $$
\begin{align} \begin{align}
Y_{dq} = \frac{Y}{(n - 1) * (n - 1)} * X_m * W_m \ Y_{dq} = \frac{Y_q}{(n - 1) * (n - 1)} * X_m * W_m \
=\frac{X_q * W_q}{(n - 1) * (n - 1)} * X_m * W_m \ =\frac{X_q * W_q}{(n - 1) * (n - 1)} * X_m * W_m \
=(\frac{X_q}{n - 1} * X_m) * (\frac{W_q}{n - 1} * W_m) \ =(\frac{X_q}{n - 1} * X_m) * (\frac{W_q}{n - 1} * W_m) \
\end{align} \end{align}
......
#!/usr/bin/env bash #!/usr/bin/env bash
# download pretrain model # download pretrain model
root_url="http://paddle-imagenet-models-name.bj.bcebos.com" root_url="https://paddle-inference-dist.bj.bcebos.com/int8/pretrain"
MobileNetV1="MobileNetV1_pretrained.zip" MobileNetV1="MobileNetV1_pretrained.zip"
ResNet50="ResNet50_pretrained.zip" ResNet50="ResNet50_pretrained.zip"
GoogleNet="GoogleNet_pretrained.tar" GoogleNet="GoogleNet_pretrained.tar"
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册