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
import paddle
import paddle.nn as nn
import paddle.nn.functional as F
nn.initializer.set_global_initializer(weight_init=None,bias_init=nn.initializer.Constant(0.))
lstm=nn.LSTM(input_size=4,hidden_size=3,num_layers=2,proj_size=2)
for a in lstm.named_parameters():
print(a)
------------------ 原始邮件 ------------------
发件人: ***@***.***>;
发送时间: 2024年12月18日(星期三) 下午2:50
收件人: ***@***.***>;
抄送: ***@***.***>; ***@***.***>;
主题: Re: [PaddlePaddle/Paddle] 3.0bata版本的lstm参数输出有误 (Issue #70245)
你好,问题已经反馈给该API负责同学排查了,有结论后尽快同步在这里呢
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you authored the thread.Message ID: ***@***.***>
bug描述 Describe the Bug
问题1:lstm输出参数时weight_ih_l1的形状为[3,2],bias_ih_l1的形状为[12,2],这明显不对啊,偏置元素比weight还多。
问题2:lstm的1.cell.weight_ho权重无法访问,我用的是lstm.1.cell.weight_ho访问的,但是报错了。用这个方法访问别的lstm权重偏置就可以唯独1.cell.weight_ho这个不行。
问题3:1.cell.weight_ho中的1和ho代表什么意思。没想明白,api文档也没说过这个问题。
复现代码如下:
其他补充信息 Additional Supplementary Information
补充:如果可以的话,能不能告知下大概多久能修复此问题。论文中使用了这个api,拖太久的话我就删掉这个proj_size参数了。感谢技术大佬。
The text was updated successfully, but these errors were encountered: