Open aturker-synnada opened 2 days ago
The current Linear model weight shape is transposed.
Aligning the Linear layer's weight shape with other frameworks will make it easier to load pre-trained weights from those frameworks.
This change will also impact LSTM and RNN layers.
Feature Request
Describe the Feature
The current Linear model weight shape is transposed.
Motivation
Aligning the Linear layer's weight shape with other frameworks will make it easier to load pre-trained weights from those frameworks.
Additional Context
This change will also impact LSTM and RNN layers.