Didan Deng

Hi, I am curious about the parameter size and FLOPs (Floating Points Operations) of adanet ensemble model.

Any hint on resume training (load state dict) for Orthogonal module?

1

Hi, authors. Thanks for providing this repo. I'm currently using the Orthogonal module and define it as part of my model weights. When I tried to resume training from a...

Could you give the code for Theoretical Insights from a two-layer Linear Model

1

I have read your paper and found it quite impressive, especially for your comparison results in Figure 6. This should be a regression task with a two-layer linear model. Unfortunately,...

Mindcv 实现的[GPSA layer代码](https://github.com/mindspore-lab/mindcv/blob/20d54a9f383a2332eb344cee748be63b0dedf437/mindcv/models/convit.py#L68) 中`get_attention` 函数没有对`attn` 进行normalization: ``` def get_attention(self, x: Tensor) -> Tensor: B, N, C = x.shape q = ops.reshape(self.q(x), (B, N, self.num_heads, C // self.num_heads)) q = ops.transpose(q,...

**Problem** During validation, if I pass a non-existent checkpoint_path to the validate.py, it won't throw an error or give a warning. The model has the initialized weights, and the validation...