thuml/Large-Time-Series-Model

为什么Timer是Decoder-only的?

Opened this issue · 0 comments

# Transformer Blocks dec_out, attns = self.decoder(dec_in) # [B * M, N, D]
forecast函数中并没有传入任何mask,而backbone中的定义也是self.decoder = Encoder(),Encoder()的forward函数默认接受的是None作为mask。
看起来完全是encoder-only的呀?