Is there a recommended approach to extract embeddings for the starting and ending regions?

[santule]

The input sequence length of 196,608 bp is given as the input sequence to the network. Out of which the embeddings are extracted for the middle 114,688 bp region mapped to 876. That means that extracting the embeddings from region 0 to 196,608 would miss out on the starting 40,960 bp embeddings. So how should we extract the embeddings for region 0 in the input sequence? Is there a recommended approach.

Thanks