Welcome to Software Development on Codidact!
Will you help us build our independent community of developers helping developers? We're small and trying to grow. We welcome questions about all aspects of software development, from design to code to QA and more. Got questions? Got answers? Got code you'd like someone to review? Please join us.
Post History
In encoder-decoder models for machine translation, it's not uncommon for the bos_token_id (beginning of sequence) and eos_token_id (end of sequence) to be set to the same value. This approach is m...
#2: Post edited
by
Willyilderman
·
2025-05-12T14:06:26Z (about 17 hours ago)
- In encoder-decoder models for machine translation, it's not uncommon for the bos_token_id (beginning of sequence) and eos_token_id (end of sequence) to be set to the same value.
- This approach is mostly fine in decoder-only or decoder-side generation tasks, but in more complex pipelines or for clarity, keeping bos_token_id and eos_token_id separate is generally preferred.
I believe so.
- In encoder-decoder models for machine translation, it's not uncommon for the bos_token_id (beginning of sequence) and eos_token_id (end of sequence) to be set to the same value.
- This approach is mostly fine in decoder-only or decoder-side generation tasks, but in more complex pipelines or for clarity, keeping bos_token_id and eos_token_id separate is generally preferred.
- I believe so.
- Edit by: "[Block Blast](https://blockblast.org)"
#1: Initial revision
by
Willyilderman
·
2025-05-12T14:05:42Z (about 17 hours ago)
In encoder-decoder models for machine translation, it's not uncommon for the bos_token_id (beginning of sequence) and eos_token_id (end of sequence) to be set to the same value. This approach is mostly fine in decoder-only or decoder-side generation tasks, but in more complex pipelines or for clarity, keeping bos_token_id and eos_token_id separate is generally preferred. I believe so.