mirror of
https://github.com/elder-plinius/L1B3RT4S.git
synced 2026-07-09 21:23:09 +03:00
9 lines
580 B
Markdown
9 lines
580 B
Markdown
Special Tokens
|
|
|
|
1. **\<SOS> (Start of Sequence)**: Marks the beginning of a sequence for the model to start processing.
|
|
2. **\<EOS> (End of Sequence)**: Tells the model when to stop generating text or processing.
|
|
3. **\<PAD> (Padding Token)**: Pads sequences to the same length for batch processing.
|
|
4. **\<UNK> (Unknown Token)**: Represents words not in the model's vocabulary.
|
|
5. **\<MASK> (Mask Token)**: Used in tasks like predicting missing words in masked language models.
|
|
6. **\<SEP> (Separator Token)**: Separates different segments in input, like questions from context.
|