Maximum and Optimal Text Size That Can Be Vectror Embedded with bge-m3?
#57
by
DracoDev
- opened
I see that BAAI/bge-m3 has dimension 1024 and sequence length of 8192. For a dense vector embedding what is the maximal and optimal size(s) for text to embed with bge-m3?
The dimension of embedding is fixed be 1024.
The sequence length depends on your device and data. If your text is long, use a large sequence length (the maximal value is 8192), but it also needs more memory to process long text.