splade-distilbert-base-uncased-gooaq-matryoshka-hard: An efficient model for semantic search
1 min read
This model is a SPLADE Sparse Encoder built on distilbert/distilbert-base-uncased. It maps text into a sparse vector space with a dimensionality of 30522 and is used for semantic search and document retrieval. Thanks to its architecture, the model efficiently finds relevant information based on the meaning of a query rather than just exact keywords. This tool is useful for developers and researchers working on search engines and text processing.
Key features
- Built on the efficient
distilbert/distilbert-base-uncasedencoder. - Supports a maximum sequence length of 256 tokens.
- Maps texts into a sparse vector space of 30,522 dimensions.
- Uses a dot product similarity function.
What you can do
- Set up semantic search for a knowledge base or documentation in your project.
- Perform sparse document retrieval using text queries.
Example
from sentence_transformers import SparseEncoder
model = SparseEncoder("tomaarsen/splade-distilbert-base-uncased-gooaq-matryoshka-hard")
sentences = [
'what language is irog ko?',
'Filipino / Tagalog language translation for the meaning of the word irog.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
Read more
DetailsBenchmarks: average dot_ndcg@10 for NanoBEIR is 0.4955.
Limitations: maximum sequence length is 256 tokens.
License: apache-2.0.










Comments