Modelpublic

yjoonjang/splade-cocondenser-ensemble-self-linear-merge

sourceHugging Faceupdated 1mo agoView on Hugging Face

0likes24downloads

Model Card

splade-cocondenser-ensemble-self-linear-merge

A demonstration model produced by the native model-merging feature (SparseEncoder.merge) added to Sentence Transformers. It is a showcase merge, not a task-tuned release.

It merges these two checkpoints with the linear method (weights [0.5, 0.5]):

How it was created

python

from sentence_transformers import SparseEncoder

merged = SparseEncoder.merge(
    models=["naver/splade-cocondenser-ensembledistil", "naver/splade-cocondenser-selfdistil"],
    weights=[0.5, 0.5],
    method="linear",
    output_path="splade-cocondenser-ensemble-self-linear-merge",
    dtype="float16",
)

Usage

python

from sentence_transformers import SparseEncoder

model = SparseEncoder("yjoonjang/splade-cocondenser-ensemble-self-linear-merge")
emb = model.encode(["What is sparse retrieval?",
                    "SPLADE produces sparse lexical embeddings."])
print(model.similarity(emb, emb))

License

This is a derivative of the two base models above; their licenses apply. See each base model's card for terms (note that some bases — e.g. the SPLADE cocondenser models — are non-commercial).