Most code search systems rely on dense embeddings. In this work, we release SPLADE-Code, learned sparse retrieval models for code retrieval, with strong generalization, high interpretability, compatibility with inverted indexes, and working across 20+ programming languages.
On the Challenges and Opportunities of Learned Sparse Retrieval for Code
@simon_lupart et al. at Naver Labs introduce a learned sparse retrieval model family for code search (600M–8B params).
📝 arxiv.org/abs/2603.22008
🤗 huggingface.co/naver/splade-c…
