SPLADE - The First Search Model to Beat BM25

Explore the latest advancements in AI-powered search with this informative video on SPLADE, the first search model to outperform BM25. Dive into the world of sparse and dense vector search, comparing their advantages and limitations. Learn how SPLADE, a cutting-edge sparse embedding model, addresses the shortcomings of traditional methods like TF-IDF and BM25. Discover how SPLADE can be used alongside dense embedding models for optimal results. Gain insights into the inner workings of SPLADE, including its use of transformers and masked language modeling. Understand the vocabulary mismatch problem and how SPLADE tackles it. Get hands-on with practical implementation examples using Python, PyTorch, and Hugging Face. Explore the Naver SPLADE library and contemplate the future of vector search in this comprehensive 29-minute tutorial.

Sparse and dense vector search
Comparing sparse vs. dense vectors
Using sparse and dense together
What is SPLADE?
Vocabulary mismatch problem
How SPLADE works transformers 101
Masked language modeling MLM
How SPLADE builds embeddings with MLM
Where SPLADE doesn't work so well
Implementing SPLADE in Python
SPLADE with PyTorch and Hugging Face
Using the Naver SPLADE library
What's next for vector search?