Tag
A 30-question breakdown explaining how embeddings, vector search, and retrieval systems work, covering similarity metrics, training methods, indexing techniques, and evaluation.
A researcher shares an experimental plan for identifying causal dependencies between capability dimensions in a 31B model using contrastive targeted SFT and circuit tracing, seeking feedback on methodology and related work.
SMART is a framework that unlocks latent multi-vector capabilities in single-vector models for multimodal retrieval, improving state-of-the-art performance with reduced computational costs via contrastive training and late-interaction inference.