Stories by cubie

A Replacement for BERT

Training and Finetuning Embedding Models with Sentence Transformers v3

Embedding Quantization: 25-45x retrieval speedup, 32x or 4x less memory usage

Attention Sinks in LLMs for endless fluency