Dense retrieval models may be used in search and recommendation engines to enable low latency retrieval of candidates with nearest neighbor search. Luyu Gao and Xueguang Ma have OSS’ed Tevatron, a set of command-line tools to develop and test dense retrievers with Transformer models. Tevatron, which is designed to support retrieval research, makes it easier to train/encode dense retrievers and implement dense index search with FAISS (an ANN framework). The toolkit also includes flexible and extendable PyTorch retriever models and a highly efficient Trainer based on HuggingFace Trainer.