MXNet.<p>It is a popular library, but I still feel that it's quite underappreciated.<p>It is constantly living in the shadows of TensorFlow and PyTorch. MXNet has a clean API (via Gluon) and has nailed the hybrid eager/graph execution model better than either of its competitors.<p>I have come across multiple ML practitioners who have never heard of it.
There are almost too many to track but, every now and then, one comes up that blows me away.<p>I have recently had a lot of success with lucene/solr and wish I had used it for years!<p>Which have you found useful and want to help bring attention to?<p>Even if relatively large but you want to emphasize, give a shout!