Good to see actual open-source models instead of these so-called "open weight" models which are still just inscrutable binary blobs that don't even provide a high-level catalog or really any description whatsoever of what went into the training data or the methods and tools used to train them.