NeuronBlocks is a PyTorch-based toolkit that abstracts commonly used neural network components into reusable blocks and provides pre-built NLP model configurations as JSON files. It supports tasks including sentence classification, sentiment analysis, question answering matching, textual entailment, slot tagging, and machine reading comprehension. The toolkit also includes knowledge distillation capabilities for model compression.