Type And Shape Inference Engine With Partial Evaluation And Data Propagation

Open Neural Network Exchange

Unique: Implements bidirectional shape inference (forward and backward propagation) combined with partial evaluation of constant subgraphs; uses operator schema registry to apply type-specific inference rules (e.g., broadcasting rules for element-wise ops, reduction rules for aggregation ops) without executing the model

vs others: More comprehensive than TensorFlow's shape inference because it handles operator-specific semantics through schema-driven rules; faster than PyTorch's symbolic shape tracing because it doesn't require model execution