PyTorch-based Fabric ML framework for training, inference, export, and runtime-aware AI deployment workflows

From-scratch PyTorch Inception-ResNet-v2 image classification with a Caltech-256 training pipeline, TensorBoard logging, checkpointing, and torch.export support.

Architecture-blind torch.export compiler prototype for exact change-aware Core AI plans, with a documented portfolio-residency limit.

PyTorch and ONNX front ends (meta-device dispatch trace, torch.export, a torch.compile backend, ONNX graph walk) for an accelerator cost model; real model configs without weights; operator coverage as a metric