·Explain dataflow firing rules and the trade-offs of static and dynamic scheduling.

·Trace a PyTorch model with torch.fx and annotate graph nodes with compute and memory cost.

·Compare how TPU, Groq, Cerebras, and IBM Spyre implement dataflow execution.

Trace the PyTorch 2 compilation path from TorchDynamo through TorchInductor.

Find and fix graph breaks and diagnose recompilation in compiled models.

Compare fusion, tiling, layout, and scheduling decisions across Inductor, XLA, and Spyre.

Explain why low-batch decode is often bandwidth-bound and estimate its arithmetic intensity.

Compare caches, scratchpads, and on-chip SRAM across accelerator architectures.

Size the KV cache and evaluate serving techniques that reduce or manage its memory cost.

Locate failures in compiled AI workloads by compiler stage and software layer.

Distinguish numerical differences from logic bugs using high-precision references and tolerances.

Use ablation, provenance, logs, and comparison to narrow the first point of divergence.

Bound operation runtime from FLOPs and bytes and place workloads on a roofline.

Profile LLM decode and separate time dominated by compute, memory movement, and overhead.

Benchmark compiled inference correctly and tune serving shapes using measured evidence.