What problem does it solve? Interviewers and hiring teams need deep, scenario-based PyTorch questions that distinguish candidates who have genuinely trained models from those who only followed tutorials. This Skill provides a structured question bank for PyTorch-specific technical interviews. ## Core Features & Use Cases - Layered Question Ladders: Covers autograd, training loops, DataLoader pipelines, mixed precision, DDP/FSDP distributed training, memory optimization, torch.compile profiling, checkpointing, ONNX export, custom operators, and LoRA fine-tuning, each with progression from basics to production debugging. - Good vs Bad Question Patterns: Reframes trivia questions into anomaly-driven scenarios (NaN loss, OOM, NCCL hangs, GPU underutilization) that test troubleshooting reasoning. - Resume-Based Hooks: Maps resume claims like "trained a 7B model" or "used DDP" to targeted follow-up probes on memory budgets, throughput numbers, and failure modes. - Use Case: When a candidate's resume mentions PyTorch or deep learning training, load this Skill to generate calibrated interview questions matched to their project scale, from single-GPU ResNet work to multi-node FSDP training. ## Quick Start Ask the AI to generate PyTorch interview questions for a candidate whose resume mentions distributed training and mixed precision fine-tuning of a 7B model.
