Skill UI

A comprehensive guide to optimizing deep learning workflows and maximizing efficiency on cloud GPU platforms like Vast.ai. Learn techniques for selecting optimal instances by performance-per-dollar, reducing startup latency, tuning data pipelines (DataLoader, pin_memory), and scaling multi-GPU training using PyTorch and Bash scripting.

This skill helps diagnose performance bottlenecks, slow jobs, and expensive queries across various data platforms (Airflow, dbt, Databricks). It uses a tiered investigation approach, starting with broad discovery to pinpoint problematic jobs or queries, bridging to affected tables, and finally drilling down to identify root causes, latency trends, and change-related regressions.