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 guide details the production-grade reference architecture for running complex GPU compute workflows on Vast.ai. It covers the foundational three-tier pattern—Orchestrator, GPU Workers, and Artifact Storage—providing blueprints for job queue management, fault-tolerant training pipelines, and robust checkpointing. Use this when designing scalable and reliable ML/AI training systems.

Guides training large language models with Volcano Engine's verl library, covering PPO/GRPO workflows, distributed backends, multi-turn rollouts, reward configs, critic-enabled PPO, and production deployments using Ray workers.

Automates wandb experiment logger tasks within ML training, providing guidance, best practices, and code or configuration examples plus validation to keep data prep, training, hyperparameter tuning, and experiment tracking consistent.

Query and download the NCI Imaging Data Commons public cancer imaging datasets via idc-index, filter by metadata, fetch DICOM files, visualize series, and respect licensing details for AI training or research.

PathML is a Python toolkit for computational pathology workflows, loading 160+ slide formats, preprocessing H&E images, constructing spatial graphs, training HoVer-Net/HACTNet models, and analyzing multiplex imaging for ML-ready pathology data.

Hardware-agnostic quantum ML framework enabling gradient-based training of quantum circuits, hybrid quantum-classical models, and seamless integration with PyTorch/JAX/TensorFlow alongside device portability across IBM/Google/Rigetti/IonQ.

PufferLib is a high-performance library designed for advanced reinforcement learning simulations and training. It excels in parallel environment simulation, offering optimized vectorization and native support for multi-agent systems. Utilizing algorithms like PuffeRL (optimized PPO+LSTM), it achieves millions of steps per second, significantly accelerating research and deployment across single-agent and multi-agent environments.

PyTorch Lightning Toolkit streamlines PyTorch model training by organizing LightningModules, LightningDataModules, Trainer configs, callbacks, logging, and distributed strategies for scalable multi-GPU/TPU runs in research or production workflows.

Stable Baselines3 is a robust, PyTorch-based library providing production-ready implementations of core reinforcement learning algorithms (PPO, SAC, DQN, etc.). It facilitates the training of single-agent RL models, supports the creation of custom Gymnasium environments, and handles advanced parallel training via vectorized environments. It is ideal for rapid prototyping and rigorous RL experiments.

This skill provides zero-shot time series forecasting using Google's TimesFM foundation model. It forecasts any univariate time series (like sales, sensor data, or weather) without requiring custom model training. The model delivers point forecasts along with calibrated prediction intervals (quantiles), offering a robust and probabilistic view of future trends. It is ideal for rapid, reliable forecasting across diverse time-based datasets.

Analyzes comprehensive fitness logs to track quantitative trends in volume, intensity, and frequency. It tracks progress across multiple disciplines (running, strength) and correlates exercise metrics with vital health indicators like blood pressure and weight, providing personalized, science-backed training and recovery advice.