Skill UI

This skill provides continuous, structured monitoring of deep learning model training metrics. It analyzes data from platforms like WandB and local logs to detect critical anomalies such as NaN values, divergence, plateaus, and sudden spikes. By providing a clear decision (CONTINUE, WAIT, or STOP) and detailed evidence, it ensures compute resources are only spent on viable training runs.

This skill automates the process of renting, managing, and terminating cloud GPU instances on vast.ai. It analyzes your AI/ML task—determining VRAM, required GPU count, and estimated runtime—and then searches for the most cost-effective and reliable GPU offers, presenting optimized options for your training workflow.

Collection of battle-tested PyTorch training recipes covering architecture selection, optimizers, LR scheduling, precision tips, and debugging to train or fine-tune LLMs, vision, diffusion, and biomedical models with high GPU throughput.

Utilize Modal, a serverless GPU cloud, for running complex machine learning workloads, including training, fine-tuning, and inference. This tool provides a zero-config environment, eliminating the need for SSH setup or Docker management. It features automatic scale-to-zero billing and local-first execution, making high-powered GPU compute accessible for debugging and small-scale experiments.

This expert tool provides deep, systematic auditing of datasets to assess completeness, consistency, accuracy, and structural validity. It performs full data profiling, detects anomalies (outliers), analyzes missing data mechanisms (MCAR/MAR/MNAR), and calculates a Data Quality Score (DQS). Use it when validating data for model training, debugging pipelines, or establishing ongoing monitoring checks.

Interactive training for the GitHub Copilot CLI with guided lessons, quizzes, scenario challenges, and a reference covering slash commands, shortcuts, agents, MCP, and configuration; say “cliexpert” to begin.

Provides programmatic and standardized access to massive, versioned datasets of single-cell and spatial transcriptomics from CZ CELLxGENE. This skill allows users to query millions of cells, analyze gene expression, and access pre-calculated embeddings across various tissues, diseases, and species without downloading entire datasets. Essential for large-scale bioinformatics analysis, machine learning model training, and comprehensive cross-dataset comparison.

Histolab is a specialized Python library for digital pathology designed to process Whole Slide Images (WSI). It automates critical steps including sophisticated tissue detection, customizable tile extraction (random, grid, score-based), and stain normalization. It is essential for preparing high-quality, structured datasets for deep learning model training and comprehensive quantitative whole slide analysis.

Modal is a serverless cloud platform designed for executing Python-based AI/ML workloads. It provides on-demand access to powerful GPUs (e.g., H100, A100), scalable serverless functions, and custom containerization. Use cases include deploying ML models, running large-scale training/inference jobs, creating web APIs, and scheduling data pipelines, all defined entirely in code.

Molfeat is a comprehensive Python library designed for molecular featurization, providing over 100 pre-trained and hand-crafted featurizers. It unifies the process of converting chemical structures (SMILES strings or RDKit molecules) into robust numerical feature vectors. This capability is essential for advanced cheminformatics tasks, including Quantitative Structure-Activity Relationship (QSAR) modeling, virtual screening, similarity searching, and training deep learning models on molecular data.

PennyLane is a quantum computing library designed for building and training quantum circuits, particularly for quantum machine learning (QML) and quantum chemistry. It offers automatic differentiation capabilities for quantum circuits, making it ideal for developing hybrid quantum-classical models. It is hardware-agnostic, supporting backends from major providers like IBM, Google, and IonQ, and integrates seamlessly with classical ML frameworks such as PyTorch and JAX. Use cases include variational algorithms (VQE, QAOA), quantum neural networks, and molecular simulation.

This guide provides comprehensive, version-aware documentation for using PufferLib in reinforcement learning pipelines. It details best practices for adapting Gymnasium/PettingZoo environments, managing different PufferLib versions (3.0.0 vs 4.0 source), and executing safe, reproducible training and evaluation routines. Key focus areas include environment contract validation, sandboxing native builds, and maintaining data provenance.