Skill UI

SimPO (Simple Preference Optimization) is a state-of-the-art, reference-free method designed for aligning Large Language Models (LLMs) using human preference data. It is an efficient alternative to DPO and PPO, notably outperforming DPO without requiring a separate reference model. It is ideal for practitioners who need faster, simpler, and more resource-efficient fine-tuning for preference alignment.

This skill provides advanced guidelines for designing robust and effective tools specifically for autonomous AI agents. It shifts the focus from traditional developer APIs to contracts optimized for Language Models (LLMs). Key topics covered include treating tool descriptions as prompt engineering, implementing namespacing, applying the Consolidation Principle, and utilizing Architectural Reduction patterns (like file system access) to maximize agent reliability and workflow complexity.

torchforge is Meta's advanced PyTorch-native library designed for agentic Reinforcement Learning (RL). It fundamentally separates core RL algorithms from complex distributed infrastructure concerns. This allows researchers to rapidly prototype and experiment with novel RL methods (like GRPO) using clean, PyTorch-native abstractions, while automatically handling large-scale tasks like distributed training across thousands of GPUs using Monarch and TorchTitan.

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.

Provides automated guidance for designing, validating, and deploying Vertex AI pipelines within Google Cloud, covering compute, storage, BigQuery, and GCP-specific services for production readiness.

Vexor is a powerful command-line interface designed for advanced semantic search across local files. By utilizing vector embeddings and integrating with large language models (like Claude/Codex), it allows users to search files based on meaning and context, rather than just exact keywords. This is ideal for large codebases or document repositories where traditional keyword search methods are insufficient.

ZAI CLI provides comprehensive access to Z.AI's advanced capabilities, including image/video analysis, real-time web search, web page extraction, and GitHub code exploration. It is designed for users needing multi-modal content analysis, error diagnosis, and complex data processing through a unified command-line interface. Ideal for developers and analysts requiring robust, API-driven AI functionality.

A comprehensive directory designed for discovering, researching, and comparing autonomous AI agents, frameworks, and tools across diverse ecosystems. This skill helps users map the current AI agent landscape, identify existing solutions, gather best practices, and pinpoint market gaps before initiating custom development or workflow design.

DeepChem is a Python library for molecular machine learning, providing featurizers, dataset loaders, splitters, and models spanning traditional ML to GNNs for property prediction, drug discovery, and materials design.

DiffDock uses diffusion-based deep learning to predict protein-ligand binding poses, confidence, and batch virtual screening workflows for structure-based drug design.

Combines ESM3 and ESM C protein language models to generate sequences, predict structures, condition functions, and extract embeddings via local or Forge API inference for protein design and engineering workflows.

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.