Skill UI

Quantizes HuggingFace LLMs to 8-bit or 4-bit with bitsandbytes, cutting memory by up to 75% while keeping accuracy, and supports QLoRA fine-tuning plus 8-bit optimizers for faster, memory-efficient training.

This workflow guides users through building advanced AI voice agents using the Retell AI SDK. It demonstrates how to implement complex call flows, integrate function calling for tasks like booking appointments, and configure voice behavior for professional telephony applications.

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.

Scans Solana and Anchor programs for six critical security vulnerabilities—arbitrary CPI, improper PDA validation, missing signer/ownership checks, sysvar spoofing, and instruction introspection issues—to support audits and pre-launch reviews.

This guide details how to set up the Speak language learning platform integration. It covers installing necessary SDKs (Speak SDK, OpenAI), configuring API keys, and establishing the connection required for sophisticated AI-powered tutoring. Use this when building applications that require real-time speech recognition, pronunciation feedback, and multilingual language support.

A comprehensive guide to leveraging Stable Diffusion models via the HuggingFace Diffusers library. This skill facilitates state-of-the-art image creation, including Text-to-Image generation from prompts, Image-to-Image translation, inpainting, outpainting, and advanced workflow building. Supports various models like SDXL and SD 3.0, ideal for professional creative and computational vision tasks.

Guides Rube MCP users through Supportivekoala automation via Composio, emphasizing tool discovery, connection checks, schema compliance, and multi-step execution for reliable workflow orchestration.

Automates TensorFlow SavedModel deployment support with step-by-step guidance, best-practice patterns, production-ready code/config generation, and validation for serving, monitoring, and optimization scenarios.

Enables Claude to manage AI/ML model versions through the model-versioning-tracker plugin, automating logging, lineage tracking, performance summaries, and version retrieval whenever users request registry or deployment support.

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.

This skill guides the development of advanced voice agents, representing the frontier of natural human-AI interaction. It covers both low-latency Speech-to-Speech (S2S) architectures (like OpenAI Realtime API) and fully controllable STT→LLM→TTS pipelines. Key focus areas include minimizing end-to-end latency (targeting <800ms), handling conversational nuances, background noise, and managing interruptions for applications like customer support, IVR, and live translation.

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.