Skill UI

Automates guidance for evaluating machine learning models, covering best practices, sample code, and validation patterns triggered by mentions of model evaluation metrics.

Combines fine-tuned models with Mergekit to blend math, coding, and chat expertise without retraining, covering SLERP, TIES, DARE, Task Arithmetic, and linear merges for rapid experimentation and deployment.

Compress large language models quickly using Wanda, SparseGPT, and N:M pruning to enforce hardware-friendly sparsity, enabling faster inference and deployment without retraining.

Automates guidance for deploying machine learning models with quantization, covering tooling, best practices, and production readiness for inference, MLops, and monitoring tasks.

This skill provides comprehensive guidance on training Mixture of Experts (MoE) models, including architectures like Mixtral and DeepSeek-V3. It details techniques for achieving large-scale model capacity with significantly reduced computational costs (e.g., 5x reduction vs. dense models). Topics covered include top-k routing, load balancing, expert parallelism, and integration with frameworks like DeepSpeed and HuggingFace for efficient, resource-constrained training.

Guides researchers through nnsight’s tracing and activation APIs so you can inspect or modify PyTorch internals locally or via NDIF-powered remote runs on 70B+ foundation models without local GPUs.

High-performance RLHF framework that combines Ray, vLLM, and ZeRO-3 to accelerate PPO/GRPO/RLOO/DPO training for 7B-70B+ models on distributed GPU clusters, providing a single-stack workflow for reward models and policy optimization.

Implements standardized function/tool calling across various large language models (LLMs) via OpenRouter. This unified approach allows developers to build sophisticated, multi-step agents and complex workflows, regardless of the underlying provider (e.g., GPT-4o, Claude 3.5). It simplifies the process of achieving structured output and tool-augmented reasoning in production systems.

This guide provides a comprehensive quickstart for integrating with OpenRouter, a unified endpoint for accessing multiple large language models (LLMs). It includes examples in cURL, Python, and TypeScript, demonstrating how to send chat completion requests, handle complex JSON responses, manage model parameters, and calculate usage cost across various providers (e.g., Llama, GPT-4o, Claude). Ideal for developers testing API connectivity or building multi-model applications.

Use this tool to query, filter, and compare over 400 available AI models from OpenRouter. It allows users to check pricing, context lengths, model capabilities (e.g., tool calling), and select the best model based on specific needs, ensuring optimal performance and cost efficiency in AI development.

Optimizes deep learning models by analyzing architecture, data, and metrics to select optimizers, learning rate schedules, and regularization strategies, then outputs optimized code and evaluates accuracy, time, and resource usage.

This guide implements comprehensive guardrails for Perplexity Sonar API. It covers crucial aspects like query content moderation, PII sanitization, enforcing user usage quotas, selecting appropriate models based on user tiers, and scoring citation reliability to ensure safe, ethical, and cost-effective AI usage.