Skill UI

A comprehensive toolkit for evaluating the performance of code generation models. It supports benchmarking against industry standards like HumanEval, MBPP, and MultiPL-E across multiple languages. Use this tool to quantitatively compare different LLM coding abilities and measure code synthesis quality using pass@k metrics.

Evaluates large language models across 60+ academic benchmarks (MMLU, HumanEval, GSM8K, TruthfulQA, HellaSwag) with standardized prompts and metrics, making it easy to benchmark HuggingFace or API models, compare releases, and track training progress.

This skill manages and refines search results obtained from Exa's neural search API. It provides comprehensive control over content extraction scope, allowing users to selectively retrieve metadata, highlights, full text, or structured summaries. Key features include implementing result caching with TTL, deduplicating citations, and crucial token budget management, ensuring the retrieved context fits within LLM context window limits for robust RAG pipelines.

Guides ML engineers through dataset prep, PEFT strategy selection, LoRA/QLoRA adapter setup, training monitoring, and deployment checks for practical LLM fine-tuning, quantization, and adapter merge workflows.

Fine-tune LLMs with TRL’s RLHF toolkit, covering SFT, DPO, PPO/GRPO and reward model training so you can align HuggingFace Transformers with human preferences or feedback in post-training workflows.

This workflow provides the primary mechanism for converting complex, live websites into clean, LLM-ready markdown content. It supports single-page scraping using `scrapeUrl`, deep multi-page site crawling with `crawlUrl`, and scalable async crawling for large datasets. Ideal for building robust content ingestion pipelines and reliably extracting structured information from the web.

This advanced workflow executes sophisticated web data gathering tasks. It supports three primary functions: LLM-powered structured data extraction using custom JSON schemas or prompts; efficient batch scraping of multiple known URLs; and rapid site mapping to discover all reachable URLs on a domain. Use this when you require reliable, typed data structures rather than raw markdown content.

A comprehensive pipeline for processing, validating, and optimizing scraped web content from Firecrawl. This tool handles markdown cleaning, structured data extraction using Zod validation, content deduplication, and chunking tailored for Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG) systems. It ensures crawled data is standardized, clean, and ready for downstream consumption or knowledge base indexing.

Firecrawl provides a comprehensive API for web data harvesting. It enables users to perform single-page scraping, multi-page crawling, site mapping (URL discovery), and advanced structured data extraction using Large Language Models (LLMs). Ideal for integrating data pipelines, building web monitoring tools, or conducting market research.

A comprehensive guide for migrating web scraping pipelines from tools like Puppeteer, Playwright, and Cheerio to Firecrawl. This skill demonstrates how to replace complex, browser-intensive code with simple API calls, handling single-page scraping, structured data extraction using LLMs, and large-scale site crawling, while eliminating the need for manual browser management and anti-bot workarounds.

This guide details integrating FoundationModels for running Large Language Models (LLMs) directly on the device (iOS 26). It covers advanced patterns including text generation, extracting structured data using @Generable, implementing custom tool calling for domain logic, and streaming responses for real-time UI updates. Key features emphasize strong privacy and offline capability, ensuring no data leaves the device.

Fuzzing Dictionary supplies domain-specific tokens to guide fuzzers when targeting parsers, protocols, configs, or file formats so they reach deeper logic; include magic bytes, headers, keywords, and combine generation from LLMs, headers, or binaries before passing via the fuzzer flag.