Skill UI

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.

This coach guides users through a progressive mastery of AI, structured across three layers: AI User (prompting), AI-Enhanced Worker (workflow integration), and AI Builder (ML foundations). It uses Socratic questioning, layered scaffolding, and spaced repetition to build genuine, transferable AI literacy, moving beyond simple copy-pasting prompts to deep conceptual understanding of models, transformers, and RAG.

This skill provides expert capabilities for training and fine-tuning transformer language models using the TRL (Transformers Reinforcement Learning) library. It supports state-of-the-art post-training methods, including Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), Group Relative Policy Optimization (GRPO), and Reward Model training. Use this skill to align and customize foundation models for advanced tasks via CLI commands.

This skill runs an automated evolution process designed to systematically optimize the seven core policy surfaces of an AI agent harness (e.g., planner, contextBuilder, reviewer). By mutating these policies and scoring the variants in a sandbox, it identifies crucial performance improvements without requiring expensive foundation model retraining. It is ideal for empirical configuration tuning and establishing optimized baselines.

A comprehensive guide for authors submitting computer vision papers (e.g., to ECCV). It outlines best practices for ensuring the full reproducibility of AI research, covering everything from detailed training recipes and hyperparameter logging to pinning foundation model dependencies and reporting variance honesty. Essential for academic publication integrity.

A comprehensive guide for academic researchers detailing how to achieve and disclose full reproducibility for Computer Vision and deep learning models. Covers essential practices such as protocol pinning for foundation models, detailed logging of datasets, hardware specifications, and variance honesty, ensuring results remain verifiable years after publication.

A comprehensive guide for positioning and refining computer vision manuscripts for the IEEE/CVF International Conference on Computer Vision (ICCV). This skill helps authors assess conference fit, structure evidence, perform benchmarking, and adopt the necessary narrative for a flagship CV submission, covering topics from foundation models to video recognition.