7.2K skills · sorted by relevance
Analyze a codebase to identify valuable assets, attack entry points, and realistic abuse paths. Produce a concise Markdown threat model that explains risk priorities, distinguishes existing protections from proposed mitigations, and makes assumptions explicit.
Train Mixture of Experts models, which activate selected specialist networks for each input. The skill guides architecture, input routing, workload balancing, and GPU distribution using DeepSpeed or HuggingFace, with examples for training configuration and inference optimization.
Use Outlines to make language models produce data in a defined format. The skill covers extracting information into JSON, restricting classification labels, and applying Pydantic models, which define fields and data types, with local model backends.
Apply HQQ quantization, which stores language model weights using fewer bits without calibration examples. Covers precision settings, inference backends, saving compressed models, deployment with HuggingFace Transformers or vLLM, and further training with LoRA.
Learn how GPT language models work by preparing text, training a small model, and generating samples. Covers CPU experiments, custom datasets, further training of pretrained GPT-2 models, and reproducing GPT-2 with multiple GPUs.
Convert language models to GGUF and reduce their size through quantization, which stores model weights at lower precision. Guides cover compression choices, local inference on CPUs and Apple Silicon, and integration with tools such as Ollama.
Adapt large language models to specific tasks through fine-tuning, or additional training. Prepares and validates datasets, configures LoRA or QLoRA to train a small subset of parameters, and supplies training settings, evaluation scripts, and deployment optimization guidance.
Train or fine-tune language and vision models on Hugging Face Jobs without a local GPU. Covers dataset checks, hardware and cost planning, training monitoring, Hub storage, and GGUF conversion for local use.
Update copies of public-company Excel models using financial disclosures and mapped input cells. Deliver an updated workbook or review workbook, with source references, change logs, unresolved items, and implications for estimates and valuation where the model supports them.
Create a Markdown threat model grounded in a codebase. Identify valuable assets, boundaries between trusted and untrusted components, realistic attack paths, and mitigations, then prioritize risks using explicit assumptions about likelihood and impact.
Implements mathematical models from a design report as reproducible code, runs calculations, and checks results. It records methods and validation findings in a results report and produces data-based PDF charts for papers.
Coordinates a mathematical modeling competition project from planning to final checks. It records formatting and language preferences, creates a plan and task checklist, and calls dedicated skills for modeling, coding, diagrams, paper writing, and verification.
Turn mathematical modeling reports into DrawIO diagrams showing solution steps, model structures, and data processing flows. Produce editable source files, export PDFs for papers when DrawIO is available, and record diagram choices and export results.
Clear the model pin, a setting that holds a session to one model, in the current Codex session through Weave Router. The session continues within the same turn, and the router's response is reported.
Choose the AI model for each agent launch using saved preferences, session instructions, agent settings, and task type. Save global or per-agent choices for future sessions and include the selected model in launch acknowledgments.
Switch automatic model selection to lower-cost options for agent tasks, either for one session or through a saved setting. Explicit model choices remain in effect, while architecture and security review tasks use a separate economy selection.
Develop Bayesian models for quantitative social science, from choosing priors to reporting posterior results. Guide MCMC, a sampling method for estimating distributions, check convergence, and assess sensitivity for hierarchical models and small datasets.
Upload model weights to Hugging Face Hub with an OpenMDW 1.1 license and a required research warning. Check the published README and license file after uploading or updating to confirm that the required information is present.
Forecast sales, sensor readings, and other single-variable time series with Google's TimesFM, without training a custom model. Accept CSV files, DataFrames, or arrays and obtain forecast values with prediction intervals that express uncertainty.
Build production systems for machine learning models, from deployment to monitoring and retraining. The skill also guides LLM integration and retrieval augmented generation, which uses retrieved information to support answers, with scripts and architecture references.
Plan experiments, analyze data, and build and evaluate predictive models. The skill also covers explaining findings to stakeholders and designing production workflows for model deployment, monitoring, and retraining to support business decisions.
Analyze a codebase to identify valuable assets, attack entry points, and realistic abuse paths. Produce a concise Markdown threat model that explains risk priorities, distinguishes existing protections from proposed mitigations, and makes assumptions explicit.
Develop healthcare prediction models from electronic health records and physiological signals. Use PyHealth to prepare data, train models for tasks such as mortality or readmission prediction, and evaluate accuracy, fairness, and how predictions can be explained.
Explain an existing machine learning model by summarizing influential input features and examining individual predictions. It helps investigate unexpected feature interactions and communicate model behavior to nontechnical stakeholders, while noting limitations and potentially misleading explanations.