Machine Learning Skills

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.

> Receive and verify Hugging Face webhooks. Use when setting up Hugging Face webhook handlers, debugging X-Webhook-Secret verification, or handling events on models, datasets, and Spaces — repo updates, new commits and tags (repo.content), config changes (repo.config), discussions, Pull Requests, and discussion comments.

> Receive and verify OpenAI webhooks. Use when setting up OpenAI webhook handlers for fine-tuning jobs, batch completions, or async events like fine_tuning.job.completed, batch.completed, or realtime.call.incoming.

CRE Due Diligence analysis suite — 7 specialist skills for multifamily property analysis including rent roll validation, expense benchmarking, market study, physical inspection, environmental review, title review, and tenant credit assessment.

Unified Kaggle skill. Use when the user explicitly mentions Kaggle, kaggle.com, a Kaggle URL, Kaggle competitions, Kaggle datasets/models/notebooks, Kaggle forums/discussions/writeups, Kaggle benchmarks, hackathons hosted on Kaggle, Kaggle badges, or Kaggle account setup. Do not use for generic ML, GPU/TPU, notebook, dataset, benchmark, or data-science tasks unless the user clearly ties them to Kaggle.

Use when the user wants to build a consistent-identity LoRA for an original character — defining the character, generating a face/body-consistent multi-angle dataset (via the gpt-image-gen skill for codex image generation), captioning it, doing base-specific homework, training on a chosen base (Pony / Z-Image / others) on a local GPU, and producing a usable LoRA. This skill ORCHESTRATES the end-to-end pipeline and gates every expensive/irreversible step; it delegates actual image generation to gpt-image-gen and never improvises training settings from memory.

Use when the user wants to test, compare, or choose local Ollama models for their machine. Checks Ollama/GPU state, recommends model sizes from available VRAM, preserves existing benchmark records, pulls only approved models, runs repeatable benchmarks, restores stopped services, and writes a markdown comparison report. NOT for hosted API model evaluation or subjective chat-quality judging without local benchmark commands.

Implement loss functions in PyTorch with proper tensor operations.

Analyzing earthquake data relative to tectonic plate boundaries using PB2002 dataset and GeoPandas.

PyTorch patterns for implementing preference optimization losses (DPO, SimPO, etc.) for LLM training.

Parse and analyze SEC 13-F filing TSV datasets to extract AUM, holdings count, and fund details by accession number.

Guide for implementing preference optimization methods (DPO, SimPO, IPO) in PyTorch. Use when implementing loss functions for RLHF-style training.

Implements the SimPO (Simple Preference Optimization) loss function from the paper "SimPO: Simple Preference Optimization with a Reference-Free Reward". Use this when implementing or understanding SimPO training objectives.

Setup and installation of TRL (Transformer Reinforcement Learning) library with compatible torch/transformers versions. Use when setting up preference optimization training environments.

How to query and filter travel itinerary datasets (restaurants, accommodations, attractions, distances) to build constraint-satisfying travel plans.

Techniques for loading and initial processing of GeoJSON and coordinate-based datasets using GeoPandas.

Manages NLP environments, handling library dependencies like PyTorch, Transformers, and custom local packages.

Implements the Simple Preference Optimization (SimPO) loss function in PyTorch, focusing on length-normalized log probabilities and target reward margins.

Provides foundational techniques for loading, projecting, and manipulating geospatial datasets using GeoPandas.

Provides strategies for traversing and filtering hierarchical JSON datasets stored in flat files.

A skill for finding best string matches in datasets using fuzzy matching libraries like fuzzywuzzy or rapidfuzz.

Using Hugging Face TRL library for building custom trainers for Preference Optimization methods like DPO, SimPO, etc.

Implement the SimPO (Simple Preference Optimization) loss function for LLM alignment. Use this skill whenever implementing SimPO training objectives, reference-free reward optimization, or Bradley-Terry preference loss with target reward margin. Triggers on: SimPO, preference optimization loss, average log probability reward, gamma margin.

Work with MultiPolygon and MultiLineString geometries from real plate boundary datasets.

Proper initialization of SimPOTrainer with model loading and args setup

Advanced filtering and validation of travel datasets with strict cuisine matching, budget constraints, and data quality checks.

Excel statistical functions for datasets including MIN, MAX, MEDIAN, AVERAGE, QUARTILE, and SUMPRODUCT

Environment setup and testing patterns for SimPO preference optimization with PyTorch.

Explore and query travel dataset files (CSV/TXT) to find cities, restaurants, accommodations, attractions, and driving distances — with precise filtering patterns and data field notes.

Build a multi-city road-trip travel itinerary from dataset files with complete budget tracking, pet-friendly constraints, cuisine coverage, and correct JSON output format.

Efficient recursive exploration and content searching across large datasets.

Improved 13F dataset analysis functions (includes better handling of TSVs).

Robust PyTorch tensor operations for implementing loss functions, including handling device placement and numerical stability.

A refined skill to implement SimPO (Simple Preference Optimization) loss in PyTorch, precisely matching the official paper implementation without a reference model.

Load citizen science and expert annotation datasets, match them by image using file_rad column, handle missing data, and prepare data for clustering evaluation.

Filter the boundary dataset to include only boundaries relevant to the Pacific plate (where PlateA or PlateB equals the Pacific plate identifier). Use this skill to exclude irrelevant boundaries before distance calculations.

Examine the plates dataset to find the exact identifier used for the Pacific plate (e.g., "PA", "Pacific", etc.). Use this skill early in the analysis to ensure consistent filtering across boundaries and plates.

Load plate boundary GeoJSON and create a GeoDataFrame with proper geometry parsing. Verify column names (PlateA, PlateB) match the dataset structure. Use this skill to prepare boundary data before filtering for Pacific plate relevance.

Inspect the actual holdings dataset to understand its schema, field names, and data patterns before applying filters. Use this to identify the correct field name that distinguishes equity securities from bonds, options, warrants, and other non-stock instruments.

Search the COVERPAGE dataset using fuzzy matching to locate a specific fund by name and extract its accession_number for subsequent analysis. Use this when you need to identify a fund's filing by partial or approximate name matching.

Retrieve and load the complete holdings dataset for a specific fund using its accession_number. Use this to access position-level details needed for AUM, stock count, and comparative analysis.

Use when implementing the `simpo_loss` function in SimPOTrainer class. Extract loss computation logic from the paper and translate it to PyTorch code that accepts the expected tensor inputs.

Understanding the structure of SEC 13F filing datasets stored in /root/2025-q2 and /root/2025-q3 folders, including available scripts and data files

Specific knowledge of the PB2002 (Bird, 2002) dataset structure for identifying the Pacific plate.

Methods for calculating or retrieving the number of tokens consumed during an LLM inference task.

Load earthquake GeoJSON and plate boundary datasets, and ensure consistent projection for distance analysis.

Implements the SimPO loss function using stable log-probability operations in the specified trainer file.

Instructions for querying the provided datasets to ensure constraints regarding pet-friendly lodging, specific cuisines, and valid locations are met.