KaggleKaggleOfficialยท2 Agent Skills Included

kaggle-benchmarks

Custom LLM evaluation tasks, assertions, and dataset benchmarking

Builds custom benchmark tasks to rigorously evaluate language models on any task you define. Replaces ad-hoc prompt testing with structured assertions, multimodal inputs, tool use, and multi-turn conversation checks. Runs evaluations across full datasets and multiple models, then publishes reproducible results to Kaggle leaderboards.
npx skills add Kaggle/kaggle-benchmarks --all -g -y
Available:

Gives your AI agent a map of the library's architecture, key API files, and operational commands so it can write and run benchmark tasks correctly.

All Skills in This Repository (2)

Pure Emerald Level Indicators

Frequently Asked Questions

FAQPage Schema
How to install kaggle-benchmarks?โ–ผ

Run `npx skills add Kaggle/kaggle-benchmarks --all -g -y` in your terminal to install all skills in this repository globally.

How to evaluate an LLM on a custom task?โ–ผ

Define a Python function with the @kbench.task decorator, prompt the model inside it, and validate outputs with kbench.assertions. Then call .run() or .evaluate() to execute it against one or more models.

Can I benchmark models over a whole dataset?โ–ผ

Yes. Pass a pandas DataFrame to .evaluate() to run your task across every row, with parallel workers, retries, and aggregate accuracy scores.

Does kaggle-benchmarks support images and tool use?โ–ผ

Yes. Tasks can send image, audio, and video inputs, request structured outputs via schemas, and give models tools like a built-in Python interpreter.

What is the Kaggle Standardized Agent Exam?โ–ผ

It is a skill that lets your agent register on Kaggle, take a 16-question timed exam over HTTP, and earn a public score on the leaderboard.

Related Repositories in Data & Analytics

View All in Data & Analyticsโ†’