Skip to content

Dataset · harborframework

A set of 100 complex tasks for evaluating AI agents

harborframework/harbor-mix

A curated selection of 100 complex and diverse agent tasks for fast and inexpensive model evaluation.

What's inside

The repository contains exactly 100 task directories selected from the Harbor Adapters benchmark pool. Each directory is expanded at the root of the repository and includes an instruction.md file with a task description. The set is designed to retain a broad signal from large-scale agent evaluations while requiring significantly lower computational costs.

What you can do with it

  • Fast benchmarking and quality evaluation of autonomous AI agents
  • Comparing models in text generation and reinforcement learning tasks
Tasks
Text generation, Reinforcement learning
Size
up to 1K rows
Downloads this month
56.4K
Likes
4
Access
on request on Hugging Face

How to load it

from datasets import load_dataset

ds = load_dataset("harborframework/harbor-mix")