Skip to main content
Welcome to Verifiers! This library provides a flexible framework for creating RL environments and evaluations with custom multi-turn interaction protocols.

​
What is Verifiers?

Verifiers enables you to:
  • Define custom interaction protocols between models and environments
  • Build agents, multi-turn conversations, tool-augmented reasoning, and interactive games
  • Create reusable evaluation environments with multi-criteria reward functions
  • Train models with the included RL trainer (via vf-rl) or integrate with other RL frameworks
Key features:
  • First-class OpenAI-compatibility for ChatCompletions and Completions
  • Extensible multi-turn interactions via MultiTurnEnv
  • Native tool calling support with ToolEnv
  • Modular reward functions through Rubric classes
  • End-to-end async compatibility with sync support where you want it
  • Full-spectrum scaling from CPU evaluations in Jupyter to multi-node GPU RL training
  • Environments as Python modules for easy installation, sharing, and reuse

​
Installation

​
Basic Installation

For evaluation and API model usage: 
uv add verifiers

​
Training Support

For RL training with the included trainer: 
uv add 'verifiers[rl]'
To use the latest main with RL extras: 
uv add 'verifiers[rl] @ git+https://github.com/PrimeIntellect-ai/verifiers.git@main'

​
Latest Development Version

To use the latest main branch: 
uv add verifiers@git+https://github.com/PrimeIntellect-ai/verifiers.git

​
Development Setup

For contributing to verifiers: 
git clone https://github.com/PrimeIntellect-ai/verifiers.git
cd verifiers
uv sync --all-extras && uv pip install flash-attn --no-build-isolation
uv run pre-commit install

​
Integration with prime-rl

For large-scale FSDP training, see prime-rl.

​
Integration with Prime Intellect Environments Hub

Coming soon.

​
Documentation

​
Getting Started

Overview — Core concepts and architecture. Start here if you’re new to Verifiers to understand how environments orchestrate interactions. Environments — Creating custom interaction protocols with MultiTurnEnv, ToolEnv, and basic rubrics.

​
Advanced Usage

Components — Advanced rubrics, tools, parsers, with practical examples. Covers judge rubrics, tool design, and complex workflows. Training — GRPO training and hyperparameter tuning. Read this when you’re ready to train models with your environments.

​
Reference

Development — Contributing to verifiers Type Reference — Understanding data structures