TinyZero logo

TinyZero

Free

Clean, minimal, accessible reproduction of DeepSeek R1-Zero

FreeFree tier
Inputs: textOutputs: text
Type
Open Source
LinksX

About TinyZero

TinyZero is a minimal reproduction of DeepSeek R1-Zero focused on countdown and multiplication tasks. Built upon the veRL (versatile reinforcement learning) library, it demonstrates how a 3B base language model can autonomously develop self-verification and search abilities through reinforcement learning — often called the "aha moment" of reasoning emergence. The project provides installation instructions, data preprocessing scripts, and training configurations for both single-GPU (1.5B model) and multi-GPU (3B+ model) setups, using the Qwen2.5 base model series. It is designed to be accessible and costs approximately $30 to run. Note: the repository is deprecated; users are directed to the latest veRL library for ongoing experiments.

Key Features

Reproduction of DeepSeek R1-Zero reinforcement learning training
Countdown and multiplication task data and training scripts
Emergence of self-verification and search abilities in a 3B base model
Built on the veRL (versatile RL) library
Supports single-GPU (1.5B) and multi-GPU (3B+) configurations
Uses Qwen2.5 base model series (0.5B, 1.5B, 3B, Instruct variant)
Includes data preprocessing for Qwen chat template support
Approximately $30 to run experiments

Pros & Cons

Pros
  • Clean, minimal codebase with low cost (~$30) to replicate results
  • Demonstrates emergence of self-verification and search in a 3B model
  • Built on the well-established veRL library for RL training
  • Open source with permissive license (MIT)
  • Supports multiple model sizes and GPU configurations
Cons
  • Repository deprecated; recommended to use latest veRL directly
  • Requires specific dependencies (vllm, flash-attn, Ray, etc.) and GPU memory
  • 0.5B base model fails to learn reasoning (as noted in docs)
  • Only two tasks (countdown, multiplication) are demonstrated
  • Not actively maintained; may not receive updates or bug fixes

Best For

Reproducing DeepSeek R1-Zero results for research and studyTraining language models to develop reasoning abilities via RLExploring emergence of self-verification and search in small modelsRL experimentation on mathematical reasoning tasks

FAQ

What is TinyZero?
A minimal reproduction of DeepSeek R1-Zero for countdown and multiplication tasks, showing how a 3B base LM develops self-verification and search via reinforcement learning.
How much does it cost to run?
Approximately $30 for compute costs, as stated in the README.
Is TinyZero actively maintained?
No, the repository is deprecated. Users are directed to the latest veRL library for ongoing RL experiments.
What base models does it support?
It uses the Qwen2.5 series (0.5B, 1.5B, 3B) and also supports Qwen2.5-3B-Instruct.
What tasks are available?
Countdown and multiplication tasks are provided with data preprocessing scripts.