Home › Language Models › Marco-o1

Marco-o1 - AI Language Models Tool

Overview

Marco-o1 is an open-source large reasoning language model developed by the MarcoPolo team at Alibaba International Digital Commerce (AIDC-AI). It is designed for open-ended, real-world problem solving by combining chain-of-thought (CoT) fine-tuning, Monte Carlo Tree Search (MCTS) and explicit self-reflection mechanisms to expand solution spaces and improve answer quality. The project provides both model checkpoints and the MCTS search code so researchers can reproduce the tree-based reasoning pipeline and create distilled training data for smaller models. ([github.com](https://github.com/AIDC-AI/Marco-o1?utm_source=openai)) Marco-o1 is built by fine-tuning a Qwen2-7B-Instruct base (reported as an ~8B-parameter checkpoint on Hugging Face) using a mix of filtered Open-O1 CoT data, proprietary Marco-o1 CoT data, and instruction datasets. Its two public iterations — v1 (released 2024-11-13) and v2 (released 2025-02-14) — focus on improving planning, math/benchmark performance, and distillation-friendly CoT data; the v2 work was later documented in a paper accepted to ACL 2025. The repository includes usage examples (transformers and vLLM), deployment hints (FastAPI), and the open-source MCTS implementation for reasoning-time search. ([huggingface.co](https://huggingface.co/AIDC-AI/Marco-o1?utm_source=openai))

GitHub Statistics

  • Stars: 1,534
  • Forks: 80
  • Contributors: 6
  • License: NOASSERTION
  • Primary Language: Python
  • Last Updated: 2025-05-30T07:51:44Z

The project shows active research-driven development with community uptake: the GitHub repository lists multiple releases, papers, and model/dataset artifacts. According to the repository and linked project pages, Marco-o1 has substantive community attention (approximately 1,534 stars, 80 forks) and a small core set of contributors (6), with recent commits through late May 2025. The README documents v1/v2 releases and links to the arXiv papers and model checkpoints for reproducibility. Overall activity indicates a research-focused open-source project with growing downstream adoption (model uploads, quantizations, and community spaces). ([github.com](https://github.com/AIDC-AI/Marco-o1?utm_source=openai))

Installation

Install via pip:

# Clone the repository
git clone https://github.com/AIDC-AI/Marco-o1
# Change into the repository
cd Marco-o1
# Install Python dependencies
pip install -r requirements.txt
# Example: load model with Hugging Face Transformers
python -c "from transformers import AutoTokenizer, AutoModelForCausalLM; tokenizer=AutoTokenizer.from_pretrained('AIDC-AI/Marco-o1'); model=AutoModelForCausalLM.from_pretrained('AIDC-AI/Marco-o1')"

Key Features

  • Chain-of-Thought fine-tuning to elicit step-by-step reasoning and produce explicit CoT traces.
  • Monte Carlo Tree Search integration to explore multiple reasoning paths and score alternative token choices.
  • Mini-step reasoning and reasoning-action strategies to vary action granularity during search.
  • Reflection/self-check mechanism prompting the model to reconsider or refine candidate solutions.
  • V2 distillation pipeline (MCTS-based data) and DPO for improved small-model post-training.

Community

Marco-o1 is research-oriented with accelerating community adoption: the GitHub project (public README, code, and MCTS) draws attention and contributions, while the model is mirrored on Hugging Face with downloads and community quantizations (ex. exllama branches and community spaces). The authors published an arXiv paper (Nov 21, 2024) describing design and benchmarks and a follow-up v2 paper (Mar 3, 2025) accepted at ACL 2025, signalling academic validation and ongoing development. Community activity includes HF downloads, third-party quantization/ports, and discussion threads—useful signs for researchers wanting to reproduce or extend the work. ([github.com](https://github.com/AIDC-AI/Marco-o1?utm_source=openai))

Last Refreshed: 2026-01-16

Key Information

  • Category: Language Models
  • Type: AI Language Models Tool
Visit Official Website

Related Tools

Qwen2.5-7B

Qwen2.5-7B is a large language model designed for text generation, featuring improvements in coding, mathematics, instruction following, long text generation, and multilingual support. It supports context lengths up to 128K tokens and is intended for sophisticated NLP tasks.
DeepSeek-V3

Open‑weight flagship LLM (base and chat) with strong reasoning and tool‑use; runs with SGLang/LMDeploy/vLLM and permissive model license.
Llama 3

Llama 3 is an open access large language model (LLM) released by Meta, available in various configurations (8B and 70B parameters) with capabilities for fine-tuning and integrations into platforms like Hugging Face, Google Cloud, and Amazon SageMaker.
UNfilteredAI-1B

A large-scale text generation model designed for creative and unconstrained content production without traditional filtering.
Shuttle-3

Shuttle-3 is a state-of-the-art language model designed for high-quality text generation, particularly suited for complex chat, multilingual communication, and reasoning tasks. It is fine-tuned from the Qwen-2.5-72b-Instruct model and designed to emulate high-quality prose similar to Claude 3 models.
WizardLM

WizardLM is a state-of-the-art large language model designed for complex chat, multilingual tasks, reasoning, and agent functionalities. It features an AI-powered pipeline (Auto Evol-Instruct) that optimizes instruction datasets for improved performance across various domains and leverages Arena Learning for an expanded learning pool of challenging instruction data.