Hugging Face researchers are trying to build a more open version of DeepSeek's AI 'reasoning' model

Barely a week after DeepSeek released its R1 “reasoning” AI model, researchers at Hugging Face have launched Open-R1, a project aimed at replicating R1 and open-sourcing all its components. Despite R1 being permissively licensed, it lacks transparency as the tools used to build it remain undisclosed. Hugging Face's initiative, led by Leandro von Werra and Elie Bakouch, seeks to address this by creating an open-source version, allowing for greater control over data and processes, which is crucial for responsible deployment and understanding biases.

The Open-R1 project plans to replicate R1 using Hugging Face’s Science Cluster and is soliciting help from the AI community on platforms like GitHub. The project has garnered significant interest, with 10,000 stars on GitHub in just three days. If successful, it could pave the way for the development of next-generation open-source reasoning models, benefiting the entire AI community. While there are concerns about potential misuse of open-source AI, Bakouch argues that the advantages of openness and collaboration outweigh the risks, promoting innovation across the field.

Key takeaways:

Hugging Face is attempting to replicate DeepSeek's R1 AI model through the Open-R1 project to promote open knowledge and transparency.
R1 is a reasoning AI model that surpasses OpenAI's o1 model in some benchmarks but lacks open-source components, making it difficult to study and replicate.
The Open-R1 project aims to fully open-source R1's architecture, using Hugging Face's Science Cluster and community collaboration to build a training pipeline.
Open-R1 seeks to establish a foundation for future open-source reasoning models, emphasizing the benefits of open-source development in AI.

Hugging Face researchers are trying to build a more open version of DeepSeek's AI 'reasoning' model | TechCrunch

Key takeaways:

Comments (0)

Newsletter