Only a week after DeepSeek released R1’s “Progress” AI model, he embraced the AI model with the market every time.
Holding the person in charge of the research, LEANDRO VON WERRA and engineers of several companies have released Open-R1. This is a project to build R1 and open source duplication, all components, including data used for training.
The engineer said that DeepSeek’s “Black Box” release philosophy was forced to act. Technically, R1 is “open” in that the model is acquired a licensed license. In other words, it can be deployed with almost no restrictions. However, R1 is not an “open source” due to the widely accepted definition because some of the tools used to build it are wrapped in mystery. Like many high -flying AI companies, DeepSeek hates to clarify the secret source.
“The R1 model is impressive, but the details of the open dataset, the details of the experiment, or the intermediate model cannot be used. This makes it more difficult to duplicate and research,” he said. ELIE BAKOUCH, one of them, told TechCrunch. “The complete architecture of the open sourcing R1 is not only transparent but also the possibility.”
Not so open
Deepseek, a Chinese AI lab that was partially funded by quantitative hedge funds that released R1 last week. In many benchmarks, R1 matches the performance of O1 inference models in OPENAI.
Since R1 is a reasoning model, there is actually a check itself. This helps avoid some pitfalls that normally stumble. It takes a little time for a reasoning model to reach a solution compared to a normal non -rational model. The advantage is that it tends to be more reliable in domains such as physics, science, and mathematics.
R1 rose to the top of the Apple App Store chart after the DeepSeek’s ChatBot app, which provides free access to R1. The speed and efficiency developed by R1 -DeepSeek released the model a few weeks after OPENAI released O1 -many of the walls’ analysts and engineers maintained the lead in the AI race. I wondered if I could.
The OPEN-R1 project is not much concerned about the US AI advantage, rather than “opening the black box of model training completely”. He stated that it was difficult to study the model in detail because R1 has not been released under training code or training instructions.
“Controling datasets and processes is important for developing models with responsibility in sensitive fields,” says Bakuch. “It also helps to understand and deal with model bias. Researchers need more fragments (…) to push up the boundaries of things possible.”
Duplicate procedure
The goal of the Open-R1 project is to replicate R1 in a few weeks. This depends on the fact that Face’s Science Cluster, a dedicated research server with 768 NVIDIA H100 GPUs, is dependent on embracing.
Hugging face engineers plan to tap scientific clusters and generate the same dataset as Deepseek used to create R1. To build a training pipeline, the team embraces the Open-R1 project hosted by the Open-R1 project, seeking help from a higher high-tech community.
“We need to confirm that algorithms and recipes (correctly) are implemented,” Von Werra told TechCrunch.
I already have a lot of interest. The Open-R1 project won 10,000 stars in just three days on GitHub. Star is a way to indicate that GitHub users like projects.
If the Open-R1 project succeeded, Bakuch said that AI researchers could build on training pipelines and work on the development of next-generation open source inference models. He hopes that the Open-R1 project will not only bring the R1’s powerful open source duplication, but also bring a better model.
“Open source development, not a Zero Sam game, can quickly make a profit to everyone, including Fronty Lab and model provider.
Some AI experts are concerned about the possibility of AI abuse of open source, but Bakouch believes that profits exceed their risks.
“If the R1 recipe is reproduced, anyone who can rent a GPU can use its own data to create a unique variant and spread technology anywhere.” “We are really excited about the recent open source release, which is strengthening the role of AI’s openness. It is that only a handful of labs can make progress and the open source is delayed. It is an important change for the field that changes the story.