Hugging Face is where DeepSeek shows its new R1 reasoning AI model
The Chinese tech trailblazer, DeepSeek, has just launched the spruced-up version of its R1 reasoning AI model. Broadcasted in a Wednesday morning WeChat message, they announced that their refurbished model is now available on the ever-popular developer platform Hugging Face.
With the flexibility of its permissive MIT license, the updated R1 can freely strut its stuff in the commercial world. Though labeled as a “minor” upgrade in DeepSeek's WeChat hot news, digging into the Hugging Face repository might leave you a tiny bit disappointed. Why? It’s only packed with configuration files and weights, the nuts and bolts directing a model’s action, not a lavish model description.
The souped-up R1 is no lightweight, tipping the scales at a whopping 685 billion parameters — or, to use jargon, "weights." Unfortunately, all these superb improvements come with a letdown: it's unlikely to be compatible with your everyday, run-of-the-mill hardware without some tinkering.
Let’s roll back a bit to the time when DeepSeek first made waves in the tech world. The introduction of R1 sent shockwaves across the AI field, even rattling the cages of top-dogs like OpenAI. Yet, not everyone's enthusiastic about this rising star. DeepSeek has been a thorn in the side of some stateside regulators, stirring up a heated debate over whether its technology puts national security at risk.