The new "reasoning" language model DeepSeek R1 is now available for private use on your own server!
It appears that DeepSeek has emerged as a significant player in the realm of OpenSource Large Language Models (LLMs). Just recently, they unveiled the DeepSeek R1 model, which draws its name from "reasoning" and is designed to mimic the cognitive processes akin to O1 models. This new offering closely resembles DeepSeek V3 but features a MoE architecture and boasts 685 billion parameters. Notably, it achieves comparable performance to leading O1 models from OpenAI and Gemini's Flash Thinking, marking a notable milestone in AI development.
This breakthrough is further underscored by the model's open-source nature, distributed under the MIT License, allowing enthusiasts and developers alike to access, download, and utilize it within platforms like Ollama.
"Reasoning" models are designed to solve complex logical problems in science, programming, and mathematics. These new models aim to overcome some of the limitations of previous versions by improving the way artificial intelligence "reasons" before generating answers. Such models are trained to spend more time thinking about problems, mimicking the human thought process, naturally improving the quality of the answer compared to a classic large language model.
A closed and secure platform with local LLM including DeepSeek R1 can be easily deployed on a company server for shared use by all employees of a web application or using API to develop functional applications. All that is needed for this is a GPU server and a little time to configure services.