DeepSeek Breaks Through Again

DeepSeek releases new AI model that can process documents with 7-20 times fewer tokens than traditional methods. Photo: The Verge .

According to SCMP , DeepSeek has released a new multi-modal artificial intelligence (AI) model that is capable of processing large and complex documents with a significantly lower number of tokens, 7-20 times less than traditional text processing methods.

Tokens are the smallest units of text that AI processes. Reducing the number of tokens means saving computational costs and increasing the efficiency of an AI model.

To achieve this, the DeepSeek-OCR (optical character recognition) model used visual perception as a means of information compression. This approach allows large language models to process huge volumes of text without incurring a proportionally increased computational cost.

“Through DeepSeek-OCR, we have demonstrated that using visual perception to compress information can achieve significant token reductions – from 7-20 times for different historical context periods, providing a promising direction,” DeepSeek said.

According to the company's blog post, DeepSeek-OCR consists of two main components, the DeepEncoder and the DeepSeek3B-MoE-A570M, which acts as a decoder.

Among them, DeepEncoder acts as the core engine of the model, helping to maintain low activation levels under high-resolution input, while achieving strong compression ratio to reduce the number of tokens.

The decoder is then a 570 million-parameter Mixture-of-Experts (MoE) model that is tasked with reproducing the original text. The MoE architecture divides the model into subnetworks that specialize in processing a subset of the input data, optimizing performance without having to activate the entire model.

On OmniDocBench, a document readability benchmark, DeepSeek-OCR outperforms major OCR models like GOT-OCR 2.0 and MinerU 2.0, while using much fewer tokens.

Source: https://znews.vn/deepseek-lai-co-dot-pha-post1595902.html