Chinese lab releases new open-use AI model


HIGH POINT— Chinese AI firm DeepSeek is out with its latest language model, DeepSeek-V3, which industry observers say is one of the most powerful open-source models to date.

Released under an extremely permissive license, TechCrunch reports that DeepSeek-V3 allows developers to download and modify it for various applications, including for commercial and enterprise use.

The model features 671 billion parameters, with 37 billion activated for each token processed. It utilizes advanced architectures that the company says enhances its overall performance.

Capable of handling a range of text-based tasks such as coding, translation, and writing essays or emails from descriptive prompts, DeepSeek V3 has surpassed both open and closed AI models in performance, according to the company’s internal testing.

The model outperforms other AI systems, including Meta’s Llama 3.1 405B, OpenAI’s GPT-4, and Alibaba’s Qwen 2.5 72B, in a subset of coding competitions on the platform Codeforces.

The model was trained on a massive dataset of 14.8 trillion tokens, with 1 million tokens equaling approximately 750,000 words.

Companies should keep in mind, however, that while DeepSeek V3’s size offers powerful performance, it also requires significant hardware to run efficiently. An unoptimized version would demand a bank of high-end GPUs to handle tasks at a reasonable speed.

Despite this, the model’s development is notable for its efficiency—DeepSeek trained DeepSeek V3 using Nvidia H800 GPUs in just two months, with a total cost of $5.5 million. This is a fraction of the cost of training models like OpenAI’s GPT-4.

DeepSeek V3 represents an advancement in AI model development that is worth keeping track of, despite its practical application being limited by hardware constraints.

See also:

a





Credit to Source link

1 thought on “Chinese lab releases new open-use AI model”

  1. Its like you read my mind You appear to know a lot about this like you wrote the book in it or something I think that you could do with some pics to drive the message home a little bit but instead of that this is fantastic blog An excellent read I will certainly be back

    Reply

Leave a Comment