DeepSeek is working on its next-gen AI model – V4, and it turns out that the new technology will run entirely on the latest Huawei AI chips. A new report shed light on the preparation of the dynamic LLM and revealed one of the most significant points.
TheInformation reported that the upcoming advanced Chinese AI LLM (Large Language Model) – DeepSeek V4 will use Huawei AI chips. It may sound like normal news, as many Chinese AI startups are now turning their heads towards Huawei.
However, it’s more than just a small piece of information!
According to the details, the V4 model will only and completely use the latest Huawei AI chips. Companies like Alibaba Group, ByteDance, and Tencent have recently ordered Ascend 950PR semiconductors in bulk.
V4 AI model is expected to use nearly 1-trillion paramater architecture with probably hundreds of thousands of Ascend 950PR chips to trigger 1.8x faster inference speeds, a 1-million-token context window, and enhanced efficiency via Engram.
Further details reveal that DeepSeek has worked with Huawei for months and Cambricon (Chinese chipmaker) to rewrite and test the model’s core code components.
Neither DeepSeek nor Huawei has confirmed the news yet. But there is a huge possibility that this information is true. DeepSeek has been using Ascend chips in its AI models for a long time. Thus, it won’t be a surprise if the AI startup completely turns to Huawei amid the conflicts of Nvidia H20 exports in China!
DeepSeek V4 will likely be released in the coming weeks and will be known for its high-performance coding + reasoning. In addition, the company is working on two more V4 variants (with China-made chips) that might debut later this year.
In the meantime, we may learn more about how many Ascend 950PR chips reside in the DeepSeek V4 AI model. Stay tuned.
(Image Credits: DeepSeek)
The post DeepSeek V4 model will run entirely on Huawei AI chips: Report appeared first on Huawei Central.
