Chinese AI Firm DeepSeek Launches Efficient Domestic Model
Chinese AI company DeepSeek has launched its latest large language model, DeepSeek-V3.2-Exp, designed to run efficiently on domestic hardware. The model's release comes amidst a growing effort by Chinese AI firms to reduce reliance on foreign technology, particularly Nvidia hardware.
DeepSeek-V3.2-Exp is optimized for Huawei's Ascend hardware and CANN software stack. It features a sparse attention mechanism to lower memory and compute requirements for long-context inference. This innovation makes it well-suited for processing lengthy sequences, as demonstrated by Cambricon's vLLM-MLU fork supporting the model.
The model's GitHub notes suggest it matches the performance of vLLM at launch, supporting both TileLang and CUDA kernels. DeepSeek has collaborated with Ascend and CANN teams to ensure smooth integration, with updates to GitHub and CANN repositories. Hygon has also announced that its DCU accelerators have been tuned for 'zero-wait' deployment with DeepSeek-V3.2-Exp through its DTK software stack.
Huawei's Ascend team and the vLLM-Ascend community have integrated DeepSeek-V3.2-Exp, providing custom operator installation steps and kernel packaging for Ascend NPUs. SGLang has also confirmed support for the model across multiple backends, including Ascend.
The rapid adoption of DeepSeek-V3.2-Exp by Chinese AI hardware companies signals a significant push towards AI sovereignty. With Baidu and other major Chinese companies supporting the model, it appears that domestic platforms are being positioned as first-class targets in the global AI landscape.
Read also:
- Web3 social arcade extends Pixelverse's tap-to-earn feature beyond Telegram to Base and Farcaster platforms.
- Trump praises the robustness of US-UK relations during his visit with Starmer at Chequers, showcasing the strong bond between the two nations.
- Navigating the Path to Tech Product Success: Expert Insights from Delasport, a Trailblazer in the Tech Industry
- German Startup Fernride Aims to Revolutionize Logistics with Autonomous Vehicles
