On April 29, Alibaba Cloud launched the Qwen3 series of open-source hybrid inference models.In less than a day, Homo Intelligent’s self-developed NPU quickly achieved efficient deployment of the Qwen3 series models (Qwen3 0.6B-14B) at the edge.This achievement fully demonstrates the significant advantages of Homo Intelligent’s NPU in ecological adaptability and rapid response capability.
Operation diagram of Qwen 3 on Homo Intelligent NPU
The Tongyi Qwen3 series, as a leading domestic hybrid inference model, highlights the innovative integration of “fast thinking” and “slow thinking” within the same model architecture. For simple requirements, it can quickly provide low-computation responses, achieving “instant replies”; when faced with complex problems, it can engage in multi-step deep thinking to gradually derive reasonable answers. Additionally, the Qwen3 series is pre-trained on massive multilingual and multimodal data and fine-tuned with high-quality data, demonstrating excellent performance in aligning with human preferences, tripling inference efficiency, and supporting API commercialization and open-source code libraries, providing users with flexible and diverse deployment options.
Homo Intelligent offers a rich selection of high-performance AI computing options through various product combinations. Based on its self-developed NPU, Homo Intelligent has launched products such as the Limo® SM30 computing module, Limo® LM30 intelligent accelerator card, and Limo® BX30 computing box, covering diverse application scenarios at the edge and in edge computing across government, industrial, consumer, and automotive sectors. These products, with their high performance and low power consumption, provide a solid computational foundation for the implementation of AI technology, meeting the needs of different users in various scenarios.
Previously, Homo Intelligent’s NPU successfully supported the DeepSeek R1 Distilled series models, showcasing its outstanding performance and broad compatibility in adapting to mainstream large models.This adaptation of the Tongyi Qwen3 series models further validates the efficiency and stability of Homo Intelligent’s NPU in handling complex AI tasks, providing strong evidence for the integrity and competitiveness of domestic technology stacks.
In the future, Homo Intelligent will continue to deepen its research in integrated storage and computing technology, continuously optimize NPU performance, strengthen cooperation with ecological partners, and promote the widespread application of domestic NPUs in the AI field. Through technological innovation and ecological co-construction, Homo Intelligent is committed to providing more users with efficient and inclusive AI computing solutions.
