huawei cloud builds multi-modal, full-series models to provide optimal solutions for business scenarios and large models
2024-09-20
한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina
today, at the huawei connect conference 2024, huawei executive director and huawei cloud ceo zhang pingan delivered a keynote speech on "leap to the cloud, ai reshaping thousands of industries", elaborating on huawei cloud's exploration and practice in building ai native cloud infrastructure, building a knowledge-centric data base, and building multimodal and multi-scale models.
zhang pingan said that the bigger the model, the better, and one large model cannot be applied to all business scenarios. it is necessary to build multi-modal and multi-size models to achieve the best match between scenarios and models to meet the needs of enterprises for the economy and professionalism of large models.
zhang pingan, executive director of huawei and ceo of huawei cloud
huawei cloud pangu big model 5.0 includes nlp, cv, multimodality, as well as prediction big models and scientific computing big models, which can meet the full business scenario needs of enterprises. at the meeting, zhang pingan announced that pangu big model 5.0 has further enhanced the stcg spatiotemporal controllable generation capability in the field of multimodal generation. taking the autonomous driving scenario as an example, pangu big model 5.0 can not only generate scenes of normal car driving, but also generate random, accidental, and adversarial scenes, greatly reducing the dependence of intelligent driving training on massive road data mining.
at the same time, pangu big model 5.0 has the pangu e series that can run on the end side, the pangu p series for efficient reasoning, the pangu u series for processing complex scenarios, and the pangu s series, the unified ai brain of the enterprise, with parameter scales ranging from billions to trillions. in the specific practice of embodied intelligence, huawei cloud has greatly reduced the complexity of end-side design through the collaboration of cloud-side intelligence and end-side intelligence, and greatly improved the generalization ability and task execution ability of embodied intelligence in industrial scenarios. through the combination of end-cloud intelligence, the global embodied intelligence industry innovation center jointly established by huawei cloud and shenzhen bao'an qianhai has improved the operating accuracy of the robot arm from millimeters to hundreds of microns, and the dynamic insertion success rate of rigid parts has reached 99.99%.
over the past three years, huawei cloud pangu big model has been applied to more than 30 industries and more than 400 application scenarios, continuously "solving problems and doing difficult things" and reshaping thousands of industries.
proofread by sheng yuanyuan
