OPPO’s multimodal algorithm expert Dr. Yu Xiyu delivered a keynote on edge‑side multimodal large‑model engineering at the AICon 2026 conference in Shenzhen (August 21‑22). His talk covered a full stack of techniques—including quantization‑aware training, sparse compression, long‑context cache eviction, decoding acceleration and zero‑order training—to address latency, pipeline complexity and model iteration challenges on devices.
The AICon event, organized by the global AI development community, will feature over 50 leading technology firms such as Tencent, Alibaba, Huawei, Baidu and Ant Group, focusing on AI infrastructure, large‑model systems, intelligent agents and data intelligence.
Separately, Lenovo and Intel equipped South China Agricultural University’s School of Mathematics and Information with AI servers (Lenovo WA5480 G3) powered by fifth‑generation Xeon Scalable processors, delivering up to 10^13 FLOPS, 64 cores, DDR5 memory and PCIe 5.0. The platform provides 7×24 reliability through redundant power and cooling, accelerating AI‑driven research and teaching workloads.
Peking University National Development Institute director Huang Yiping noted that while AI has not yet boosted total factor productivity, it is expected to become a core driver of economic growth, urging a “invest‑in‑people” strategy to develop talent that combines domain expertise with AI tools.
At AICon, Huawei AI open‑source director Huang Zhipeng and panelists discussed AI‑infra challenges, highlighting increasing model complexity—especially multimodal models such as MoE and full‑modality systems—and the need to manage heterogeneous compute resources across GPUs, NPUs and emerging domestic chips.
In the consumer‑robot sector, Qiyuan Robot completed a dual‑track rollout within a month: showcasing its humanoid robot at the UN AI for Good summit on July 10 and preparing for a debut at ChinaJoy on July 31. The robot blends open‑source DIY capabilities with entertainment features like IP collaborations and companion filming, illustrating a strategy that links global AI policy platforms with youth‑centric market exposure.
Entities: AICon 2026 Shenzhen · Huang Yiping · Huang Zhipeng · Intel · Lenovo · Mo Zifeng · Oppo · Qiyuan Robot · Yu Xiyu
Claims
What the coverage asserts, and how well corroborated each claim is across sources.
- [○ 1 SOURCE] The OPPO presentation described a full‑stack pipeline using quantization‑aware training, sparse compression, long‑context cache eviction, decoding acceleration, QALFT automation and zero‑order on‑site
- [○ 1 SOURCE] AICon 2026 Shenzhen hosted a panel on AI infrastructure featuring Huawei AI open‑source ecosystem director Huang Zhipeng, vLLM committer Mo Zifeng and Fourth Paradigm researcher Yang Shouren.
- [○ 1 SOURCE] Panel speakers suggested defining SLA/SLO and cost budgets, potentially delegating infrastructure tasks to AI models for efficiency.
- [○ 1 SOURCE] Yu Xiyu is an OPPO multimodal algorithm expert who discussed on‑device large multimodal model challenges and solutions.
- [○ 1 SOURCE] The panel identified heterogeneous compute management and kernel consistency across GPUs, NPUs and domestic accelerators as a key difficulty.
- [○ 1 SOURCE] OPPO presented a talk titled “OPPO端侧多模态大模型工程化实践” at AICon 2026 Shenzhen on August 21‑22.
- [○ 1 SOURCE] The panel highlighted the increasing complexity of model architectures, such as Mixture‑of‑Experts and multimodal VLM Omni, as a major AI infrastructure challenge.
- [● 2 SOURCES] AICon conference took place in Shenzhen, China on August 21‑22 2026.
- [○ 1 SOURCE] AICon 2026 will be held in Shenzhen on August 21‑22 with over 50 tech firms including Tencent, Alibaba, Huawei, Baidu and Ant Group speaking on AI infrastructure, large models and agents (fcf815dd-abaa-43b0-a868-76767b79ce2c)
- [○ 1 SOURCE] Lenovo and Intel supplied South China Agricultural University with WA5480 G3 servers powered by 5th‑gen Xeon Scalable CPUs offering up to 10^13 FLOPS and 64 cores (e0ce1c44-51c6-48dd-a5d8-712efc53778e)
- [○ 1 SOURCE] The robot features open‑source DIY capabilities and entertainment functions such as IP collaborations and companion filming (0a95b83e-6d28-4cf8-9f8e-0fdd92d2fa6c)
- [○ 1 SOURCE] Huang Yiping said AI has not yet significantly raised total factor productivity but will become a core driver of economic growth, recommending an ‘invest‑in‑people’ strategy (6a5b6a3e-1e39-406b-9fe4-1046f5fca256)