Unlocking AI Large Model Potential: Key Takeaways from Shanghai HCDG AI Model Salon

Amid the rapid surge of artificial intelligence, large‑model technologies are becoming the driving force behind digital transformation, prompting a call to further unleash their imagination and accelerate practical applications.

On May 26, 2024, Huawei Cloud, Kunpeng Club, and Chuangxiang Dream Space co‑hosted the HCDG City Tour Shanghai Station – AI Large Model Special Practice Salon, gathering developers and technical professionals to discuss the latest trends and experiences with Huawei Cloud Ascend services, Retrieval‑Augmented Generation (RAG) models, and LLM large models.

Huawei Cloud Shanghai Developer Ecosystem Director delivered a keynote titled “Open‑Source LLM Application Practice Based on ModelArts,” demonstrating one‑click deployment of the ChatGLM2‑6B inference model via AI Gallery on Ascend Cloud and building a knowledge‑base application using the LangChain framework.

CEO Zhang Yingfeng of InfiniFlow (Shanghai) analyzed the current state and future of RAG, introducing the open‑source RAGFlow product and the Infinity AI‑native database, which integrates vector search, engine search, and structured data queries to enhance multi‑path retrieval for enterprise‑level search, recommendation, dialogue, and personal assistants.

Observability Cloud product architect Liu Rui presented best practices for monitoring LangChain‑based LLMs, emphasizing the importance of tracing internal model states, analyzing input/output parameters, and fine‑tuning prompts. He also highlighted Guance Cloud’s end‑to‑end tracing platform, supporting over 400 technology stacks with components such as DataKit, DQL, GuanceDB, and an intelligent monitoring suite.

The event concluded that large‑model technology will continue to empower various industries, and Huawei Cloud will keep collaborating with partners to explore limitless AI possibilities and drive innovation in the digital and intelligent era.