KubeCon + CloudNativeCon + OpenInfra Summit + PyTorch Conference China 2026
2026-09-07..2026-09-09
Shanghai, China
as the organisers put it
Join CNCF, OpenInfra, and PyTorch communities to shape the future of next-gen AI and open source with OpenStack, Kubernetes, and PyTorch.
| title | author | length |
|---|---|---|
| Tidal Auto-scaling for Training and Inference Based on Kubernetes + KEDA | Jun Zheng, Tan Pei Xiang | 07m |
CMB AI assistants exhibit distinct peak and off-peak traffic patterns. During off-peak hours, the platform allocates idle GPUs to LoRA incremental training. As… | ||
| From Open Source Adoption To Open Source Innovation: China’s Next Chapter | Lu Shouqun | 02m |
Traces the shift from copying foreign code to upstream contributions and community-run projects, citing growing participation in Kubernetes and PyTorch… | ||
| Evolving Argo Workflows for AI: New Features, Tuning, and Multi-Tenant Isolation | S. Tian, Y. Su | 25m |
Over the past year, Argo Workflows adoption in AI has surged, particularly in China. Key adopters include BYD, which runs autonomous driving pipelines on Argo… | ||
| 10 Years of Cilium: Connecting, Securing, and Simplifying the Cloud Nat... | L. Huang, T. Ye, Y. Su | 31m |
Join us as we celebrate a decade of Cilium, now the de-facto standard CNI for Kubernetes and a cornerstone of cloud native networking and security. This… | ||
| Dragonfly V2.5.0 - Intro, Updates, Data Distribution in AI Infrastructure | Wenbo Qi, Chenyu Zhang | 30m |
Dragonfly provides efficient, stable, and secure file distribution and image acceleration using P2P technology within cloud-native architectures. This talk… | ||
| Project Lightning Talk: Atlantis: Terraform Pull Request Automation for Cloud Native Teams | R. Chen | 06m |
Atlantis is a CNCF Sandbox project that provides Terraform and OpenTofu pull request automation for cloud native teams. In this 5-minute Project Lightning… | ||
| Project Lightning Talk: KubeEdge Everywhere: Latest Project Update With Industrial Cases | H. Zhang | 09m |
Since becoming the first cloud-native edge project to graduate from the CNCF, KubeEdge adoption has exploded. From intelligent transportation and smart cities… | ||
| Operating Frontier Intelligence at Scale | Chris Aniszczyk, Xiao Zhang | 20m |
Machine workloads shift from training to inference with commits jumping from ~1.4B to 2.9B in months and web traffic driven more by bots and agents, causing… | ||
| Beyond Multimodal: Building Full‑Modal AGI Via Open‑Weight | Ryan Lee | 04m |
Beyond Multimodal: Building Full‑Modal AGI Via Open‑Weight - Ryan Lee, MiniMax | ||
| Hibernate and Evolving: Taming OpenClaw With OpenKruise Agents | Zhang Zhen | 34m |
OpenClaw's rapid adoption has highlighted critical "Day-2" operational challenges: high costs of idle agents, sandbox instability, and complex version… | ||
| Beyond Model Sharding: Atomic Scheduling and Disaggregated LLM Serving With ... | K. Yan, C. Zicong | 39m |
LLM serving on Kubernetes is moving beyond model sharding. In production, one inference replica may be a coordinated group of pods across nodes, GPUs, and… | ||
| Project Lightning Talk: Closing Remarks | Miley Fu | 01m |
This is the final scheduled talk followed by an afternoon session of lightning talks at 4:15 p.m. in Pearl Hall. A staffed Women in Tech booth, table eight… | ||
| Project Lightning Talk: Let’s Talk About Headlamp, the Extensible Open-Source Kubernete... | E. Chen | 05m |
Headlamp is a CNCF Sandbox Kubernetes UI available as a desktop app or in-cluster web UI that uses Kubernetes RBAC so users only see what their kubeconfigs… | ||
| Project Lightning Talk: Run Sandbox Securely and Cost-Effectively With OpenKruise Agents | Z. Zhen | 05m |
Running sandboxes for LLM-generated code in production forces an uncomfortable three-way trade-off between security, latency, and cost. The sandbox executes… | ||
| Project Lightning Talk: Multi-cluster Orchestration System: Karmada Updates and ... | H. Ren, Y. Ci | 07m |
Karmada (Kubernetes Armada) is a Kubernetes management system that enables you to run your cloud-native applications across multiple Kubernetes clusters and… | ||
| Project Lightning Talk: Architecture Without Gatekeepers: Decoupling for Neutrality | Sergey Pronin | 06m |
Project moves from a Percona‑centric monolith to a tiny core with a modular plugin system that removes operator lock‑in, donated to CNCF sandbox, adds a simple… | ||
| Project Lightning Talk: Opening Remarks | Miley Fu | 01m |
Host introduces a lightning track of CNCF projects with five-minute lightning talks and points to a GitHub link for Alaris, an AI operating system that aims to… | ||
| Agile Readiness: Leveraging CI Relay and Full-Scale Test Reuse | Jiawei Li, Jingwei Huang | 30m |
Adapting to PyTorch’s rapid bi-monthly release cycle is a challenge for heterogeneous accelerators. This session shares Ascend’s practices in maintaining high… | ||
| Cybertwin-based Cloud Native Network (CCNN): Network Architecture Innovation ... | D. Liang, W. Lan | 25m |
Driven by IoT and AI, future networks demand an architectural revolution. A CCNN is proposed that converges RAN, IP bearer and data center networks, using… | ||
| Self-Healing Rollouts: Automating Production Fixes With Agentic AI | Kevin Dubois, Daniel Oh | |
Demonstrates an Argo Rollouts plugin calling an agentic pipeline that uses Quarkus Java + a Java LangChain to analyze Prometheus metrics, logs and… | ||
| Minutes, Not Weeks: Long-Term Memory for AI Applications With PowerMem | Yihang Li | 28m |
Every AI app eventually needs memory. Most teams hit the same walls: choosing a backend locks them into a later rewrite, wiring retrieval and lifecycle bleeds… | ||
| KernelAgent: Hardware-Guided GPU Kernel Optimization Via Multi-Agen... | K. Cheng, J. Khuu, L. Wang | 18m |
KernelAgent adds a hardware-guided optimization layer on top of an agentic multi-agent loop that runs real Triton kernels in a GPU sandbox, calls vendor… | ||
| Exploring HiFloat8: A Tapered Format Complementing the FP8 Ecosystem for Robust Model T... | Y. Zhao | 26m |
Standard FP8 formats suffer from frequent gradient overflows and heavy reliance on complex Delayed Scaling, which often lead to training instability or… | ||
| Closing Remarks | 01m | |
Schedule note: tomorrow returns at the same time with more keynotes and more demos, some of which are explicitly uncertain — demo reliability is unknown and… | ||
| Building an Agent Runtime With Open Infrastructure | Yaya Xia, Xu Wang | 09m |
An agent runtime has to create a working environment on demand, isolate code the platform may not trust, and release resources when the task ends. Much of this… | ||
| From Containers To Agents: The Next Cloud Native | Di Xu | 06m |
Argues agents become a new unit of autonomous work and proposes an agent infrastructure that treats them as first-class Kubernetes workloads, not just LLM… | ||
| PD Disaggregation VLLM Deployment on Alternative AI Accelerators Using Llm-d | Wang Jifei, Mengxuan Li | 09m |
PD Disaggregation VLLM Deployment on Alternative AI Accelerators Using Llm-d - 纪飞 王 & Mengxuan Li | ||
| ⚡ Eliminating Adaptation Latency: Day-0 Inference for New Models on Multi-Backends With... | X. Wang | 06m |
As LLM iterations accelerate, a significant "adaptation latency"—the days or weeks required to port models from GPU to heterogeneous hardware delays product… | ||
| ⚡Edge Resource Control in KubeEdge:Safely Holding and Releasing Pod Updates for Critical... | F. Gao | 04m |
KubeEdge v1.22.0 adds an annotation edge.kubeedge.io/hold-upgrade: "true" that makes edged intercept and hold new Pod creations in memory until local logic… | ||
| ⚡ Can Stateful AI Agents Scale To Zero? | Peng Li | 05m |
AI agents are often treated as always-on workloads because they keep session context, invoke tools, and may depend on sandboxed runtimes. Knative, by contrast… | ||
| 📚 Building Portable Large Models on Heterogeneous AI Accelerators Wit... | M. Lyu, H. Zhang, Y. Lyu | 01h 07m |
Porting large models across diverse accelerators and running them efficiently remains difficult. Developers can download LLMs and multimodal models such as… | ||
| Dragonfly's CNCF Graduation Journey and Large-Scale AI Model Distribution in Pra... | F. Lyu, W. Qi | 25m |
In 2025, after nearly eight years of incubation within CNCF, the Dragonfly project officially graduated. But getting there wasn't easy. Growing from an… | ||
| Spec-Driven Evolution: How AI Agents Are Rewriting the OpenStack/K8s Pla... | B. Souville, M. Hamza | 30m |
Spec-driven workflow uses three AI agents (requirements, implementation analysis, runtime evidence) that read specs, code and live clusters to produce… | ||
| Architecting Secure Agentic Workflows on Kubernetes: A Financial Sector... | V. Caldeira, M. Foster | 40m |
Kubernetes-based reference architecture uses Rosoctl CRD + sidecar to inject short‑lived SPIFFE identities, Keycloak JWTs and route tool calls through an MCP… | ||
| 8 Million Tasks Daily: Driving Efficiency in Multi-Tenant Data Platforms at Ho... | J. Sun, S. Tian | 25m |
As autonomous driving shifts to end-to-end models, Horizon Robotics faces unprecedented scale. Thousands of concurrent, sub-minute workflows pushed… | ||
| From Chatbots To Agentic AI: Running NVIDIA Dynamo the Kubernetes Way | Xianglong Lu, Bo Chu | 38m |
Agentic AI is redefining inference on Kubernetes. Long tool-call workflows create sustained KV-cache pressure; parallel subagents need shared prefixes and… | ||
| Towards Device-agnostic PyTorch: Building Unified Infrastructure for a Multi-Backend Ecos... | W. Li | 05m |
Cambricon describes hardening PyTorch for PrivateUse1 backends by contributing device-agnostic fixes upstream: over 100 pull requests across PyTorch, NiteI… | ||
| Zuul and OpenClaw, Safe Agentic Coding Via Human Collaboration Tools | Monty Taylor | 38m |
Claim: treat AI agents like untrusted contributors and use existing review and CI tooling to control them. Uses OpenClaw as agent gateway, Matrix for chat… | ||
| 📚 Advanced NCCL Tutorial: Beyond Collectives | Jeff Hammond, Hanyue He, Ke Wen | 01h 17m |
NCCL now exposes device-side primitives that let CUDA kernels read/write remote buffers and drive RDMA: symmetric memory avoids per-call registration to cut… | ||
| Gravity of Ecosystems: Pollinating the Modern AI Orchestrator | Wenjia Zhang, Bo Fu | 36m |
Claims that shifting core contributors into Ray, Slurm, PyTorch over a year produces concrete Kubernetes changes: implemented Dynamic Resource Allocation and a… | ||
| Building AI4S Systems for Scientific Research on Kubernetes | Jiamu Liu, Yanjun Chen | 37m |
Researchers are exploring how AI workflows can accelerate science in recent AI waves. Most existing workflows are built for AI/ML model training, and they fall… | ||
| Zero-Trust Traffic Governance for Kubernetes AI Agent Sandboxes | Bingshen Wang, Bo Kang Li | 30m |
When AI agents run in Kubernetes, network access can no longer be treated as a static allowlist problem. An agent may execute generated code, choose tools at… | ||
| Scaling Digital Employees at China Merchants Bank in Production: A Controlled AI Agents A... | J. Xu | 42m |
China Merchants Bank runs 20,000+ digital employees with centralized templates and an AI Agent-as-a-Service platform that enforces identity, stateful… | ||
| Ascend & PyTorch: Pioneer New AI Open Ecosystem | Liang Zhang | 05m |
As AI models grow larger and their architectures become increasingly sophisticated, the industrial ecosystem faces new opportunities and challenges. Drawing on… | ||
| Redefining LLM Training Through Atomic Components and Serverless TaaS | Tan Pei Xiang | 39m |
RL (e.g., GRPO) is vital for LLM reasoning, yet current tools are either too rigid or engineering-heavy. Twinkle, co-developed by ModelScope and China… | ||
| What AI Agents Need From Open Infrastructure | Yaya Xia | 05m |
Maps 84 agent-infra projects and finds 32 user-facing apps, 31 runtime projects and 23 new entries since May (13 brand new), arguing agents create short-lived… | ||
| Building Frontier Intelligence in the Open | Mark Collier | 11m |
PyTorch Foundation now hosts six projects and PyTorch sees 44% contributor growth in one year while VLM grows 185%. Open models get day‑zero support for Kimiko… | ||
| Training Through Failures: How Meta Keeps 100k-GPU Jobs Alive With Open-Sou... | T. Rice, A. Afzali | 25m |
Meta reports keeping 100k‑GPU training jobs alive by treating GPU failures as expected and recovering per step in under 30 seconds for many cases. Open-source… | ||
| Escaping the Vendor Trap: A Journey for Migrating Legacy Infrastructure To Op... | T. Nguyen, Q. Le | 33m |
By 2018 Viettel runs ~18,000 servers across 11 countries on VMware and multi‑vendor storage, low 20–25% utilization and huge MTTR force a move to OpenStack… | ||
| A Cloud Native Stack From Bare Metal To Tokens for Large-Scale AI Inference | Nguyen Trong Vinh | 04m |
Platform unifies a mixed fleet of T4, L40S, A100, H200 and B200 GPUs into one managed stack by layering OpenInfra (OIF) at bare metal, CNCF tooling for GPU… | ||
| Building Frontier AI Infra: SGLang and Miles | Ke Bao | 06m |
A model’s real-world performance gets decided in the serving and post-training layer: how fast it runs, what it costs, and how quickly people can build on it… | ||
| Meet the Community Behind the Open Source AI Stack | Jane Lyu, Fupan Li, Zesheng Zong, Hiu Yeung, Sunny Chan | 10m |
No single project, company, or community builds the AI stack alone. Behind the models, frameworks, cloud native technologies, and infrastructure powering AI is… | ||
| HyperParallel: A SuperPoD-Aware Distributed Acceleration Library | Teng Su, Shendi Wang | 09m |
PyTorch-based library that optimizes distributed training for Ascend SuperPoD and demonstrates on a Huawei Atlas 800T cluster with a Qwen3-30B-A3B workload… | ||
| What Powers Frontier Intelligence | Thierry Carrez | 10m |
Claims that infrastructure is the gating factor for frontier AI and that open source is the practical path to control, cost savings and compliance. Cites… | ||
| ⚡ Speedy but Sturdy PyTorch Optimizers | Jane Xu | 06m |
PyTorch reduces memory round trips with horizontal fusion via torch.foreach and vertical fusion via torch.fusedadamw so fewer kernels run; on CUDA 13+ kernels… | ||
| ⚡Empower Open Edge Computing on RISC-V | Tiejun Chen | 05m |
LF Edge - Linux Foundation Edge, is aiming to establish an open, interoperable software framework for edge computing, independent of hardware, silicon, cloud… | ||
| VeRL: Extreme Optimization Practices for Ultra-Large-Scale MoE Models | Carson Wu | 31m |
Several leading internet companies (including ByteDance, JD.com, etc.) have built core RL training infrastructure based on veRL to support AI businesses… | ||
| Welcome Back + Opening Remarks | Jonathan Bryce, Horace Li | 03m |
Concrete claims about how to build "token factories" for AI workloads appear: neo clouds use OpenStack to provision specialized hardware, Kata containers to… | ||
| VLLM KV Cache Management: From Cache Reuse To Agent Scenario Optimization | Mengqing Cao, Wang Xiyuan | 34m |
As LLM Agents proliferate, the conflict between surging KV Cache and limited VRAM in multi-turn dialogues and tool-chaining has become a key bottleneck. This… | ||
| Turning Fragmented GPU Clusters Into One Elastic Compute Pool | Chongkang Tan | 33m |
As AI infrastructure evolves, fragmented GPU clusters and bursty AI workloads make large-scale multi-cluster operations increasingly difficult.
In this talk… | ||
| Redesigning OpenStack I18n for the AI Era: Weblate & Zero-GPU AI | Seongsoo Cho, DaGyeong Kim | 30m |
Migrated decades of PO/POT data across 50+ languages into Weblate migration while preserving plural rules, header normalization and compileability using… | ||
| To Cache or Not To Cache? a Tiered KVCache Storage System for Agent ... | J. Zhang, W. Cong, y. bai | 25m |
Abstract:
KVCache reuse is a critical lever for reducing inference costs, with scenarios like OpenClaw and Code Agents achieving reuse rates exceeding 90%… | ||
| Secure AI Agent Sandboxing With OpenStack Zun and Kata Container | Wenjie Bao | 28m |
A secure sandbox built natively on OpenStack runs hundreds of ephemeral Kata-backed containers managed by Zun, using Glance for images and Neutron for… | ||
| Serving Qwen at Scale: Multi-Cluster AI Infrastructure on Karmada | Jionghang Cai | 04m |
Serving Qwen requires infrastructure at massive scale: training and serving large language models across multiple Kubernetes clusters and vast GPU resources… | ||
| Solving Industrial Challenges With KubeEdge: A Post-Graduation Report | Yue Bao, Huan Wei, Yin Ding | 26m |
Following its graduation within the CNCF, KubeEdge has solidified its position as the premier platform for extending Kubernetes to the edge. In this session… | ||
| Multi-cluster Orchestration System: Karmada Updates and Use Cases | Hongcai Ren, Zongqing Li | 34m |
Karmada (Kubernetes Armada) is a Kubernetes management system that enables you to run your cloud-native applications across multiple Kubernetes clusters and… | ||
| ⚡ Before VLLM Starts: Preflight Checks for LWS for LLM Inference on K8S | Peter Pan | 07m |
Defines an LWS-based startup gate that runs init-container preflight checks and keeps them decoupled from vLLM, with a concrete KEP implementation LWS KEP #813… | ||
| What We Learned Securing AI Agents at Scale on Multi-Tenant Kubernetes Clusters | Pengfei Ni | 25m |
Giving every engineer a personal AI agent sounds great until you put hundreds of them on a shared Kubernetes cluster. Each agent executes arbitrary code, calls… | ||
| The State of the PyTorch Ecosystem in 2026: Global Trends and China’s Rising Role | Wei Wang | 18m |
In 2026, PyTorch stands at the center of a rapidly expanding global AI ecosystem. Through a data-driven analysis of activity across GitHub, Hugging Face, and… | ||
| Closing Remarks | 03m | |
A working token factory is described that stitches OpenStack, Kubernetes, LLMD, Hammy, PyTorch and vLLM into a single pipeline that actually generates tokens… | ||
| Build a Unified Heterogeneous AI Computing Ecosystem for PyTorch | Zesheng Zong | 04m |
AI computing hardware faces severe heterogeneity challenges with high adaptation costs and lack of unified standards. Co-chaired by Huawei and Intel, PyTorch… | ||
| Road From Kata Containers To Confidential Containers + GPUs: From First Commit To CNC... | Z. Kaiser | 05m |
Starts from first issue Nov 2021 and a first PR and delivers working Confidential Containers that run unmodified GPU workloads inside a VM-based trust boundary… | ||
| Helion's New CuTe DSL and Pallas Backends and How We Built Them Via Agents | Dunfan Lu, Jason Ansel | 41m |
Helion adds two new backends: a CuteDSL backend targeting NVIDIA Blackwell and a Pallas TPU lowering that emits VM-prefetched pipelines. Autotuner chooses… | ||
| KV Cache: Accelerating AI Inference on Intel CPU | Bin Yang | 15m |
Serving LLM is bottlenecked by one scarce resource: HBM. The KV Cache grows with every token and every concurrent request, capping context length, throughput… | ||
| Accelerating RL With AgentCube: Cloud Native Multi-Agent Collaboration | Zhencheng Lee | 19m |
Reinforcement learning workloads often require multiple agents to start concurrently, interact with environments, evaluate rewards, and iteratively update… | ||
| ⚡Building a Multi-Cluster Progressive Delivery Platform With Karmada & Argo | Zhuang Zhang | 06m |
As enterprises scale across hybrid clouds, the gap between "cluster scheduling" and "release control" remains a significant hurdle. While many understand why… | ||
| ⚡ Fast Restarts, Not Just Fast Starts: Accelerating Pod Recovery | Baofa Fan | 08m |
Pod startup optimization is no longer enough; restart latency is now a core bottleneck for reliability and cost, especially in AI/ML and batch workloads. This… | ||
| Beyond Static Pods: Dynamic GPU Sharing and Low-Latency Model Switching for LLM ... | Xi Wang, J. Duan | 34m |
Kubernetes revolutionized microservice orchestration, but Model-as-a-Service exposes a fundamental gap, where MaaS must handle thousands of heterogeneous… | ||
| End-to-End Observability for LLM Inference: From Token To GPU | Jared Tan, Murphy Chen, Nicole Li | 32m |
LLM inference is becoming the most expensive—and hardest to observe—workload on cloud-native clusters. Unlike stateless microservices, responses are streamed… | ||
| Accelerating Sequential Recommender Models in the PyTorch Ecosystem | Zan Huang, Jiayu Sun | 34m |
Accelerating Sequential Recommender Models in the PyTorch Ecosystem - Zan Huang & Jiayu Sun, NVIDIA
Sequential recommendation models share a natural… | ||
| Behind the Scenes of Building a Gateway API Dataplane... | Norwin Schnyder, Huabing Zhao | 31m |
Compares two Envoy-based Gateway API implementations and gives concrete trade-offs: default deployment uses one Envoy fleet per gateway in controller land… | ||
| Write Once, Run Anywhere: PyTorch’s Generalization Journey | Yu Guangye, Eikan Wang | 31m |
PyTorch builds a unified path so one model runs unchanged across CPU, MPS, and new vendor backends by replacing fragmented pieces with concrete components… | ||
| From Cloud To AI: How Kata Containers 4.0 Reinvents the Sandbox for the Agent Era | F. Li, X. Niu | 25m |
Kata Containers 4.0 replaces Go with a Rust-based runtime that collapses multiple processes into one, adds multi-hypervisor support (Dragonball, QEMU, Cloud… | ||
| In-Place PVC Re-Binding: Zero-Downtime Disk Migration Using Only the Kub... | M. Nazarenko, N. Bora | 22m |
Migrates ~500 Azure managed disks to the CSI driver across many clusters in ~2 months with no severity-1 incidents by using only standard Kubernetes APIs… | ||
| Why Your TTFT Lies: Diagnosing PD-Disaggregated LLM Inference With Minimal Cross... | N. Li, K. Liu | 31m |
Why is inference slow? You have metrics from frameworks/engines/GPUs, but they don't tell you what's wrong. Prefill queueing? Decode memory‑bound? KV transfer… | ||
| Pathless First: Redefining Runc Container Security Against ProcFS-Based Attacks | Fubang Li | 30m |
ProcFS has been the most persistent attack surface for runc container escapes, accounting for 60% of path-related runc CVEs from 2017 to 2025. Incremental… | ||
| VLLM-Helion: SOTA LLM Performance by Advanced Autotuning and Fine-grained Dispatching | Sean Chen | 26m |
vLLM integrates PyTorch Helion to produce pre-tuned, platform-specific kernels via an offline autotuner that searches representative input shapes and a… | ||
| Project Lightning Talk: From Static Slices To Elastic GPUs: Dynamic MIG With HAMI | Wang Jifei | 07m |
Project Lightning Talk: From Static Slices To Elastic GPUs: Dynamic MIG With HAMI - 纪飞 王, Dynamia
In Kubernetes, NVIDIA MIG is typically pre-partitioned… | ||
| Open Source for the AI Era | Jonathan Bryce, Horace Li | 11m |
Introduces a claim that open source can span the full AI stack from chip to agent and shows concrete tooling connecting those layers: OpenStack, Kubernetes and… | ||
| From Disk Images To Container Images: Deploying Bootc To Bare Metal With Kubernetes an... | S. Baker | 23m |
Demonstrates deploying a bootable OS packaged as a bootc OCI container image so OpenStack Ironic (Flamingo/Gaspacho) with Metal3/metal‑cubed autodetect pulls… | ||
| One Size Fits None: Lightweight Workload-Specialized Confidential Containers | Kailun Qin | 25m |
Benchmarks show Confidential Containers (CoCo) spends about 7 seconds startup and incurs roughly ~250 MB overhead per Pod VM because early Linux boot and… | ||
| Serving AI at Massive Scale: The Cloud Native Inference Plane Behind Huawei... | Kevin Wang, Jun Ding | 39m |
Serving AI at Massive Scale: The Cloud-Native Inference Plane Behind Huawei Celia - Kevin Wang & Jun Ding, Huawei
As AI shifts to integrated mobile… | ||
| How Intsig Serves Billions of Document Scans: GPU Virtualization at... | Mengxuan Li, Walter Duan | 22m |
How Intsig Serves Billions of Document Scans: GPU Virtualization at Scale with HAMi Mengxuan Li, Dynamia.ai & Walter Duan, intsig information co. ltd
Intsig… | ||
| Inside VLLM: Production Best Practices, Model Integration and Road Map | Kaichao You | 02m |
vLLM presents concrete production moves: a new Rust frontend, a KV connector, and staged reinforcement learning rollouts aimed at lower latency and easier… | ||
| Volcano: A Unified Scheduling Platform for Cloud Native AI | Chen Zicong, DongYang Wang | 30m |
Volcano: A Unified Scheduling Platform for Cloud Native AI - Chen Zicong, Volcano Maintainer, Software Engineer at Huawei Cloud, Huawei & DongYang Wang, Cloud… | ||
| 8 Million Tasks Daily: Driving Efficiency in Multi-Tenant Data Platforms at Hori... | J. Sun, S. Tian | |
8 Million Tasks Daily: Driving Efficiency in Multi-Tenant Data Platforms at Horizon Robotics - JIanxiang Sun, Horizon Robotics & Shuangkun Tian, Alibaba… | ||
| Kubernetes DRA Architecture: Scheduling, Status, and Topology at Scale | Paco Xu, Kang Zhang | 34m |
AI/HPC workloads need more than “two GPUs”: they need GPUs, NICs, NUMA locality, and PCIe/fabric relationships. This session presents Kubernetes Dynamic… | ||
| Non-Invasive AI Agent Observability With OBI | Haibin Zhang, Endre Sara | 29m |
OBI uses eBPF kernel probes to intercept network traffic and extract plain‑text MCP JSON‑RPC, MCP Streamable HTTP, embedding and reranking calls, and tool… | ||
| All-or-Nothing No More: Taming Kubernetes Impersonation With Granular Controls | Jian Qiu | 24m |
KEP-5284 rolls out constrained impersonation in beta, adding new authorization verbs that limit which identities can be impersonated and which API verbs the… | ||
| Declarative Underlays: Scaling Purpose-Built Infrastructure(Clusters)... | HanSol Park, Kangsub Song | 31m |
Runs OpenStack on Kubernetes using Cluster API to declare and manage many purpose-built clusters, consolidating control planes into a management cluster where… | ||
| Project Lightning Talk: Let's Talk About Perses, the Open Specification for... | Leon Nunes | 04m |
Perses is an open CNCF sandbox specification for dashboards that runs as dashboards as code (JSON or code), provides embeddable NPM components, and is… | ||
| nothing matches | ||