Follow us to discover more Baidu tech insights.
Within two days, Baidu’s Baige and Kunlun Chip teams adapted the 309‑billion‑parameter MiMo Flash V2 model—featuring a hybrid SWA+Sink and Full Attention mechanism—to run efficiently on the Kunlun P800 XPU using the vLLM‑Kunlun Plugin, achieving lossless performance comparable to GPU inference.
The article explains how Baidu’s Baige team integrated a Context Parallelism strategy into DeepSeek V3.2, detailing the DSA architecture, the limitations of traditional tensor and sequence parallelism, and how CP distributes computation and memory across GPUs to achieve up to an 80 % reduction in token‑to‑first‑token latency for ultra‑long 128K‑token contexts.
The vLLM‑Kunlun Plugin, jointly released by Baidu Baige and Kunlun Chip, provides a high‑performance, zero‑intrusion solution for deploying open‑source large language models on domestic Kunlun XPU hardware, includes fused operators, precision‑validation and profiling tools, and supports over twenty mainstream and multimodal models.
This report analyzes the memory bottleneck of DeepSeek‑V3.2‑Exp’s sparse‑attention decoder, proposes the Expanded Sparse Server (ESS) to offload the latent cache to CPU memory, and demonstrates through high‑fidelity simulation that the approach dramatically improves decode throughput while keeping latency within acceptable limits.
This article explains how traditional kernel network stacks struggle with high‑concurrency, low‑latency cloud data‑center workloads and introduces Baidu Intelligent Cloud’s Polar solution—Polar‑TCP and Polar‑RDMA—which combine user‑space DPDK drivers, a lightweight TCP stack, and an industrial RPC framework to achieve near‑RDMA performance while preserving compatibility with existing TCP ecosystems.
In a round‑table hosted by Baidu Intelligent Cloud, industry leaders dissect how AI agents are transforming Chinese retail and overseas brand expansion, addressing challenges such as rising traffic costs, low repurchase rates, localization hurdles, and demonstrating concrete use cases in content generation, intelligent customer service, and automated marketing that promise to make AI agents an essential, not optional, component of modern commerce.
This article details Baidu's e‑commerce risk‑control overhaul, explaining how traditional rule‑based and manual reviews struggled with multimodal violations, ambiguous semantics, and poor merchant experience, and how a new AI‑driven pipeline combining large multimodal models, rule engines, and knowledge‑base queries achieved full‑automation, real‑time feedback, and high explainability.
This article presents an LLM‑driven intelligent data platform ingestion solution that automates schema recognition, mapping, quality rule extraction, and package building, reducing integration cycles from three months to three days while eliminating manual effort and enhancing scalability and control.
This article details Baidu APP's massive data‑warehouse overhaul, describing the two‑step strategy that stabilized log cleaning, modernized the ETL framework, introduced wide‑table architectures, and implemented tiered storage to dramatically improve processing speed, reliability, and cost efficiency for petabyte‑scale workloads.
This article explains how Baidu’s Feed real‑time data warehouse was rebuilt using a pure streaming architecture, detailing the limitations of the previous stream‑batch design, the technical solutions—including core/non‑core data separation, metric calculation in streaming, and Parquet storage with Apache Arrow—and the resulting cost reductions, latency improvements, and future roadmap.
