Official account of the Xiaohongshu tech team, sharing tech innovations and problem insights, advancing together.
In a 1024 Programmer's Day AMA, Xiaohongshu’s tech leaders and Hugging Face co‑founder discussed AI coding as a core skill, open‑source as a catalyst for rapid advancement, and the challenges of achieving human‑like AGI, offering practical takeaways for developers and product teams.
The article details Xiaohongshu’s Dataverse platform evolution into a Data+AI system, highlighting inefficiencies in algorithm and data‑science workflows, the introduction of an interactive notebook, comprehensive data lineage, AI‑coding assistance, and future DataAgent plans to automate data engineering tasks.
FireRedTTS-2, the latest open‑source dialogue TTS model from Xiaohongshu’s audio team, upgrades its speech tokenizer and text‑to‑speech architecture to enable low‑latency, per‑sentence generation, robust multi‑speaker switching, and natural prosody across multiple languages, outperforming rivals in both objective and subjective tests.
RedHub is a next‑generation database proxy built by Xiaohongshu that unifies fragmented access methods, leverages PolarDB‑X for distributed SQL execution, and delivers high‑performance, highly available, and easily observable database connectivity, enabling seamless migration and advanced features for massive‑scale workloads.
The paper introduces Single Trajectory Distillation (STD), a novel training framework that aligns full PF‑ODE trajectories from a fixed noisy state, uses a Trajectory Bank to cut training cost, and adds an Asymmetric Adversarial Loss to markedly improve style consistency and aesthetic quality while accelerating image and video style‑transfer diffusion models.
DynamicFace introduces a controllable face‑swapping framework that leverages composable 3D facial priors, dual‑stream identity injection, and a FusionTVO module to achieve superior image and video quality, identity preservation, and temporal consistency, outperforming existing state‑of‑the‑art methods on benchmark datasets.
This article details how XiaoHongShu's infrastructure team built a keyless architecture that offloads CPU‑intensive TLS private‑key signing to Intel QAT hardware, achieving massive HTTPS throughput gains, lower server costs, and valuable insights for similar high‑traffic TLS offload scenarios.
This article explains how XiaoHongShu’s database team built a lightweight Binlog Server to accelerate semi‑synchronous MySQL replication beyond 300 MB/s, achieve RPO=0 data‑loss‑free failover, and improve high‑availability without manual intervention, backed by performance tests and detailed architecture diagrams.
dots.vlm1, the first open‑source multimodal large model from Xiaohongshu hi‑lab, combines a 1.2‑billion‑parameter NaViT visual encoder with DeepSeek V3 LLM, achieving near‑state‑of‑the‑art visual understanding and reasoning while remaining competitive on text tasks, and is available on GitHub and HuggingFace.
dots.ocr is a 1.7 billion-parameter multilingual document-parsing model that unifies layout detection and content recognition within a single visual-language model, delivering state-of-the-art performance across text, tables, formulas and reading order while remaining efficient and extensible for future multimodal AI research.
