Tagged articles
real-time inference
36 articles · Page 1 of 1
miHoYo and Tsinghua University open-source GameScaling, a series of general game-playing models (0.8B–27B) that achieve 63.3% success on Minecraft benchmarks and demonstrate that multi-game training enables 2.8× better transfer to unseen 3D worlds, though scaling alone fails to guarantee far transfer, while real-time inference outperforms frontier models.
GameScalingMinecraftcross-game transfer
0 likes · 13 min read

This article details the architecture of a production-ready AI voice cloning system that clones a voice from just 1 minute of audio, covering audio frontend processing, speaker encoding with WavLM or ECAPA-TDNN2, Flow Matching acoustic modeling, BigVGAN vocoder, inference pipeline latency under 2 seconds, and key engineering challenges like noise robustness, multilingual support, and safety.
BigVGANCosyVoiceECAPA-TDNN2
0 likes · 13 min read

The article explains two deployment patterns for integrating machine learning models into Palantir Foundry's Ontology—real-time function calls and batch inference—so predictions become regular fields on business objects, enabling business users to consume AI outputs without technical expertise, while highlighting governance practices like versioning, confidence intervals, and wrapper functions for safe AI agent usage.
AI governanceBatch InferenceMachine Learning Deployment
0 likes · 10 min read

The article surveys ten community-built demos of the Jev model — a specialized decision-making LLM from a former OpenAI researcher — showing it driving 50 parallel Subway Surfers games, beating Super Mario Bros, piloting a rocket in MuJoCo, analyzing 724 ads in 40 seconds, and filtering AI tool-call histories, all at sub-cent costs and ~0.7 s per decision.
Jevbrowser automationdecision-making LLM
0 likes · 13 min read

The article introduces Jev, a fast, cheap AI model from TypeSafe that handles reflexive decisions while LLMs handle reasoning, showcasing five demos: context compression, ad analysis, real-time Mario gameplay, probability-based animations, and autonomous rocket landing — all at fractions of LLM cost and latency.
AI agentsCost OptimizationJev
0 likes · 7 min read


AntTechAug 20, 2026 · Artificial Intelligence
Ling-3.0-flash is a newly open‑sourced 124B‑parameter MoE model that offers multiple quantized versions, API, single‑machine private deployment, and high‑performance GPU inference exceeding 1100 tokens/s, with detailed benchmarks, optimization techniques, and real‑world use‑case analyses for agents, coding, and sensitive data processing.
Agent applicationsLLMLing-3.0-flash
0 likes · 15 min read

TurboVLA, a real‑time vision‑language‑action model from Huazhong University of Science and Technology and Huawei, bypasses the large language model bottleneck by directly fusing visual and language features, achieving 32 Hz online action prediction on a single RTX 4090 with only 0.2 B parameters and 0.9 GB VRAM, while maintaining high success rates across LIBERO, RoboTwin 2.0, and real‑robot tasks.
LIBEROLLMRTX 4090
0 likes · 11 min read

Qwen-Audio-3.0-TTS launches two variants—Flash with ~300 ms latency and Plus with higher naturalness—offering multilingual support for 16 languages, superior WER/CER and speaker similarity scores, free‑style natural‑language control, fine‑grained tag editing, and robust performance in noisy environments, all backed by benchmark results that crown Plus as the top performer on the Artificial Analysis leaderboard.
AI ModelMultilingual TTSQwen-Audio-3.0-TTS
0 likes · 7 min read

Lychee-FD introduces a native end‑to‑end full‑duplex speech large model that overcomes gradient conflicts and semantic dilution through hierarchical acoustic‑semantic decoupling, delivers up to 28.5% benchmark gains, and is released with open‑source code and demos for digital humans and robots.
ACL 2026full-duplex speechhierarchical acoustic-semantic modeling
0 likes · 13 min read

OmniVoice introduces a single‑stage, diffusion‑style language model that maps text directly to multi‑codebook acoustic tokens, achieving zero‑shot voice cloning for over 600 languages with high intelligibility and real‑time factor as low as 0.025, making it suitable for large‑scale multilingual deployment.
Acoustic tokenMultilingual speech synthesisOmniVoice
0 likes · 8 min read


SuanNiJun 6, 2026 · Artificial Intelligence
JoyAI‑Echo introduces a cross‑modal audio‑visual memory bank, a three‑stage post‑training pipeline, and a Director Agent to enable consistent, high‑quality, real‑time generation of minute‑level videos, achieving up to 7.5× inference speedup and state‑of‑the‑art benchmark scores.
Director AgentJoyAI-Echoaudio-visual AI
0 likes · 13 min read


SuanNiMay 31, 2026 · Artificial Intelligence
Gamma‑World introduces a multi‑agent world model that solves identity, interaction, and real‑time inference challenges with parameter‑free geometric encoding, sparse hub attention, and teacher‑student distillation, enabling zero‑shot generalization from two to four agents and achieving 24 FPS interactive video generation.
Gamma-WorldSimplex Rotary Agent EncodingSparse Hub Attention
0 likes · 11 min read

The article details how the product team tackled AI‑driven challenges by designing a two‑layer, event‑driven Function‑Centric Agent architecture that unifies workflow orchestration and capability supply, enabling real‑time inference for billions of items, cutting development cycles to one person‑week, and boosting search conversion rates.
AI agentAIFunctionFunction Calling
0 likes · 29 min read

The paper identifies real‑time latency as the main obstacle for deploying VLA models on robots, proposes the TTFA metric and the FASTER framework with a Horizon‑Aware Schedule, mixed scheduling and streaming inference, and demonstrates through extensive GPU and task experiments that TTFA and reaction time can be cut by up to three‑fold without sacrificing motion quality.
Embodied AIFASTERRobotics
0 likes · 14 min read

OpenAI’s GPT‑Realtime‑2 embeds GPT‑5‑class reasoning into a continuous‑audio loop, achieving 96.6% accuracy on Big Bench Audio, offering adjustable inference intensity with latency from 1.12 s to 2.33 s, a 128 K context window, and demonstrable gains in real‑world call success rates, while prompting industry debate over pricing and competitive impact.
GPT-5GPT-Realtime-2latency
0 likes · 5 min read

PiLoT, a CVPR 2026 Highlight paper, introduces a neural pixel‑to‑3D registration framework that lets a single‑camera UAV achieve drift‑free 6‑DoF pose and real‑time target locking over 10 km without GNSS, running at 25‑30 FPS on an NVIDIA Jetson Orin and outperforming existing hybrid and absolute‑pose methods.
GNSS-denied navigationUAV localizationmonocular vision
0 likes · 12 min read

GR4AD, Kuaishou’s generative recommendation system, redesigns the entire ad pipeline—from tokenizing multimodal ad material to value‑aware learning, lazy decoding, and dynamic beam search—delivering over 4 % revenue lift, higher eCPM, and sub‑100 ms latency for more than 400 million users.
Advertisinggenerative recommendationonline learning
0 likes · 17 min read


AIWalkerMar 9, 2026 · Artificial Intelligence
The article dissects EFSI‑DETR, a UAV small‑object detector that combines simulated frequency processing with dynamic semantic enhancement to overcome pixel scarcity, static fusion, and ignored frequency cues, delivering 188 FPS and a 5.8% APₛ gain on VisDrone while remaining lightweight.
DETRUAV visiondynamic fusion
0 likes · 16 min read

This article explains Apache Flink’s stream processing fundamentals, introduces the open‑source Flink Agents framework for building event‑driven AI agents, details Alibaba Cloud’s Flink AI Function for real‑time LLM inference, and showcases demos, architecture, integration patterns, and practical use cases such as VOC analysis, live‑stream analytics, and intelligent operations.
Apache Flinkbig datacloud computing
0 likes · 24 min read

The Qwen3‑ASR series, now open‑sourced by Alibaba, offers three models (1.7B, 0.6B, and a 0.6B forced aligner) that cover 52 languages and 22 Chinese dialects, support streaming and offline inference, achieve an RTF of 0.064 with 2000× realtime throughput, handle singing with background music, and provide detailed deployment guides, benchmarks, and comparisons with other ASR solutions.
Qwen3-ASRforced alignermultilingual speech recognition
0 likes · 15 min read

Resemble AI’s open‑source Chatterbox‑Turbo reduces TTS generation from ten steps to one, enabling high‑sample‑rate, lossless voice cloning from a 5‑10 second reference while supporting emotional control, side‑language tags, and embedded watermarking for real‑time applications across chatbots, games, podcasts, and education.
Chatterbox‑TurboNon-autoregressiveText-to-Speech
0 likes · 7 min read

Tencent TEG Shannon Lab won the NTIRE 2025 UGC Video Enhancement competition with a progressive training framework that combines adaptive color enhancement, high‑speed denoising, and temporal stability under bitrate constraints, achieving top subjective scores, significant inference speed‑ups, and successful INT8 quantization for real‑time deployment.
AI video codecNTIRE2025UGC
0 likes · 18 min read

iQIYI’s minute‑level online deep‑learning framework overcomes stability, timeliness, compatibility, delayed feedback, catastrophic forgetting, and i.i.d. constraints through high‑availability pipelines, TensorFlow Example serialization, rapid P2P model distribution, flexible scheduling, disaster‑recovery rollbacks, PU‑loss adjustment, and knowledge‑distillation, delivering a 6.2% revenue boost.
AdvertisingCTR predictionModel Serving
0 likes · 9 min read

The Dynamic Graph Sampling (DGS) service, built on GraphLearn, delivers sub‑20 ms latency for real‑time GNN inference on large, constantly evolving graphs by separating storage from computation, using event‑driven pre‑sampling, lazy multi‑hop concatenation, and a publish‑subscribe architecture that scales linearly across distributed workers.
Alibaba CloudGraph Neural NetworksGraphLearn
0 likes · 12 min read

This article explains the concept of edge intelligence, its advantages over cloud‑based AI, the technical challenges of deploying AI on mobile devices, Ant Group's development timeline, core technology stack, and future directions for edge‑cloud collaboration.
ai-optimizationedge AImobile AI
0 likes · 10 min read

The presentation by Tencent expert Yuan Yi details the Wuliang deep learning system for recommendation, covering its background, technical challenges such as massive data and real‑time requirements, the parameter‑server based solutions for training and inference, model compression techniques, and continuous online deployment strategies.
deep learninglarge-scale trainingparameter server
0 likes · 14 min read

To enable real‑time bullet‑comment passthrough on Youku’s mobile app, the team built a million‑scale portrait dataset and designed the AirSegNet series—CPU, GPU, and server variants—using VGG‑style nets, edge‑aware losses, and hybrid CPU‑GPU inference, achieving 0.98 IoU and sub‑15 ms latency on most devices.
Computer VisionEdge ComputingMNN Framework
0 likes · 13 min read

The article details how to create the AI‑driven “Object Hunt Battle” game by processing data, designing and training a PP‑LCNet model with PaddleClas, converting it for Paddle.js, and integrating real‑time WebGL inference on mobile devices, achieving sub‑50 ms latency and encouraging developers to explore further.
AI game developmentPaddle.jsPaddleClas
0 likes · 9 min read

The talk presents an overview of graph neural network development, explains 极验's graph modeling research and evolution, and details offline and real‑time GCN solutions, including self‑supervised training, large‑scale handling, and performance comparisons, highlighting practical applications in fraud detection and risk control.
GCNGraph ModelingGraph Neural Networks
0 likes · 26 min read

This article details Weibo's multimodal content understanding platform, covering its massive data challenges, heterogeneous model support, standardized pipelines, platformization, workflow architecture, GPU heterogeneous cluster management, resource scheduling, performance optimization, and full‑stack monitoring to achieve stable, low‑latency AI services at scale.
GPU clusterModel ServingWeibo
0 likes · 18 min read

This article explains Alibaba’s User Interest Center approach to real‑time model serving, detailing how it separates offline sequence modeling from lightweight online inference, uses an online interest‑embedding store, and dramatically reduces latency for recommendation models such as DIEN and MIMN.
AlibabaEmbeddingModel Serving
0 likes · 8 min read

Deepthought is iQIYI’s end‑to‑end machine‑learning platform that unifies distributed frameworks, decouples pipeline stages, integrates with Tongtian Tower, and offers visual drag‑and‑drop configuration, evolving from a fraud‑detection prototype to a generic system with real‑time inference, automated hyper‑parameter optimization, and support for large‑scale data across anti‑fraud, recommendation, and analytics workloads.
AI PlatformAutoMLData Engineering
0 likes · 13 min read

The article explains how the fourth paradigm's distributed machine learning framework GDBT tackles the massive data, high‑dimensional features, and real‑time requirements of modern recommendation systems by leveraging heterogeneous computing, parameter servers, RDMA networking, and optimized workloads.
GDBTRDMAdistributed ML
0 likes · 18 min read

The paper presents a lightweight, anchor‑free CenterNet‑based object‑ness detector for WeChat’s Scan feature, built on a ShuffleNetV2 backbone with enlarged 5×5 depth‑wise convolutions, a streamlined detection head, and a Pyramid Interpolation Module, then quantized, ONNX‑converted and NCNN‑deployed to achieve a 436 KB model running in ~15 ms per frame on an iPhone 8 CPU.
CenterNetObject DetectionShuffleNetV2
0 likes · 12 min read

This article details how Alibaba leveraged AI-driven hand‑gesture detection and a lightweight SSD‑based object detection model to create an interactive rock‑paper‑scissors game for Double‑11, addressing challenges of undefined gestures, real‑time mobile performance, and data collection, and achieving over 16 million page views and high accuracy.
Object DetectionSSDfeature pyramid network
0 likes · 22 min read

This paper introduces a Deep Feedforward Sequential Memory Network (DFSMN) for statistical parametric speech synthesis that matches BLSTM quality with only a quarter of the model size and four times faster inference, making it ideal for memory‑constrained, real‑time IoT devices.
DFSMNIoT devicesdeep learning
0 likes · 10 min read
