Tag with "deep learning"

August 5

Alibaba and Nankai University Unveil LLaVA-Scissor for Video Model Compression

2025年8月5日

Alibaba's Tongyi Lab and Nankai University have jointly developed LLaVA-Scissor, a groundbreaking video large model compression technology. The innovation addresses challenges in video processing efficiency by reducing token numbers while preserving semantic information. The method shows superior performance in benchmark tests, particularly at low token retention rates, marking a significant advancement in video AI.

DAMN

2025

July 28

Alibaba Open-Sources MoE-Based Video Generation Model

2025年7月28日

Alibaba's cloud division has open-sourced Tongyi Wanxiang Wan 2.2, the world's first video generation model using Mixture of Experts (MoE) architecture. The release includes three core models offering significant improvements in computational efficiency and cinematic quality, with one model deployable on consumer GPUs.

DAMN

2025

July 23

Kimi K2 Technical Report Reveals Model's Open-Source Dominance

2025年7月23日

The Kimi team has released the technical report for their K2 model, detailing its 1 trillion parameters and innovative training techniques that propelled it to win a global open-source competition. Key advancements include the MuonClip optimizer and a unique 'restatement method' for improved data efficiency.

DAMN

2025

July 9

Vidu Q1 Upgrade: AI Video Generation Now Supports Seven Reference Images

2025年7月9日

Vidu, a leader in AI video generation, has unveiled a major upgrade to its Q1 model, introducing a 'Reference-to-Video' feature that allows users to upload up to seven images for highly consistent 1080p video output. This breakthrough addresses multi-scene and multi-subject consistency challenges, offering creators unprecedented flexibility.

DAMN

2025

July 8

Recurrent Models Breakthrough: 500 Steps to Handle Ultra-Long Sequences

2025年7月8日

Researchers have developed a method enabling recurrent models to process sequences up to 256k in length with just 500 training steps. This breakthrough challenges Transformer dominance in long-sequence tasks, offering new potential for AI applications in data-intensive fields.

DAMN

DLoRAL: Breakthrough Open-Source Video HD Enhancement Tool

2025年7月8日

Hong Kong Polytechnic University and OPPO Research Institute have jointly developed DLoRAL, an open-source framework using diffusion models for efficient video super-resolution. The dual LoRA architecture ensures temporal consistency and spatial detail enhancement, offering tenfold faster inference speeds than traditional methods while maintaining superior visual quality.

DAMN

2025

July 4

DeepSeek R1 Enhanced Version Boosts Efficiency by 200%

2025年7月4日

TNG's enhanced DeepSeek-TNG-R1T2-Chimera model achieves a 200% improvement in inference efficiency and reduces costs using the innovative AoE architecture. This hybrid model outperforms standard versions in benchmark tests, showcasing advanced reasoning capabilities and cost-effectiveness.

DAMN

2025

July 1

TEN VAD Open-Source Release: A Leap in Voice Detection Tech

2025年7月1日

The TEN Agent team has open-sourced its enterprise-level voice activity detector, TEN VAD, offering superior accuracy and low latency compared to existing solutions. This tool enhances real-time dialogue systems and supports cross-platform deployment, marking a significant advancement in voice AI technology.

DAMN

2025

June 10

NVIDIA and HKU Breakthrough: New AI Model Speeds Up High-Res Image Generation 84x

2025年6月10日

NVIDIA and The University of Hong Kong have developed GSPN, a revolutionary visual attention mechanism that accelerates high-resolution image generation by over 84 times while preserving spatial coherence, setting new benchmarks in computer vision.

DAMN

2025

May 12

CAS Team Unveils MCA-Ctrl: A Game-Changer for AI Image Customization

2025年5月12日

Researchers from the Chinese Academy of Sciences have developed MCA-Ctrl, a revolutionary text-to-image technology that enables precise customization without model fine-tuning. The breakthrough addresses key industry challenges in controllability and complex scene handling.

DAMN

AI D-A-M-N

Tag with "deep learning"

Tags

August 5

Alibaba and Nankai University Unveil LLaVA-Scissor for Video Model Compression

July 28

Alibaba Open-Sources MoE-Based Video Generation Model

July 23

Kimi K2 Technical Report Reveals Model's Open-Source Dominance

July 9

Vidu Q1 Upgrade: AI Video Generation Now Supports Seven Reference Images

July 8

Recurrent Models Breakthrough: 500 Steps to Handle Ultra-Long Sequences

DLoRAL: Breakthrough Open-Source Video HD Enhancement Tool

July 4

DeepSeek R1 Enhanced Version Boosts Efficiency by 200%

July 1

TEN VAD Open-Source Release: A Leap in Voice Detection Tech

June 10

NVIDIA and HKU Breakthrough: New AI Model Speeds Up High-Res Image Generation 84x

May 12

CAS Team Unveils MCA-Ctrl: A Game-Changer for AI Image Customization

AI DAMN

Latest Updates