本文前置知识: DDPM: Denoising Diffusion Probabilistic Model. DiffSinger: Singing Voice Synthesis via Shallow Diffusion Me

2025.03.17: 更新Reverse Process中的描述. DDPM: Denoising Diffusion Probabilistic ModelDDPM Overview DDPM: Denoising Diffusio

Introduction: Vector QuantizationVector QuantizationAutoEncoder(AE)由Encoder和Decoder组成, Encoder将图像压缩为一个低维的隐向量(Latent), 再由

本文前置知识: Vision & Language Pretrained Model 总结. 2025.05.06: 应评论区要求, 更新了Qwen-VL系列(Qwen-VL, Qwen2-VL, Qwen2.5-VL). Mu

本文前置知识: 通用信息抽取(上) - UIE, USM, InstructUIE. 通用信息抽取(下) - UniEX, Mirror, RexUIE本文为介绍通用信息抽取领域经典模型的下篇, 将会介绍了UniEX, Mirror

2024.5.27: 稍微补充了UIE的其中一个改进版MetaRetriever. 本文前置知识: T5: Exploring the Limits of Transfer Learning with a Unified Text-to

2024.4.21: 添加CoCa, 并修改对WPA的描述. 2024.4.23: 增加了BLIP-2的部分描述. Vision & Language Pretraining 总结本文只是以总结的形式梳理了近期比较有代表性的VL

大模型并行优化为什么要并行优化?大就是好, 虽然丛2019年人们的认识普遍就是大就是好, 这个概念在当今依然没有被改变, 只是有了更深刻的认识. 所以, 为什么要并行? 虽然大就是好, 模型太大显存吃不消(空间). 虽然大就是好

Query-based Instance Discrimination Network for Relational Triple Extraction本文是论文Query-based Instance Discrimination Net

UniRel: Unified Representation and Interaction for Joint Relational Triple Extraction本文是论文UniRel: Unified Representation

OneEE: A One-Stage Framework for Fast Overlapping and Nested Event Extraction本文是论文OneEE: A One-Stage Framework for Fast

W2NER: Unified Named Entity Recognition as Word - Word Relation Classification本文是论文Unified Named Entity Recognition as W