Deliberação no Espaço Latente via Augmentação de Cache Diferenciável
Deliberation in Latent Space via Differentiable Cache Augmentation
December 23, 2024
Autores: Luyang Liu, Jonas Pfeiffer, Jiaxing Wu, Jun Xie, Arthur Szlam
cs.AI
Resumo
Técnicas que permitem que grandes modelos de linguagem (LLMs) "pensem mais" gerando e atendendo a etapas intermediárias de raciocínio têm mostrado promessa na resolução de problemas complexos. No entanto, as abordagens padrão geram sequências de tokens discretos imediatamente antes de responder, o que pode acarretar custos significativos de latência e ser desafiador de otimizar. Neste trabalho, demonstramos que um LLM congelado pode ser aumentado com um coprocessador offline que opera no cache chave-valor do modelo. Esse coprocessador aumenta o cache com um conjunto de embeddings latentes projetados para melhorar a fidelidade da decodificação subsequente. Treinamos esse coprocessador usando a perda de modelagem de linguagem do decodificador em dados padrão de pré-treinamento, mantendo o próprio decodificador congelado. Essa abordagem permite que o modelo aprenda, de forma diferenciável de ponta a ponta, como destilar computação adicional em seu cache chave-valor. Como o decodificador permanece inalterado, o coprocessador pode operar offline e de forma assíncrona, e o modelo de linguagem pode funcionar normalmente se o coprocessador não estiver disponível ou se um determinado cache não precisar de computação extra. Mostramos experimentalmente que, quando um cache é aumentado, o decodificador alcança menor perplexidade em numerosos tokens subsequentes. Além disso, mesmo sem nenhum treinamento específico da tarefa, nossos experimentos demonstram que a ampliação do cache consistentemente reduz a perplexidade e melhora o desempenho em uma variedade de tarefas intensivas em raciocínio.
English
Techniques enabling large language models (LLMs) to "think more" by
generating and attending to intermediate reasoning steps have shown promise in
solving complex problems. However, the standard approaches generate sequences
of discrete tokens immediately before responding, and so they can incur
significant latency costs and be challenging to optimize. In this work, we
demonstrate that a frozen LLM can be augmented with an offline coprocessor that
operates on the model's key-value (kv) cache. This coprocessor augments the
cache with a set of latent embeddings designed to improve the fidelity of
subsequent decoding. We train this coprocessor using the language modeling loss
from the decoder on standard pretraining data, while keeping the decoder itself
frozen. This approach enables the model to learn, in an end-to-end
differentiable fashion, how to distill additional computation into its
kv-cache. Because the decoder remains unchanged, the coprocessor can operate
offline and asynchronously, and the language model can function normally if the
coprocessor is unavailable or if a given cache is deemed not to require extra
computation. We show experimentally that when a cache is augmented, the decoder
achieves lower perplexity on numerous subsequent tokens. Furthermore, even
without any task-specific training, our experiments demonstrate that cache
augmentation consistently reduces perplexity and improves performance across a
range of reasoning-intensive tasks.Summary
AI-Generated Summary
DeepSeek-R1: Incentivizando a Capacidade de Raciocínio em LLMs via
Aprendizado por ReforçoDeepSeek-R1: Incentivizing Reasoning Capability in LLMs via
Reinforcement Learning
DeepSeek-R1: Incentivizando a Capacidade de Raciocínio em LLMs via
Aprendizado por Reforço
DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via
Reinforcement Learning
DeepSeek-AI, Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, Xiaokang Zhang, Xingkai Yu, Yu Wu, Z. F. Wu, Zhibin Gou, Zhihong Shao, Zhuoshu Li, Ziyi Gao, Aixin Liu, Bing Xue, Bingxuan Wang, Bochao Wu, Bei Feng, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, Damai Dai, Deli Chen, Dongjie Ji, Erhang Li, Fangyun Lin, Fucong Dai, Fuli Luo, Guangbo Hao, Guanting Chen, Guowei Li, H. Zhang, Han Bao, Hanwei Xu, Haocheng Wang, Honghui Ding, Huajian Xin, Huazuo Gao, Hui Qu, Hui Li, Jianzhong Guo, Jiashi Li, Jiawei Wang, Jingchang Chen, Jingyang Yuan, Junjie Qiu, Junlong Li, J. L. Cai, Jiaqi Ni, Jian Liang, Jin Chen, Kai Dong, Kai Hu, Kaige Gao, Kang Guan, Kexin Huang, Kuai Yu, Lean Wang, Lecong Zhang, Liang Zhao, Litong Wang, Liyue Zhang, Lei Xu, Leyi Xia, Mingchuan Zhang, Minghua Zhang, Minghui Tang, Meng Li, Miaojun Wang, Mingming Li, Ning Tian, Panpan Huang, Peng Zhang, Qiancheng Wang, Qinyu Chen, Qiushi Du, Ruiqi Ge, Ruisong Zhang, Ruizhe Pan, Runji Wang, R. J. Chen, R. L. Jin, Ruyi Chen, Shanghao Lu, Shangyan Zhou, Shanhuang Chen, Shengfeng Ye, Shiyu Wang, Shuiping Yu, Shunfeng Zhou, Shuting Pan, S. S. Li, Shuang Zhou, Shaoqing Wu, Shengfeng Ye, Tao Yun, Tian Pei, Tianyu Sun, T. Wang, Wangding Zeng, Wanjia Zhao, Wen Liu, Wenfeng Liang, Wenjun Gao, Wenqin Yu, Wentao Zhang, W. L. Xiao, Wei An, Xiaodong Liu, Xiaohan Wang, Xiaokang Chen, Xiaotao Nie, Xin Cheng, Xin Liu, Xin Xie, Xingchao Liu, Xinyu Yang, Xinyuan Li, Xuecheng Su, Xuheng Lin, X. Q. Li, Xiangyue Jin, Xiaojin Shen, Xiaosha Chen, Xiaowen Sun, Xiaoxiang Wang, Xinnan Song, Xinyi Zhou, Xianzu Wang, Xinxia Shan, Y. K. Li, Y. Q. Wang, Y. X. Wei, Yang Zhang, Yanhong Xu, Yao Li, Yao Zhao, Yaofeng Sun, Yaohui Wang, Yi Yu, Yichao Zhang, Yifan Shi, Yiliang Xiong, Ying He, Yishi Piao, Yisong Wang, Yixuan Tan, Yiyang Ma, Yiyuan Liu, Yongqiang Guo, Yuan Ou, Yuduan Wang, Yue Gong, Yuheng Zou, Yujia He, Yunfan Xiong, Yuxiang Luo, Yuxiang You, Yuxuan Liu, Yuyang Zhou, Y. X. Zhu, Yanhong Xu, Yanping Huang, Yaohui Li, Yi Zheng, Yuchen Zhu, Yunxian Ma, Ying Tang, Yukun Zha, Yuting Yan, Z. Z. Ren, Zehui Ren, Zhangli Sha, Zhe Fu, Zhean Xu, Zhenda Xie, Zhengyan Zhang, Zhewen Hao, Zhicheng Ma, Zhigang Yan, Zhiyu Wu, Zihui Gu, Zijia Zhu, Zijun Liu, Zilin Li, Ziwei Xie, Ziyang Song, Zizheng Pan, Zhen Huang, Zhipeng Xu, Zhongyu Zhang, Zhen Zhang•Jan 22, 2025•3685
Relatório Técnico do Qwen2.5Qwen2.5 Technical Report
Relatório Técnico do Qwen2.5
Qwen2.5 Technical Report
Qwen, An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, Huan Lin, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jingren Zhou, Junyang Lin, Kai Dang, Keming Lu, Keqin Bao, Kexin Yang, Le Yu, Mei Li, Mingfeng Xue, Pei Zhang, Qin Zhu, Rui Men, Runji Lin, Tianhao Li, Tingyu Xia, Xingzhang Ren, Xuancheng Ren, Yang Fan, Yang Su, Yichang Zhang, Yu Wan, Yuqiong Liu, Zeyu Cui, Zhenru Zhang, Zihan Qiu•Dec 19, 2024•36311
MiniMax-01: Dimensionamento de Modelos de Fundação com Atenção RelâmpagoMiniMax-01: Scaling Foundation Models with Lightning Attention
MiniMax-01: Dimensionamento de Modelos de Fundação com Atenção Relâmpago
MiniMax-01: Scaling Foundation Models with Lightning Attention
MiniMax, Aonian Li, Bangwei Gong, Bo Yang, Boji Shan, Chang Liu, Cheng Zhu, Chunhao Zhang, Congchao Guo, Da Chen, Dong Li, Enwei Jiao, Gengxin Li, Guojun Zhang, Haohai Sun, Houze Dong, Jiadai Zhu, Jiaqi Zhuang, Jiayuan Song, Jin Zhu, Jingtao Han, Jingyang Li, Junbin Xie, Junhao Xu, Junjie Yan, Kaishun Zhang, Kecheng Xiao, Kexi Kang, Le Han, Leyang Wang, Lianfei Yu, Liheng Feng, Lin Zheng, Linbo Chai, Long Xing, Meizhi Ju, Mingyuan Chi, Mozhi Zhang, Peikai Huang, Pengcheng Niu, Pengfei Li, Pengyu Zhao, Qi Yang, Qidi Xu, Qiexiang Wang, Qin Wang, Qiuhui Li, Ruitao Leng, Shengmin Shi, Shuqi Yu, Sichen Li, Songquan Zhu, Tao Huang, Tianrun Liang, Weigao Sun, Weixuan Sun, Weiyu Cheng, Wenkai Li, Xiangjun Song, Xiao Su, Xiaodong Han, Xinjie Zhang, Xinzhu Hou, Xu Min, Xun Zou, Xuyang Shen, Yan Gong, Yingjie Zhu, Yipeng Zhou, Yiran Zhong, Yongyi Hu, Yuanxiang Fan, Yue Yu, Yufeng Yang, Yuhao Li, Yunan Huang, Yunji Li, Yunpeng Huang, Yunzhi Xu, Yuxin Mao, Zehan Li, Zekang Li, Zewei Tao, Zewen Ying, Zhaoyang Cong, Zhen Qin, Zhenhua Fan, Zhihang Yu, Zhuo Jiang, Zijia Wu•Jan 14, 2025•2836