Explore · 每周精选
发现最受关注的研究论文,追踪研究趋势,订阅感兴趣的期刊与关键词。
Keivan Alizadeh, Iman Mirzadeh, Dmitry Belenko +5 authors
Efficient inference for large language models on devices with limited DRAM by optimizing data transfer and access from flash memory.
50 篇论文 · 按点赞排序
Albert Gu, Tri Dao
Mamba, a novel SSM-based model, outperforms Transformers in inference speed and scalability across various modalities by selectively propagating information and using efficient hardware-aware algorithms.
Yuxiang Wei, Zhe Wang, Jiawei Liu +2 authors
Magicoder, using OSS-Instruct to incorporate open-source code snippets, achieves superior performance on coding benchmarks while reducing bias in synthetic data generation.
Akio Kodaira, Chenfeng Xu, Toshiki Hazama +7 authors
StreamDiffusion, a real-time diffusion pipeline, enhances interactive image generation by implementing batching denoising, residual classifier-free guidance, and stochastic similarity filtering, which boost throughput and reduce energy consumption.
Dahyun Kim, Chanjun Park, Sanghoon Kim +15 authors
A novel technique, depth up-scaling (DUS), efficiently enhances large language models (LLMs) without complex changes, building SOLAR 10.7B that outperforms existing open-source models in various NLP tasks, including instruction-following.
Zhen Li, Mingdeng Cao, Xintao Wang +3 authors
PhotoMaker is an efficient personalized text-to-image generation method that encodes ID images into a unified embedding, achieving high ID fidelity, speed improvements, and generalization capabilities with an ID-oriented data pipeline.
Zhengzhong Liu, Aurick Qiao, Willie Neiswanger +25 authors
LLM360 initiative promotes full transparency and reproducibility in LLM training by open-sourcing training code, data, model checkpoints, and intermediate results.
Xueyao Zhang, Liumeng Xue, Yuancheng Wang +10 authors
Amphion is a toolkit for audio, music, and speech generation that includes model visualizations, vocoders, and evaluation metrics to support reproducible research and training for researchers.
Chi Zhang, Zhao Yang, Jiaxuan Liu +5 authors
A novel LLM-based multimodal agent learns to operate smartphone apps through autonomous exploration or imitation, demonstrating proficiency across diverse tasks.
Gemini Team, Rohan Anil, Sebastian Borgeaud +939 authors
Gemini, a family of multimodal models, achieves state-of-the-art performance across various benchmarks, including human-expert performance on MMLU, through advanced cross-modal reasoning and language understanding.
Zhaojian Yu, Xin Zhang, Ning Shang +5 authors
A Generator-Discriminator framework using LLMs generates high-quality instruction data, enhancing the fine-tuning of code-related tasks and model generalization.
Julian D. Parker, Janne Spijkervet, Katerina Kosta +6 authors
A transformer-based, non-autoregressive model generates musically coherent audio by responding to context, achieving audio quality comparable to text-conditioned models.
Dan Kondratyuk, Lijun Yu, Xiuye Gu +28 authors
VideoPoet is a language model that synthesizes high-quality video and audio from diverse inputs using a transformer architecture, showcasing superior zero-shot video generation capabilities.
Yixin Song, Zeyu Mi, Haotong Xie +1 authors
PowerInfer, a high-speed LLM inference engine for personal computers, enhances efficiency using hotspot neuron analysis, GPU-CPU hybrid computation, adaptive predictors, and neuron-aware sparse operators, achieving performance close to server-grade GPUs.
Vladimir Arkhipkin, Andrei Filatov, Viacheslav Vasilev +6 authors
Kandinsky 3.0, a large-scale text-to-image model based on latent diffusion, improves quality and realism through a larger architecture and advanced text understanding.
Renat Aksitov, Sobhan Miryoosefi, Zonglin Li +10 authors
A ReAct-style LLM agent with iterative training using reinforcement learning and self-distillation improves complex question-answering performance with fewer parameters.
Bingbin Liu, Sebastien Bubeck, Ronen Eldan +5 authors
A synthetic dataset of math problems paired with solutions enables small language models to achieve high accuracy in mathematical reasoning, outperforming larger models.
Róbert Csordás, Piotr Piękos, Kazuki Irie +1 authors
SwitchHead reduces memory and compute for Transformers by using Mixture-of-Experts layers, matching performance with significant speedup.
Luka Ribar, Ivan Chelombiev, Luke Hudlass-Galley +3 authors
SparQ Attention reduces memory bandwidth requirements in LLM attention blocks, enhancing inference throughput without accuracy loss.
Mengyang Feng, Jinlin Liu, Kai Yu +11 authors
DreaMoving is a diffusion-based framework that generates high-quality, customized human dance videos by controlling motion and preserving identity.
Hyeonho Jeong, Geon Yeong Park, Jong Chul Ye
The VMC framework enhances video generation by tuning temporal attention layers in diffusion models to accurately reproduce and diversify motion with a motion distillation objective.
Juan A. Rodriguez, Shubham Agarwal, Issam H. Laradji +4 authors
StarVector integrates CodeLLMs and vision models to generate complex SVGs by aligning visual and code tokens, enhancing visual quality and complexity handling.
Sondos Mahmoud Bsharat, Aidar Myrzakhan, Zhiqiang Shen
Guiding principles streamline querying and prompting of large language models, enhancing understanding and effectiveness across different scales.
Quan Sun, Yufeng Cui, Xiaosong Zhang +8 authors
A large-scale generative multimodal model with 37 billion parameters demonstrates strong few-shot in-context learning and achieves state-of-the-art performance on multimodal tasks through scaling-up and instruction tuning.
Peng Wang, Yichun Shi
ImageDream is a multi-view diffusion model for 3D object generation that uses image prompts to produce high-quality 3D models with improved visual geometry accuracy.
Chengshu Li, Jacky Liang, Andy Zeng +7 authors
Chain of Code (CoC) enhances language models' reasoning capabilities by using pseudocode and selective code emulation, enabling them to tackle complex semantic and logical tasks more effectively.
Zeyi Sun, Ye Fang, Tong Wu +6 authors
Alpha-CLIP enhances CLIP by adding an auxiliary alpha channel for attentive region suggestion, enabling precise control over image content across various tasks like open-world recognition and multimodal generation.
Zehua Chen, Guande He, Kaiwen Zheng +2 authors
Bridge-TTS introduces a novel text-to-speech system using a clean deterministic prior in diffusion models, significantly improving synthesis quality and sampling efficiency over existing methods.
Collin Burns, Pavel Izmailov, Jan Hendrik Kirchner +9 authors
Naive finetuning of strong pretrained models with weak supervision can improve performance but is insufficient to match full model capabilities, highlighting challenges for aligning superhuman models.
Tero Karras, Miika Aittala, Jaakko Lehtinen +3 authors
Modifications to network layers in the ADM diffusion model architecture improve training stability and synthesis quality, reducing FID from 2.41 to 1.81, and a novel method for post-hoc EMA parameter tuning is introduced.
北京市昌平区探索星信息技术及软件开发工作室
京ICP备2026059466号