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。
Anton Razzhigaev, Matvey Mikhalchuk, Elizaveta Goncharova +4 authors
Transformer decoders exhibit near-perfect linear relationships between layers, which can be reduced with cosine-similarity-based regularization, leading to improved performance on benchmarks.
50 篇论文 · 按点赞排序
Chameleon Team
Chameleon is a mixed-modal early-fusion token-based model that achieves state-of-the-art performance across various tasks, including image captioning, text generation, and long-form mixed-modal generation, using a unified architecture.
Seungone Kim, Juyoung Suk, Shayne Longpre +7 authors
Prometheus 2 is an advanced open-source evaluator LM that aligns closely with human and proprietary LM judgments, supporting both direct assessment and pairwise ranking with customizable criteria.
Justin Zhao, Timothy Wang, Wael Abid +7 authors
LoRA fine-tuning reduces computational resources while maintaining high performance across various tasks and models, and LoRAX allows for efficient deployment of multiple LLMs on a single GPU.
Wei Chen, Zhiyuan Li
The Octopus v4 model uses functional tokens to integrate and direct queries to task-specific open-source language models, achieving SOTA performance with models under 10B parameters.
Ziming Liu, Yixuan Wang, Sachin Vaidya +5 authors
Kolmogorov-Arnold Networks (KANs) outperform Multi-Layer Perceptrons (MLPs) in accuracy and interpretability by using learnable activation functions and spline-based weights.
Hugo Laurençon, Léo Tronchon, Matthieu Cord +1 authors
Idefics2, a vision-language model with 8 billion parameters, achieves state-of-the-art performance on multimodal benchmarks through extensive experimental validation.
Florian Bordes, Richard Yuanzhe Pang, Anurag Ajay +38 authors
Introduction to vision-language models (VLMs) covering their applications, training, evaluation, and extension to videos, addressing challenges in mapping visual data to language.
Dan Biderman, Jose Gonzalez Ortiz, Jacob Portes +9 authors
LoRA, a parameter-efficient finetuning method for large language models, underperforms full finetuning in target domains but provides better regularization and maintains diverse generation compared to other techniques.
Fabian Gloeckle, Badr Youbi Idrissi, Baptiste Rozière +2 authors
Training large language models to predict multiple future tokens simultaneously enhances their sample efficiency and performance, particularly on generative benchmarks and small algorithmic tasks, with decreased inference time.
Chanran Kim, Jeongin Lee, Shichang Joung +2 authors
InstantFamily uses masked cross-attention and multimodal embeddings to generate images with preserved multi-IDs and high visual coherence.
Hanze Dong, Wei Xiong, Bo Pang +7 authors
Online iterative reinforcement learning from human feedback achieves state-of-the-art performance in large language models using open-source datasets and proxy preference models.
Wenting Zhao, Xiang Ren, Jack Hessel +3 authors
Chatbots such as GPT-4 and ChatGPT are now serving millions of users. Despite their widespread use, there remains a lack of public datasets showcasing how these tools are used by a population of users in practice. To bridge this gap, we offered free access to ChatGPT for online users in exchange for their affirmative, consensual opt-in to anonymously collect their chat transcripts and request headers. From this, we compiled WildChat, a corpus of 1 million user-ChatGPT conversations, which consists of over 2.5 million interaction turns. We compare WildChat with other popular user-chatbot interaction datasets, and find that our dataset offers the most diverse user prompts, contains the largest number of languages, and presents the richest variety of potentially toxic use-cases for researchers to study. In addition to timestamped chat transcripts, we enrich the dataset with demographic data, including state, country, and hashed IP addresses, alongside request headers. This augmentation allows for more detailed analysis of user behaviors across different geographical regions and temporal dimensions. Finally, because it captures a broad range of use cases, we demonstrate the dataset's potential utility in fine-tuning instruction-following models. WildChat is released at https://wildchat.allen.ai under AI2 ImpACT Licenses.
Yupeng Zhou, Daquan Zhou, Ming-Ming Cheng +2 authors
StoryDiffusion, combining consistent self-attention and semantic motion prediction, enables generation of coherent and stable images and videos from textual descriptions.
Sean McLeish, Arpit Bansal, Alex Stein +8 authors
Transformers achieve state-of-the-art performance on large arithmetic tasks and other reasoning tasks by addressing positional tracking with embeddings and integrating architectural modifications.
Byung-Kwan Lee, Chae Won Kim, Beomchan Park +1 authors
Meteor, an efficient large language and vision model, enhances understanding and answering capabilities by embedding multifaceted rationales using the Mamba architecture, leading to improved vision language performance without increasing model size or using additional vision encoders.
Jihwan Kim, Junoh Kang, Jinyoung Choi +1 authors
FIFO-Diffusion generates long text-conditional videos through iterative denoising with latent partitioning and lookahead strategies to minimize the training-inference gap.
Ting Jiang, Shaohan Huang, Shengyue Luo +8 authors
MoRA, a high-rank updating method using square matrices, enhances the ability of large language models to learn and memorize new knowledge, especially in memory-intensive tasks, compared to LoRA.
Richard Yuanzhe Pang, Weizhe Yuan, Kyunghyun Cho +3 authors
An iterative preference optimization method using a modified DPO loss improves reasoning accuracy on various datasets by optimizing winning and losing reasoning steps in Chain-of-Thought candidates.
Ge Zhang, Scott Qu, Jiaheng Liu +42 authors
MAP-Neo, a fully open-sourced bilingual LLM with 7B parameters, achieves performance comparable to existing state-of-the-art LLMs, enhancing transparency and fostering innovation in the open research community.
Fu-Yun Wang, Zhaoyang Huang, Alexander William Bergman +9 authors
The Phased Consistency Model (PCM) addresses limitations in Latent Consistency Models (LCM) and outperforms them in high-resolution, text-conditioned image and few-step text-to-video generation.
Huajian Xin, Daya Guo, Zhihong Shao +6 authors
Generating extensive Lean 4 proof data from mathematical competition problems improved DeepSeekMath 7B's theorem-proving capabilities over GPT-4 and other methods.
Ruiqi Gao, Aleksander Holynski, Philipp Henzler +5 authors
CAT3D generates consistent 3D views from limited input using a multi-view diffusion model, enabling fast and accurate 3D reconstruction.
Chunjiang Ge, Sijie Cheng, Ziming Wang +6 authors
ConvLLaVA addresses excessive visual tokens and quadratic complexity in high-resolution multimodal models by using ConvNeXt as a hierarchical backbone, optimizing pretrained ConvNeXt for high resolution, and adding a successive stage for further compression, achieving competitive performance.
Boshi Wang, Xiang Yue, Yu Su +1 authors
Transformers can learn implicit reasoning through extensive training, with varying degrees of generalization for different reasoning types, and parametric memory proves more effective for complex reasoning tasks compared to non-parametric memory.
Jian Hu, Xibin Wu, Weixun Wang +3 authors
OpenRLHF is an open-source RLHF framework that enhances training efficiency and accessibility for researchers and practitioners.
Joshua Engels, Isaac Liao, Eric J. Michaud +2 authors
Research explores multi-dimensional features in language models, discovering interpretable circular representations in GPT-2, Mistral 7B, and Llama 3 8B, which are used for modular arithmetic tasks.
Abhijit Bendale, Michael Sapienza, Steven Ripplinger +3 authors
SUTRA, a multilingual Large Language Model architecture, achieves superior performance on multilingual tasks by decoupling conceptual understanding from language-specific processing using a Mixture of Experts framework.
Andreas Koukounas, Georgios Mastrapas, Michael Günther +11 authors
A novel multi-task contrastive training method improves CLIP model performance on both text-image and text-text retrieval tasks.
Mu Cai, Jianwei Yang, Jianfeng Gao +1 authors
Large Multimodal Models (LMMs) such as LLaVA have shown strong performance in visual-linguistic reasoning. These models first embed images into a fixed large number of visual tokens and then feed them into a Large Language Model (LLM). However, this design causes an excessive number of tokens for dense visual scenarios such as high-resolution images and videos, leading to great inefficiency. While token pruning/merging methods do exist, they produce a single length output for each image and do not afford flexibility in trading off information density v.s. efficiency. Inspired by the concept of Matryoshka Dolls, we propose M3: Matryoshka Multimodal Models, which learns to represent visual content as nested sets of visual tokens that capture information across multiple coarse-to-fine granularities. Our approach offers several unique benefits for LMMs: (1) One can explicitly control the visual granularity per test instance during inference, e.g. , adjusting the number of tokens used to represent an image based on the anticipated complexity or simplicity of the content; (2) M3 provides a framework for analyzing the granularity needed for existing datasets, where we find that COCO-style benchmarks only need around ~9 visual tokens to obtain accuracy similar to that of using all 576 tokens; (3) Our approach provides a foundation to explore the best trade-off between performance and visual token length at sample level, where our investigation reveals that a large gap exists between the oracle upper bound and current fixed-scale representations.
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