Baize: An Open-Source Chat Model with Parameter-Efficient Tuning on Self-Chat Data
Proposes a pipeline that can automatically generate a high-quality multi-turn chat corpus by leveraging ChatGPT to engage in a conversation with itself
Can improve customer service chat bots and enable businesses to create their own chat models for customer interactions
Vision Transformers with Mixed-Resolution Tokenization
Introduces a novel image tokenization scheme for Vision Transformers, improving accuracy gains on image classification while controlling computational budget
Can improve image recognition tasks and enable businesses to process images more efficiently
Follow Your Pose: Pose-Guided Text-to-Video Generation using Pose-Free Videos
Presents a two-stage training scheme for obtaining pose-controllable character videos using pre-trained text-to-image models and easily obtained datasets
Can improve digital character creation for businesses in various industries, including gaming and entertainment
ReMoDiffuse: Retrieval-Augmented Motion Diffusion Model
ReMoDiffuse proposes a diffusion-model-based motion generation framework that integrates a retrieval mechanism to refine the denoising process for diverse motion generation
ReMoDiffuse enhances the generalizability and diversity of text-driven motion generation which is crucial for the creative industry, providing a better balance between text-motion consistency and motion quality
LLMMaps - A Visual Metaphor for Stratified Evaluation of Large Language Models
LLMMaps proposes a novel visualization technique that enables users to evaluate Large Language Models' (LLMs) performance with respect to question and answer (Q&A) datasets
LLMMaps provides detailed insights into LLMs' knowledge capabilities in different subfields, revealing subfields where hallucinations are more likely to occur and guiding their further development