Zhiyuan Liu

A new approach to make language models smaller and faster using 1-bit quantization.

ProSparse improves activation sparsity in LLMs for better efficiency and performance.

Examining grokking, double descent, and emergent abilities in deep learning models.

The Yi model family showcases strong language and multimodal processing capabilities.

New model improves image processing in multimodal systems.

EREN enhances the accuracy of language models through effective editing techniques.

New models enhance reasoning skills across various tasks, improving AI performance.

UltraMedical collections improve medical language models and address data shortages.

GUICourse aims to improve interaction with digital interfaces through targeted datasets for GUI agents.

This article discusses new approaches to improve predictions in chemical reactions using technology.

Examining how LLMs can add numbers without explicit steps.

Research aims to develop language models with unique personalities for better human-like interactions.

A new framework enhances evaluation of RAG systems in specialized domains.

MiniCPM-V enhances AI use on mobile devices with efficiency and performance.

A new approach to tokenization enhances analysis of ancient scripts.

Exploring the efficiency and adaptability of language models through modular design.

Exploring activation sparsity to improve language model efficiency.

KBAlign helps machines learn faster and more effectively through self-questioning techniques.

Discover how reward models are changing the way machines learn and perform.

Explore how large language models are becoming more efficient and accessible.

A new method combines autoregressive and diffusion models for better media generation.