🏢 Tsinghua University

Boosting code generation accuracy with more unit tests! This research shows that increasing the number of unit tests used to evaluate code generated by LLMs significantly improves accuracy, especially…

VisionReward, a novel reward model, surpasses existing methods by precisely capturing multi-dimensional human preferences for image and video generation, enabling more accurate and stable model optimi…

New benchmarks, HumanEval Pro and MBPP Pro, reveal LLMs struggle with self-invoking code generation, highlighting a critical gap in current code reasoning capabilities.

FoPE enhances attention’s periodic extension for better length generalization in language models by addressing spectral damage in RoPE using Fourier Series and zeroing out destructive frequencies.

Distilled Decoding (DD) drastically speeds up image generation from autoregressive models by using flow matching to enable one-step sampling, achieving significant speedups while maintaining acceptabl…

ReMoE: Revolutionizing Mixture-of-Experts with fully differentiable ReLU routing, achieving superior scalability and performance.

Token-level editing prevents language model collapse from synthetic data by theoretically bounding test error and empirically improving model performance.

LLaVA-UHD v2 enhances MLLMs by integrating high-resolution visual details using a hierarchical window transformer.

Self-play method SPAR enhances LLMs instruction following abilities, beating GPT-4 on IFEval

ColorFlow, a new AI model, accurately colorizes black-and-white image sequences while preserving character identity.

SynerGen-VL: A simpler, more powerful unified MLLM for image understanding and generation.

InternVL 2.5, a new open-source multimodal LLM, surpasses 70% on the MMMU benchmark, rivaling top commercial models through model, data, and test-time scaling strategies.

GENMAC: Multi-agent collaboration revolutionizes compositional text-to-video generation, achieving state-of-the-art results by iteratively refining videos via specialized agents.

LLMs’ training quality is exponentially improving, enabling models with half the parameters to match state-of-the-art performance every 3 months, thus reducing inference costs.

MIDI: a novel multi-instance diffusion model generates compositional 3D scenes from single images by simultaneously creating multiple 3D instances with accurate spatial relationships and high generali…

2DGS-Room: Seed-guided 2D Gaussian splatting with geometric constraints achieves state-of-the-art high-fidelity indoor scene reconstruction.

X-Prompt: a novel autoregressive vision-language model achieves universal in-context image generation by efficiently compressing contextual information and using a unified training framework for super…

Unified 3D latent representation (SLAT) enables versatile high-quality 3D asset generation, significantly outperforming existing methods.

Train high-performing Process Reward Models (PRMs) cheaply using only outcome-level labels, eliminating the need for costly step-by-step annotations!

AlphaTablets: A novel 3D plane representation enabling accurate, consistent, and flexible 3D planar reconstruction from monocular videos, achieving state-of-the-art results.