Weili Xu

I am a sophomore in Computer Engineering, currently pursuing a dual degree from University of Illinois Urbana-Champaign and Zhejiang University.

I am fortunate to collaborate with Wenhao Chai and Enxin Song, working on Efficient Long Video Understanding. We built LongVidRWKV, a hybrid MLLM that efficiently handles hour-long videos on a single consumer GPU while achieving comparable performance to its Transformer counterparts on multiple video understanding benchmarks such as MLVU, MovieChat-1k and VDC.

I’m interested in various aspects of machine learning and computer systems:

Hardware-aware efficient algorithms
Exploiting sparsity for training and inference acceleration
Applications of multi-modal (video, audio, text, etc.) long-context modeling

news

Jun 25, 2025	One paper accepted by ICCV 2025, see you in Hawaii!
Mar 31, 2025	One paper accepted by the second CVPR workshop on Efficient Large Vision Models
Jan 21, 2025	Start to work as Teaching Assistant for ECE 220 Computer Systems & Programming with Prof. Ujjal Bhowmik

selected publications

ICCV 2025
Bringing RNNs Back to Efficient Open-Ended Video Understanding

Weili Xu, Enxin Song, Wenhao Chai, and 3 more authors

To appear at International Conference on Computer Vision (ICCV) 2025 , Oct 2025

Abs Bib PDF Supp

Transformer-based long video understanding models are hindered by their high computational complexity and prohibitive memory cost, since the memory and computation scale quadratically with input sequence length. We propose LongVidRWKV to address this challenge by replacing the LLM component in MLLMs with RWKV, an RNNl ike language model that handles input sequences of arbitrary length with constant-size hidden states. To reduce the gap between RWKV’s 4k context length and the extended token sequences typical of long videos, we combine visual token merge with linear RNN models and reorder merged visual tokens. Despite having only 2B parameters and being trained exclusively on public data, LongVidRWKV achieves performance comparable to Transformerbased models of similar size trained on private datasets across multiple video benchmarks. This demonstrates the potential of efficient, linear RNNs to lower the computation entry barrier for long video understanding. To our knowledge, we are the first to use an RWKV LLM backbone in a LLaVA-like model for open-ended video understanding.
@article{xu2024longvidrwkv, title = {Bringing RNNs Back to Efficient Open-Ended Video Understanding}, author = {Xu, Weili and Song, Enxin and Chai, Wenhao and Wen, Xuexiang and Ye, Tian and Wang, Gaoang}, publisher = {preprint}, month = oct, year = {2025}, }