Papers by Haoyi Wu
Layer-Condensed KV Cache for Efficient Inference of Large Language Models (2024.acl-long)
Copied to clipboard
| Challenge: | Using a key-value cache, memory consumption is a bottleneck for high-throughput language models. |
| Approach: | They propose a method that only computes and caches the KVs of a small number of layers, thus saving memory consumption and improving inference throughput. |
| Outcome: | The proposed method achieves higher throughput and competitive performance than standard transformers and is orthogonal to existing transformer memory-saving techniques. |
Why Vision Language Models Struggle with Visual Arithmetic? Towards Enhanced Chart and Geometry Understanding (2025.findings-acl)
Copied to clipboard
| Challenge: | Vision Language Models struggle with visual arithmetic, seemingly simple tasks like object counting or length comparison, which are essential for relevant complex tasks like chart understanding and geometric reasoning. |
| Approach: | They propose a novel post-training strategy inspired by Piaget’s theory of cognitive development that trains VLMs to recognize invariant properties under visual transformations. |
| Outcome: | The proposed approach outperforms supervised fine-tuning methods while requiring 60% less training data. |
Probabilistic Transformer: A Probabilistic Dependency Model for Contextual Word Representation (2023.findings-acl)
Copied to clipboard
| Challenge: | Syntactic structures were deemed essential in natural language processing . but since the deep learning revolution, NLP has been dominated by neural models that do not consider syntactical structures in their design. |
| Approach: | They propose a model that models latent representations of words in a sentence . they use a conditional random field to model latent and dependency arcs . |
| Outcome: | The proposed model performs competitively to transformers on small to medium sized datasets. |
Parallel Continuous Chain-of-Thought with Jacobi Iteration (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to continuous CoT rely on sequential decoding of latent thought tokens, which leads to long training time and low inference speed. |
| Approach: | They propose a parallel continuous chain-of-thought which updates latent thought tokens iteratively in parallel instead of sequentially and improves both training and inference efficiency. |
| Outcome: | The proposed method saves 50% of training and inference time while maintaining stability and robustness in training. |
Evaluating Cultural and Social Awareness of LLM Web Agents (2025.findings-naacl)
Copied to clipboard
Haoyi Qiu, Alexander Fabbri, Divyansh Agarwal, Kung-Hsiang Huang, Sarah Tan, Nanyun Peng, Chien-Sheng Wu
| Challenge: | Existing benchmarks often overlook cultural and social awareness . current evaluations focus on task completion, often ignoring the diverse cultural and socio-cultural backgrounds. |
| Approach: | They propose a benchmark to assess LLM agents’ sensitivity to cultural and social norms across two web-based tasks: online shopping and social discussion forums. |
| Outcome: | The proposed framework evaluates LLM agents’ ability to detect and appropriately respond to norm-violating user queries and observations across two web-based tasks. |
A Systematic Study of Cross-Layer KV Sharing for Efficient LLM Inference (2025.naacl-short)
Copied to clipboard
| Challenge: | Recent studies have shown that sharing key-value (KV) cache across layers is effective in efficient inference of large language models. |
| Approach: | They propose a unified framework that covers several recent methods and their novel variants to investigate cross-layer KV sharing. |
| Outcome: | The proposed framework achieves higher throughput and better performance when reducing the size of the key-value cache by 2 while maintaining competitive performance. |
Look Both Ways and No Sink: Converting LLMs into Text Encoders without Training (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods for converting large language models into powerful text encoders require extensive training on large datasets. |
| Approach: | They propose a training-free approach that enables bidirectional attention and suppresses the attention sink phenomenon, resulting in superior performance. |
| Outcome: | The proposed approach enables bidirectional attention and suppresses the attention sink phenomenon, resulting in superior performance. |
Conic10K: A Challenging Math Problem Understanding and Reasoning Dataset (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing benchmarks or datasets require only a few steps of reasoning, making it difficult to analyse AI’s behaviour with reference to different problems within a specific topic in detail. |
| Approach: | They propose a conic10K math problem dataset that requires only a few steps of reasoning to be analysed. |
| Outcome: | The proposed dataset shows that existing language models exhibit weak performance on complex reasoning. |