📝 Biography

🔥 Recent News

• Sep, 2024. From Instance Training to Instruction Learning: Task Adapters Generation from Instructions accepted to NeurIPS 2024!


• Oct, 2023. Dynamic Weighted Neural Bellman-Ford Network for Knowledge Graph Reasoning accepted to CCKS 2023!


• Jun, 2023. Graduated from NCEPU with honor: awarded Outstanding Graduate by Beijing Ministry of Education and by NCEPU!


• Dec, 2022. Awarded National Scholarship by Ministry of Education!


• Dec, 2021. Awarded National Scholarship by Ministry of Education!

📒 Publications  

Check my Google Scholar for more information!

(* stands for equal contribution. Listed in reverse chronological order.)

• SKIntern: Internalizing Symbolic Knowledge for Distilling Better CoT Capabilities into Small Language Models   
  Huanxuan Liao , Shizhu He, Yupu Hao, Xiang Li, Yuanzhe Zhang, Kang Liu, Jun Zhao   
  Preprint , 2024.
  
   

   

   

   

   

   

   

   

   

   
  @article{liao2024textit, title={$$\backslash$textit $\{$SKIntern$\}$ $: Internalizing Symbolic Knowledge for Distilling Better CoT Capabilities into Small Language Models}, author={Liao, Huanxuan and He, Shizhu and Hao, Yupu and Li, Xiang and Zhang, Yuanzhe and Liu, Kang and Zhao, Jun}, journal={arXiv preprint arXiv:2409.13183}, year={2024} } Small Language Models (SLMs) are attracting attention due to the high computational demands and privacy concerns of Large Language Models (LLMs). Some studies fine-tune SLMs using Chains of Thought (CoT) data distilled from LLMs, aiming to enhance their reasoning ability. Furthermore, Some CoT distillation methods introduce external symbolic knowledge into the generation process to improve the limited knowledge memory, reasoning ability and out-of-domain (OOD) generalization of SLMs. However, the introduction of symbolic knowledge increases computational overhead and introduces potential noise. In this paper, we introduce SKIntern, an innovative approach that empowers SLMs to internalize symbolic knowledge and few-shot examples gradually through a progressive fine-tuning process, guided by a predefined linear decay schedule under curriculum learning. By efficiently internalizing knowledge, SKIntern reduces computational overhead and speeds up the reasoning process by focusing solely on the question during inference. It outperforms state-of-the-art baselines by over 5\%, while reducing inference costs (measured in FLOPs) by up to 4× across a wide range of SLMs in both in-domain (ID) and out-of-domain (OOD) tasks.


• Neural-Symbolic Collaborative Distillation: Advancing Small Language Models for Complex Reasoning Tasks   
  Huanxuan Liao , Shizhu He, Yao Xu, Yuanzhe Zhang, Kang Liu, Jun Zhao   
  Preprint , 2024.
  
   

   

   

   

   

   

   

   

   

   
  @article{liao2024neural, title={Neural-Symbolic Collaborative Distillation: Advancing Small Language Models for Complex Reasoning Tasks}, author={Liao, Huanxuan and He, Shizhu and Xu, Yao and Zhang, Yuanzhe and Liu, Kang and Zhao, Jun}, journal={arXiv preprint arXiv:2409.13203}, year={2024} } In this paper, we propose Neural-Symbolic Collaborative Distillation (NesyCD), a novel knowledge distillation method for learning the complex reasoning abilities of Large Language Models (LLMs, e.g., \textgreater 13B). We argue that complex reasoning tasks are difficult for Small Language Models (SLMs, e.g., ≤ 7B), as these tasks demand not only general cognitive abilities but also specialized knowledge, which is often sparse and difficult for these neural-based SLMs to effectively capture. Therefore, NesyCD distills the general capabilities and specialized knowledge in LLMs using different manners. On the one hand, we distill only general abilities from teacher LLMs into the student SLMs of parameterized neural networks. On the other hand, for the specialized abilities and uncommon knowledge of a complex reasoning task, we employ a symbolic knowledge distillation approach to obtain and store the specialized knowledge within a symbolic knowledge base (KB). By decoupling general and specialized capabilities, the proposed NesyCD can achieve superior performance cost-effectively, utilizing smaller models and blending parameterized neural networks with symbolic KB. Moreover, the specialized KB generalizes well and is comprehended and manipulated by humans. Our experiments show that NesyCD significantly boosts SLMs' complex reasoning performance on in-domain (BBH, GSM8K) and out-of-domain (AGIEval, ARC) datasets. Notably, our approach enabled the LLaMA3-8B and Qwen2-7B to surpass GPT-3.5-turbo in performance and come close to matching LLaMA3-70B, despite the latter having nine times more parameters.


• CITI: Enhancing Tool Utilizing Ability in Large Language Models without Sacrificing General Performance   
  Yupu Hao, Pengfei Cao, Zhuoran Jin, Huanxuan Liao , Yubo Chen, Kang Liu, Jun Zhao   
  Preprint , 2024.
  
   

   

   

   

   

   

   

   

   

   
  @article{hao2024citi, title={CITI: Enhancing Tool Utilizing Ability in Large Language Models without Sacrificing General Performance}, author={Hao, Yupu and Cao, Pengfei and Jin, Zhuoran and Liao, Huanxuan and Liu, Kang and Zhao, Jun and others}, journal={arXiv preprint arXiv:2409.13202}, year={2024} } Tool learning enables the Large Language Models (LLMs) to interact with the external environment by invoking tools, enriching the accuracy and capability scope of LLMs. However, previous works predominantly focus on improving model's tool-utilizing accuracy and the ability to generalize to new, unseen tools, excessively forcing LLMs to adjust specific tool-invoking pattern without considering the harm to model's general performance. This deviates from the actual applications and original intention of integrating tools to enhance model. To tackle this problem, we dissect the capability trade-offs by examining the hidden representation changes and the gradient-based importance score of model's components. Based on the analysis result, we propose a Component Importance-based Tool-utilizing ability Injection method (CITI). According to the gradient-based importance score of different components, it alleviates the capability conflicts caused by fine-tuning process by applying distinct training strategies to different components. CITI applies Mixture-Of-LoRA (MOLoRA) for important components. Meanwhile, it fine-tunes the parameters of few components deemed less important in the backbone of the LLM, while keeping other parameters frozen. CITI can effectively enhance the model's tool-utilizing capability without excessively compromising its general performance. Experimental results demonstrate that our approach achieves outstanding performance across a range of evaluation metrics.


• [NeurIPS 2024] From Instance Training to Instruction Learning: Task Adapters Generation from Instructions
  Huanxuan Liao , Shizhu He, Yao Xu, Yuanzhe Zhang, Yanchao Hao, Shengping Liu, Kang Liu, Jun Zhao   
  In Advances in Neural Information Processing Systems 38.
  
   

   

   

   

   

   

   

   

   

   

  
  @article{liao2024instance, title={From Instance Training to Instruction Learning: Task Adapters Generation from Instructions}, author={Liao, Huanxuan and Xu, Yao and He, Shizhu and Zhang, Yuanzhe and Hao, Yanchao and Liu, Shengping and Liu, Kang and Zhao, Jun}, journal={arXiv preprint arXiv:2406.12382}, year={2024} } Large language models (LLMs) have acquired the ability to solve general tasks by utilizing instruction finetuning (IFT). However, IFT still relies heavily on instance training of extensive task data, which greatly limits the adaptability of LLMs to real-world scenarios where labeled task instances are scarce and broader task generalization becomes paramount. Contrary to LLMs, humans acquire skills and complete tasks not merely through repeated practice but also by understanding and following instructional guidelines. This paper is dedicated to simulating human learning to address the shortcomings of instance training, focusing on instruction learning to enhance cross-task generalization. Within this context, we introduce Task Adapters Generation from Instructions (TAGI), which automatically constructs the task-specific model in a parameter generation manner based on the given task instructions without retraining for unseen tasks. Specifically, we utilize knowledge distillation to enhance the consistency between TAGI developed through Learning with Instruction and task-specific models developed through Training with Instance, by aligning the labels, output logits, and adapter parameters between them. TAGI is endowed with cross-task generalization capabilities through a two-stage training process that includes hypernetwork pretraining and finetuning. We evaluate TAGI on the Super-Natural Instructions and P3 datasets. The experimental results demonstrate that TAGI can match or even outperform traditional meta-trained models and other hypernetwork models, while significantly reducing computational requirements.


• Imagination Augmented Generation: Learning to Imagine Richer Context for Question Answering over Large Language Models   
  Huanxuan Liao , Shizhu He, Yao Xu, Yuanzhe Zhang, Kang Liu, Shengping Liu, Jun Zhao   
  Preprint , 2024.
  
   

   

   

   

   

   

   

   

   

   

  
  @article{liao2024imagination, title={Imagination Augmented Generation: Learning to Imagine Richer Context for Question Answering over Large Language Models}, author={Liao, Huanxuan and He, Shizhu and Xu, Yao and Zhang, Yuanzhe and Liu, Kang and Liu, Shengping and Zhao, Jun}, journal={arXiv preprint arXiv:2403.15268}, year={2024} } Retrieval-Augmented-Generation and Gener-ation-Augmented-Generation have been proposed to enhance the knowledge required for question answering over Large Language Models (LLMs). However, the former relies on external resources, and both require incorporating explicit documents into the context, which increases execution costs and susceptibility to noise data. Recent works indicate that LLMs have modeled rich knowledge, albeit not effectively triggered or awakened. Inspired by this, we propose a novel knowledge-augmented framework, Imagination-Augmented-Generation (IAG), which simulates the human capacity to compensate for knowledge deficits while answering questions solely through imagination, thereby awakening relevant knowledge in LLMs without relying on external resources. Guided by IAG, we propose an imagine richer context method for question answering (IMcQA). IMcQA consists of two modules: explicit imagination, which generates a short dummy document by learning from long context compression, and implicit imagination, which creates flexible adapters by distilling from a teacher model with a long context. Experimental results on three datasets demonstrate that IMcQA exhibits significant advantages in both open-domain and closed-book settings, as well as in out-of-distribution generalization.

• LMTuner: An user-friendly and highly-integrable Training Framework for fine-tuning Large Language Models   
  Yixuan Weng, Zhiqi Wang, Huanxuan Liao , Shizhu He, Shengping Liu, Kang Liu, Jun Zhao   
  Preprint , 2023.
  
   

   

   

   

   

   

   

   

   

   

  
  @article{weng2023lmtuner, title={LMTuner: An user-friendly and highly-integrable Training Framework for fine-tuning Large Language Models}, author={Weng, Yixuan and Wang, Zhiqi and Liao, Huanxuan and He, Shizhu and Liu, Shengping and Liu, Kang and Zhao, Jun}, journal={arXiv preprint arXiv:2308.10252}, year={2023} } With the burgeoning development in the realm of large language models (LLMs), the demand for efficient incremental training tailored to specific industries and domains continues to increase. Currently, the predominantly employed frameworks lack modular design, it often takes a lot of coding work to kickstart the training of LLM. To address this, we present "LMTuner", a highly usable, integrable, and scalable system for training LLMs expeditiously and with minimal user-input. LMTuner comprises three main modules - the Interaction, Training, and Inference Modules. We advocate that LMTuner's usability and integrality alleviate the complexities in training large language models. Remarkably, even a novice user could commence training large language models within five minutes. Furthermore, it integrates DeepSpeed frameworks and supports Efficient Fine-Tuning methodologies like Low Rank Adaptation (LoRA), Quantized LoRA (QLoRA), etc., enabling the training of language models scaling from 300M to a whopping 130B parameters using a single server. The LMTuner's homepage (this https URL screencast video (this https URL) are now publicly available.


• [CCKS 2023] Dynamic Weighted Neural Bellman-Ford Network for Knowledge Graph Reasoning   
  Huanxuan Liao , Shizhu He, Yao Xu, Kang Liu, Jun Zhao   
  Accepted to the 8th China Conference on Knowledge Graph and Semantic Computing.
  
   

   

   

   

   
  @InProceedings{10.1007/978-981-99-7224-1_1, author="Liao, Huanxuan and He, Shizhu and Xu, Yao and Liu, Kang and Zhao, Jun", editor="Wang, Haofen and Han, Xianpei and Liu, Ming and Cheng, Gong and Liu, Yongbin and Zhang, Ningyu", title="Dynamic Weighted Neural Bellman-Ford Network for Knowledge Graph Reasoning", booktitle="Knowledge Graph and Semantic Computing: Knowledge Graph Empowers Artificial General Intelligence", year="2023", publisher="Springer Nature Singapore", address="Singapore", pages="3--16", isbn="978-981-99-7224-1" } Recent studies have shown that subgraphs of the head entity, such as related relations and neighborhoods, are helpful for Knowledge Graph Reasoning (KGR). However, prior studies tend to focus solely on enhancing entity representations using related relations, with little attention paid to the impact of different relations on different entities and their importance in various reasoning paths. Meanwhile, conventional Graph Neural Networks (GNNs) utilized for KGR consider simultaneously neighboring nodes and connected relations of the head entity but typically use a standard message-passing paradigm over the entire Knowledge Graph (KG). This results in over-smoothed representations and limits efficiency. To address the above-mentioned limitations of existing methods, we propose a Dynamic Weighted Neural Bellman-Ford Network (DyNBF) for KGR, which utilizes relation weights generated from subgraphs to compute only the most relevant relations and entities. This way, we can integrate multiple reasoning paths more flexibly to achieve better interpretable reasoning, while scaling more easily to more complex and larger KGs. DyNBF consists of two key modules: 1) a transformer-based relation weights generator module, which computes the weights of different relations on the path with a sequence-to-sequence model, and 2) an NBFNet-based logic reasoner module, which obtains entity representations and conducts fact prediction with dynamic weights from the previous module. Empirical results on three standard KGR datasets demonstrate that the proposed approach can generate explainable reasoning paths and obtain competitive performance.

🎖️ Awards

• Beijing Outstanding Graduate Awards (北京市优秀毕业生), Beijing Ministry of Education, 2023.
  Top graduate in Beijing.


• National Scholarship (国家奖学金), Ministry of Education, 2022.
  Top scholarship in China; 0.2% domestically.


• National Scholarship (国家奖学金), Ministry of Education, 2021.
  Top scholarship in China; 0.2% domestically.


• SiFang Society Scholarship (四方社会奖学金), North China Electric Power University, 2020.
  Top society scholarship in NCEPU.

💻 Professional Services

• Conference Reviewing : ACL ARR 2024

Copyright© 2024 Huanxuan Liao(廖桓萱).