Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileAcademic lineage
Possible advisorsa guess from early papers, not confirmed
- Xiaodan ZhuSuggested from co-authorship
Is this you? Claim this profile to confirm or dismiss it.
Works83 from public data
- Qwen2.5 Technical Report5,408
The open-weight flagship Qwen2.5-72B-Instruct outperforms a number of open and proprietary models and demonstrates competitive performance to the state-of-the-art open-weight model, Llama-3-405B-Instruct, which is around 5 times larger.
- Qwen Technical Report4,443
Qwen is a comprehensive language model series that encompasses distinct models with varying parameter counts, and includes Qwen, the base pretrained language models, and Qwen-Chat, the chat models finetuned with human alignment techniques.
- Qwen3-VL Technical Report2,675
Qwen3-VL is introduced, the most capable vision-language model in the Qwen series to date, achieving superior performance across a broad range of multimodal benchmarks, and three key upgrades are introduced, including an enhanced interleaved-MRoPE for stronger spatial-temporal modeling across images and video.
- Qwen2 Technical Report2,548
A comprehensive suite of foundational and instruction-tuned language models, encompassing a parameter range from 0.5 to 72 billion, featuring dense models and a Mixture-of-Experts model, which surpasses most prior open-weight models and demonstrates robust multilingual capabilities.
- Qwen2.5-Math Technical Report: Toward Mathematical Expert Model via Self-Improvement1,174
A series of math-specific large language models that possess advanced mathematical reasoning capabilities, including Chain-of-Thought (CoT) and Tool-Integrated Reasoning (TIR), and are evaluated on 10 mathematics datasets in both English and Chinese.
- BigCodeBench: Benchmarking Code Generation with Diverse Function Calls and Complex Instructions724
An extensive evaluation of 60 LLMs shows that LLMs are not yet capable of following complex instructions to use function calls precisely, with scores up to 60%, significantly lower than the human performance of 97%, which underscores the need for further advancements in this area.
- OLMoE: Open Mixture-of-Experts Language Models352
This work introduces OLMoE, a fully open, state-of-the-art language model leveraging sparse Mixture-of-Experts (MoE), and presents various experiments on MoE training, analyzes routing in the model showing high specialization, and open-source all aspects of the work: model weights, training data, code, and logs.
- Large Language Models are Versatile Decomposers: Decomposing Evidence and Questions for Table-based Reasoning288
This work exploits large language models (LLMs) as decomposers for effective table-based reasoning, and proposes a novel ''parsing-execution-filling'' strategy to decompose a complex question into simper step-by-step sub-questions by generating intermediate SQL queries as a bridge.
- 173
- 135
- CodeElo: Benchmarking Competition-level Code Generation of LLMs with Human-comparable Elo Ratings120
This work introduces CodeElo, a standardized competition-level code generation benchmark that effectively addresses all these challenges for the first time and provides the Elo ratings of 30 existing popular open-source and 3 proprietary LLMs for the first time.
- 111
- A Survey on Text-to-SQL Parsing: Concepts, Methods, and Future Directions110
A comprehensive review on deep learning approaches for text-to-SQL parsing, including a systematical overview of pre-trained language models and existing methods, is presented.
- Lemur: Harmonizing Natural Language and Code for Language Agents98
Lemur and Lemur-Chat are introduced, openly accessible language models optimized for both natural language and coding capabilities to serve as the backbone of versatile language agents, demonstrating balanced proficiencies in both domains, unlike existing open-source models that tend to specialize in either.
- 95
- Qwen3-Coder-Next Technical Report83
This work explores how far strong training recipes can push the capability limits of models with small parameter footprints and presents Qwen3-Coder-Next, an open-weight language model specialized for coding agents that activates only 3 billion parameters during inference, enabling strong coding capability with efficient inference.
- 76
- Collaborative Graph Convolutional Networks: Unsupervised Learning Meets Semi-Supervised Learning76
This paper proposes a multi-task graph learning model, called collaborative graph convolutional networks (CGCN), composed of an attributed graph clustering network and a semi-supervised node classification network, designed by combining variational graph auto-encoder with Gaussian mixture models (GMM-VGAE).
- PolyLM: An Open Source Polyglot Large Language Model75
PolyLM, a multilingual LLM trained on 640 billion (B) tokens, avaliable in two model sizes: 1.7B and 13B, surpasses other open-source models such as LLaMA and BLOOM on multilingual tasks while maintaining comparable performance in English.
- Latent Heterogeneous Graph Network for Incomplete Multi-View Learning69
A novel Latent Heterogeneous Graph Network (LHGN) for incomplete multi-view learning is proposed, which aims to use multiple incomplete views as fully as possible in a flexible manner and explores the complex relationship between samples and latent representations.
- 63
- Proton: Probing Schema Linking Information from Pre-trained Language Models for Text-to-SQL Parsing56
A novel framework to elicit relational structures from large-scale pre-trained language models (PLMs) via a probing procedure based on Poincaré distance metric, and use the induced relations to augment current graph-based parsers for better schema linking is proposed.
- Large Language Models Meet Symbolic Provers for Logical Reasoning Evaluation52
It is shown that state-of-the-art LLMs struggle to solve ProverQA problems, even with CoT prompting, highlighting the dataset's challenging nature, and the value of the proposed data generation framework.
- Dynamic Hybrid Relation Exploration Network for Cross-Domain Context-Dependent Semantic Parsing52
This paper presents a dynamic graph framework that is capable of effectively modelling contextual utterances, tokens, database schemas, and their complicated interaction as the conversation proceeds, and employs a dynamic memory decay mechanism that incorporates inductive bias to integrate enriched contextual relation representation.
- Prompting Large Language Models to Tackle the Full Software Development Lifecycle: A Case Study45
DevBench is proposed, a comprehensive benchmark that evaluates LLMs across various stages of the software development lifecycle, including software design, environment setup, implementation, acceptance testing, and unit testing.
- 44
- 44
- A Preliminary Study of the Intrinsic Relationship between Complexity and Alignment40
This paper proposes Tree-Instruct to systematically enhance the instruction complexity in a controllable manner by adding a specified number of nodes to instructions’ semantic trees, which not only yields new instruction data from the modified tree but also allows us to control the difficulty level of modified instructions.
- DEEM: Diffusion Models Serve as the Eyes of Large Language Models for Image Perception38
Compared to the state-of-the-art interleaved content generation models, DEEM exhibits enhanced robustness and a superior capacity to alleviate model hallucinations while utilizing fewer trainable parameters, less pre-training data, and a smaller base model size.
- Geometry-Guided Dense Perspective Network for Speech-Driven Facial Animation38
A non-linear face reconstruction representation is introduced as a guidance of latent space to obtain more accurate deformation, which helps solve the geometry-related deformation and is good for generalization across subjects.
- 37
- ExecRepoBench: Multi-level Executable Code Completion Evaluation29
A novel framework for enhancing code completion in software development through the creation of a repository-level benchmark ExecRepoBench and the instruction corpora Repo-Instruct, aim at improving the functionality of open-source large language models (LLMs) in real-world coding scenarios that involve complex interdependencies across multiple files.
- Evaluating and Aligning CodeLLMs on Human Preference29
A rigorous human-curated benchmark CodeArena is presented to emulate the complexity and diversity of real-world coding tasks and reveals a notable performance gap between open SOTA code LLMs and proprietary LLMs, underscoring the importance of the human preference alignment.
- 28
- SWE-RM: Execution-free Feedback For Software Engineering Agents27
SWE-RM is introduced, an accurate and robust reward model adopting a mixture-of-experts architecture with 30B total parameters and 3B activated during inference that substantially improves SWE agents on both TTS and RL performance.
- 27
- 24
- BIRD-INTERACT: Re-imagining Text-to-SQL Evaluation for Large Language Models via Lens of Dynamic Interactions22
BIRD-INTERACT is introduced, a benchmark that restores realism to multi-turn interactions in large language models by coupling each database with a hierarchical knowledge base, metadata files, and a function-driven user simulator, enabling models to solicit clarifications, retrieve knowledge, and recover from errors without human supervision.
- SWE-Flow: Synthesizing Software Engineering Data in a Test-Driven Manner20
This work introduces **SWE-Flow**, a novel data synthesis framework grounded in Test-Driven Development (TDD) that automatically infers incremental development steps directly from unit tests, enabling the generation of a structured, step-by-step *development schedule*.
- VideoAgentTrek: Computer Use Pretraining from Unlabeled Videos20
VideoAgentTrek is presented, a scalable pipeline that automatically mines training data from publicly available screen-recorded videos at web scale, eliminating the need for manual annotation, providing a scalable alternative to expensive manual annotation.
- 18
- Model Inversion Attack via Dynamic Memory Learning16
This work proposes a novel Dynamic Memory Model Inversion Attack (DMMIA) to leverage historically learned knowledge, which interacts with samples (during the training) to induce diverse generations and performs better than state-of-the-art MI attack methods.
- 15
- 15
- SWE-Universe: Scale Real-World Verifiable Environments to Millions14
SWE-Universe, a scalable and efficient framework for automatically constructing real-world software engineering (SWE) verifiable environments from GitHub pull requests (PRs), utilizes a building agent powered by an efficient custom-trained model to ensure the reliable generation of high-fidelity, verifiable tasks.
- TrojanSQL: SQL Injection against Natural Language Interface to Database14
TrojanSQL is proposed, a backdoor-based SQL injection framework for text-to-SQL systems, to show how state-of-the-art text-to-SQL parsers can be easily misled to produce harmful SQL statements that can invalidate user queries or compromise sensitive information about the database.
- Multi-Agent Collaboration for Multilingual Code Instruction Tuning12
This work introduces a novel multi-agent collaboration framework to enhance multilingual instruction tuning for code LLMs, where multiple language-specific intelligent agent components with generation memory work together to transfer knowledge from one language to another efficiently and effectively.
- 12
- 11
- Transaudio: Towards the Transferable Adversarial Audio Attack Via Learning Contextualized Perturbations11
A novel contextualized attack with deletion, insertion, and substitution adversarial behaviors, namely TransAudio, which achieves arbitrary word-level attacks based on the proposed two-stage framework and strengthens the attack transferability.
- SUN: Exploring Intrinsic Uncertainties in Text-to-SQL Parsers11
A model uncertainty constraint is proposed to refine the query representations by enforcing the output representations of different perturbed encoding networks to be consistent with each other and improve the generalizability and stability of neural text-to-SQL parsers.
- MegaFlow: Large-Scale Distributed Orchestration System for the Agentic Era10
Large-scale distributed orchestration system that enables efficient scheduling, resource allocation, and fine-grained task management for agent-environment workloads and addresses a critical infrastructure gap in the emerging agentic AI landscape.
- 10
- 8
- WorldPM: Scaling Human Preference Modeling7
Through evaluations on 7 benchmarks with 20 subtasks, it is found that WorldPM broadly improves the generalization performance across human preference datasets of varying sizes, with performance gains exceeding 5% on many key subtasks.
- 7
- Scaling Agentic Verifier for Competitive Coding6
Agentic Verifier is proposed, an execution-based agent that actively reasons about program behaviors and searches for highly discriminative test inputs that expose behavioral discrepancies among candidate solutions to solve competitive programming problems correctly in a single attempt.
- 6
- 6
- 5
- BigCodeArena: Unveiling More Reliable Human Preferences in Code Generation via Execution5
The evaluation shows that most LLMs have superior performance in judging coding preferences when the execution results are available, and proposes AutoCodeArena, an automatic Elo rating benchmark designed to assess the coding quality of LLMs without human involvement.
- 5
- 5
- DialCLIP: Empowering Clip As Multi-Modal Dialog Retriever4
This paper introduces a multi-modal context prompt generator to learn context features which are subsequently distilled into prompts within the pre-trained vision-language model CLIP, and introduces domain prompt to mitigate the disc repancy from the downstream dialog data.
- 4
- 4
- Teaching LLM to Reason: Reinforcement Learning from Algorithmic Problems without Code3
This work proposes TeaR, which aims at teaching LLMs to reason better by leveraging careful data curation and reinforcement learning to guide models in discovering optimal reasoning paths through code-related tasks, thereby improving general reasoning abilities.
- 3
- 2
- 2
- Evaluating and Achieving Controllable Code Completion in Code LLM1
This work presents the first instruction-guided code completion benchmark, Controllable Code Completion Benchmark (C3-Bench), comprising 2,195 carefully designed completion tasks and develops a straightforward data synthesis pipeline that leverages Qwen2.5-Coder to generate high-quality instruction-completion pairs for supervised fine-tuning (SFT).
- Fine-Tuning Language Models with Collaborative and Semantic Experts1
A Collaborative and Semantic Experts (CoE) approach for supervised fine-tuning (SFT), which employs a two-phase training strategy, demonstrating improved performance and expert collaboration in diverse tasks, significantly outperforming traditional SFT methods.
- 1
- VDialogUE: A Unified Evaluation Benchmark for Visually-grounded Dialogue1
The VDialogUE benchmark, along with the evaluation scripts and baseline models, are proposed, which are believed to accelerate the development of visually-grounded dialog systems and lead to theDevelopment of more sophisticated and effective pre-trained models.
- 1
- 1
- –
- –
- –
- –
- –
- –
- –
Publication data from OpenAlex, with missing venues and authors filled in from Crossref; citation counts are the higher of OpenAlex and Semantic Scholar, last synced 2026-10-14. One-sentence summaries under some papers are written by Semantic Scholar’s model. Citation counts may be lower than on Google Scholar, which indexes more sources.
Report an error
Wrong papers, two people merged into one, or a profile that should not be here? Tell us and we will fix or hide it.