🔥 News
- 2026.03: 📄 Released the preprint RealChart2Code: Advancing Chart-to-Code Generation with Real Data and Multi-Task Evaluation.
- 2026.03: 🚀 Contributed as a major contributor (Speakn0w) to the release of Qwen3-Coder-Next, a highly efficient open-weight coding-agent model with 80B total parameters but only 3B activated parameters, delivering performance comparable to models with 10–20× more active parameters, strong long-horizon tool-use ability, and 256K native context for real IDE and CLI workflows.
- 2026.02: 📄 Released the preprint Scaling Agentic Verifier for Competitive Coding.
- 2026.02: 🚀 Contributed to the release of Qwen3.5, a native multimodal agent model family led by a 397B-A17B MoE release and later expanded to smaller sizes, with strong support for deep research, web dev, and adaptive tool use; my main contributions were in web development and agentic coding capabilities.
- 2026.01: 📄 Released the preprint MegaFlow: Large-Scale Distributed Orchestration System for the Agentic Era.
- 2026.01: 🎉 One paper accepted to WWW 2026: What Should I Cite? A RAG Benchmark for Academic Citation Prediction.
- 2026.01: 📄 Released two new preprints: Evaluating and Achieving Controllable Code Completion in Code LLM and From Completion to Editing: Unlocking Context-Aware Code Infilling via Search-and-Replace Instruction Tuning.
- 2025.11: 📄 Released the survey preprint From Code Foundation Models to Agents and Applications: A Comprehensive Survey and Practical Guide to Code Intelligence.
- 2025.11: 📄 Released the preprint PlotCraft: Pushing the Limits of LLMs for Complex and Interactive Data Visualization.
- 2025.08: 🎉 One paper accepted to EMNLP 2025 Findings: LAGCL4Rec: When LLMs Activate Interactions Potential in Graph Contrastive Learning for Recommendation.
- 2025.06: 🎉 One paper accepted to ICML 2025: SWE-Flow: Synthesizing Software Engineering Data in a Test-Driven Manner.
- 2025.05: 🚀 Contributed to the release of Qwen3-Coder, an open coding model with strong agentic coding performance, 256K native context (extendable to 1M with YaRN), and competitive results on repository-scale coding and browser-use tasks.
- 2025.05: 🎉 One paper accepted to NeurIPS 2025: Negative Feedback Really Matters: Signed Dual-Channel Graph Contrastive Learning Framework for Recommendation.
- 2024.12: 📄 Released the preprint ExecRepoBench: Multi-level Executable Code Completion Evaluation.
- 2024.09: 🚀 Contributed to the release of Qwen2.5-Coder, a strong open code model family trained on 5.5T tokens of code-centric data, achieving state-of-the-art performance across code generation, completion, reasoning, and repair.
- 2024.03: 🎉 One paper accepted to CIKM 2024: Evolving to the Future: Unseen Event Adaptive Fake News Detection on Social Media.