Recent News
Recent News
All updates in reverse chronological order.
Event · 12 Paper · 24 Project · 11 Talk · 7 Service · 9 Intern · 3 Full Time · 1 Honor · 6
- 2026/10/19 Paper Our paper ORL: On-Policy Residual Learning for Diffusion Language Models is on arXiv and under review. Project, code, inference engine and models are now available.
- 2026/10/19 Project Together Post-Training is live! Flexible RL and SFT with custom training loops, multi-LoRA and full-weight training. Excited to share what our team is building.
- 2026/10/05
- 2026/10/05 Paper Linear RNN Scaling Laws: When Longer Sequences Beat More Sequences is available on arXiv.
- 2026/09/30
- 2026/09/30 Paper Our paper, When Does a Speech Model Know to Hand Off? Reading Competence from Hidden States for Real-Time Escalation, has been accepted at NeurIPS 2026 Workshop RTCA.
- 2026/09/28 Paper Orthros: Phase-Aware Heterogeneous Attention for Efficient Transformers has been accepted at NeurIPS 2026.
- 2026/09/24 Paper OSCAR: Offline Spectral Covariance-Aware Rotation for 2-bit KV Cache Quantization has been accepted at NeurIPS 2026.
- 2026/09/22 Service Serving as a reviewer for WACV 2027 (IEEE/CVF Winter Conference on Applications of Computer Vision).
- 2026/09/11 Paper Osprey: Target-agnostic Pre-training Makes Stronger Drafters in Speculative Decoding is accepted at EMNLP 2026. The paper and official code are now available. Osprey builds transferable drafters from pretrained small language models through target-agnostic pre-training and target-specific adaptation.
- 2026/09/03 Talk I gave an invited virtual talk at Pluralis Research on September 30, 2026: Toward AGI: Efficient Compression across the Data–PostTraining–Inference Loop. The talk covered compression-centered algorithm–architecture–system co-design across data, post-training, and inference.
- 2026/09/02
- 2026/09/01 Paper Our Massive Activations in Hybrid Linear Attention Large Language Models paper was featured in an in-depth interview by Synced (机器之心 Pro), reaching 3,839 reads. The coverage highlights our discovery of pre-attention spikes and inter-spike plateaus in hybrid linear-attention LLMs.
- 2026/08/28 Service Invited to serve as a reviewer for The Fifteenth International Conference on Learning Representations (ICLR 2027).
- 2026/08/18 Project We wrote 0 train-infer mismatch for Open-weight MoE RL: aligning XoRL and SGLang to bitwise-identical logprobs, which lifts Wordle solve rate from 63.9% to 77.4%.
- 2026/08/12 Paper Our new paper, Massive Activations in Hybrid Linear Attention Large Language Models: Pre-Attention Spikes and Inter-Spike Plateaus, is now on arXiv and under review. We present the first systematic study of massive activations in layer-interleaved hybrid linear-attention LLMs, identifying pre-attention spikes and inter-spike plateaus across architectures, hybridization configurations, data domains, and model scales.
- 2026/08/01
- 2026/07/22 Talk I gave an invited talk at Amazon: Compression-Centric Systems for Large-Model Post Training & Inference. Thank you to the Amazon team for the invitation and engaging discussion!
- 2026/07/09 Paper Our papers I-DLM: Introspective Diffusion Language Models and Squeeze Evolve: Unified Multi-Model Orchestration for Verifier-Free Evolution have been accepted by COLM 2026!
- 2026/06/26 Project Happy to see community interest in bringing OSCAR toward vLLM serving workflows. Grateful to the contributors pushing broader support for deployable rotation-based KV-cache quantization in mainstream LLM serving stacks.
- 2026/06/26 Paper Two new scaling-law papers are now on arXiv and under review: Sketched Linear Contrastive Learning: Approximation, Optimization, and Statistical Scaling, and From One-Pass SGD to Data Reuse: Mini-Batch Scaling Laws in Sketched Linear Regression. Congrats to Andy!
- 2026/06/19 Project OSCAR has been covered by multiple tech media outlets, including Towards AI, ModelScope, MarkTechPost, QbitAI (量子位), and Synced (机器之心), reaching 100,000+ reads across platforms. Grateful for the community interest in deployable 2-bit KV-cache quantization.
- 2026/06/18 Project Released Taylor-Calibrate: Principled Initialization for Hybrid Linear Attention Distillation. The codebase distills Qwen/Llama softmax-attention Transformers into hybrid linear-attention students built on GatedDeltaNet, using Taylor-series-informed initialization before staged distillation.
- 2026/06/17 Service Nominated as a reviewer for The 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026). Looking forward to contributing to the NLP and efficient language-model community.
- 2026/06/15 Paper OSCAR: Offline Spectral Covariance-Aware Rotation for 2-bit KV Cache Quantization is available on arXiv.
- 2026/06/10 Project Huge thanks to the open-source community — OSCAR has reached 500 stars on GitHub and now supports local llama.cpp usage, making 2-bit KV-cache quantization easier to try on local LLM deployments.
- 2026/05/29 Service Excited to help organize the 1st International Workshop on Sustainable and Efficient Language, Vision, and Action Models (SELVA 2026) at ACL 2026 — joining as a Program Committee member. SELVA brings together researchers working on efficient foundation models, multimodal systems, model compression, efficient inference, embodied AI, and related topics. Looking forward to reading the submissions!
- 2026/05/29 Talk I will be giving a talk at the AI-tonomy Summit (Models & Agents) on June 5 at the Plug and Play Tech Center in Silicon Valley — speaking in the AI Researcher Forum on Frontier Research: Scaling Agentic RL & Verified Reasoning. Looking forward to connecting with everyone building the next phase of agentic systems!
- 2026/05/26 Project Huge thanks to the open-source community — OSCAR has crossed 100 ⭐ on GitHub in its first week! Grateful to everyone who tried the rotation zoo, filed issues, or shared the project. The 2-bit KV-cache work is just getting started; more SGLang integration and additional reasoning-model recipes coming.
- 2026/05/19 Project Thrilled to release OSCAR: Offline Spectral Covariance-Aware Rotation for 2-bit KV Cache Quantization. INT2 KV-cache serving at 2.28 effective bits per element — ~8× KV-memory reduction and up to ~7× higher large-batch throughput with near-BF16 accuracy on Qwen3-4B/8B/32B and GLM-4.7-FP8.
- 2026/05/18
- 2026/04/03 Event I will be attending the International Conference on Learning Representations (ICLR 2026) in Rio. Looking forward to connecting with amazing researchers — see you in Rio!
- 2026/02/25
- 2026/02/25 Project Happy to share CoderForge-Preview: SOTA open dataset for training efficient coding agents.
- 2026/02/04
- 2026/01/30 Event I will be attending HPCA / PPoPP / CGO / CC 2026 as the web chair of HPCA 2026. Happy to see everyone there.
- 2026/01/27
- 2026/01/27
- 2025/11/13 Talk I will join a panel discussion on AI in the Cross-Border Digital Technologies Ecosystem: Infrastructure, Platforms & Global Launchpads.
- 2025/10/01 Service Joined 32nd IEEE International Symposium on High-Performance Computer Architecture (HPCA 2026) organization as Web Chair.
- 2025/09/02 Paper My first theoretical work on RL has been released on arXiv. Feel free to discuss!
- 2025/07/18 Full Time I am a Senior Research Scientist, Turbo & Kernel at Together AI, where I have worked since May 2024.
- 2025/07/01 Service Nominated as Reviewer for Thirty-ninth Conference on Neural Information Processing Systems (NeurIPS 2025).
- 2025/06/15 Honor I passed my Progress Evaluation Meeting as satisfactory or excellent — thank you USYD! Preparing for my thesis!
- 2025/05/02
- 2025/04/24 Event I will be attending the International Conference on Learning Representations (ICLR 2025) in Singapore. Looking forward to connecting with amazing researchers — see you in Singapore!
- 2025/03/16 Event I will be attending the NVIDIA GPU Technology Conference (GTC 2025) in San Jose, California. Excited to meet fellow GPU enthusiasts — see you there!
- 2024/12/07
- 2024/09/26
- 2024/07/10 Event I will be attending the USENIX Annual Technical Conference (ATC 2024) online. Looking forward to connecting!
- 2024/06/15 Honor I passed my Progress Evaluation Meeting as satisfactory or excellent — thank you USYD!
- 2024/06/06
- 2024/05/01 Intern Joined Together Computer as a Senior Research Scientist.
- 2024/03/01 Intern Joined Dolby as a Research Intern.
- 2023/11/22 Honor Received APR Intern Program Scholarship (SC3600). Thank you USYD and Dolby!
- 2023/10/06
- 2023/07/13 Service Joined ACM International Conference on Architectural Support for Programming Languages and Operating Systems Artifact Evaluation Committee (ASPLOS'24 AEC).
- 2023/06/20
- 2023/06/15 Honor I passed my Progress Evaluation Meeting as satisfactory or excellent — thank you Dolby and USYD!
- 2023/04/04
- 2023/04/03 Service Nominated as Reviewer for Thirty-seventh Conference on Neural Information Processing Systems (NeurIPS 2023).
- 2023/03/25 Event I will be attending the ACM International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS 2023) online. Excited to see you there!
- 2023/03/01 Intern Joined DeepSpeed Team, Microsoft as a Research Intern.
- 2023/03/01 Service Joined Computer Science Research Methods (CSRM 2023) — INFO5993 / INFO4990 at the University of Sydney — Program Committee.
- 2022/11/13 Event I will be attending the International Conference for High Performance Computing, Networking, Storage and Analysis (SC 2022) online. Looking forward to engaging with the global HPC community!
- 2022/10/01 Honor Received The Jingdong Technology (JD) Co Ltd Research Scholarship in Artificial Intelligence. Thank you JD and USYD!
- 2021/12/16 Event I will be attending the China National Computer Congress (CNCC 2021) online.
- 2021/11/14 Event I will be attending the International Conference for High Performance Computing, Networking, Storage and Analysis (SC 2021) online. See you online at SC — always an inspiring HPC gathering!
- 2021/07/01 Honor Received SYSU Overseas Visiting and Collaborative Research Program Funding Plan. Thank you SYSU!
- 2021/06/24 Event I will be attending the ACM International Conference on Supercomputing (ISC 2021) online. Looking forward to exchanging ideas with HPC experts worldwide!
- 2021/06/14 Event I will be attending the International Symposium on Computer Architecture (ISCA 2021) online.
- 2020/07/11 Talk I will give a talk on JSidentify at ICSE 2020. Looking forward to connecting!
- 2020/05/23 Event I will be attending the International Conference on High Performance Big Data and Intelligent Systems (HPBD&IS 2020) online.