| Jun 08, 2026 | \(A^3\) is now accepted in ICML 2026! It compresses LLMs memory, FLOPs and KV caches through principled low-rank decomposition. Amazing work from the team!  |
| Mar 14, 2026 | Team-of-Thought is accepted by MAIGAI@ICLR 2026 as an oral presentation! We introduced a novel framework that assembles diverse models to function as a Multi-Agent Team through tool-calling interfaces. |
| Mar 14, 2026 | Secondary Attention Sinks is accepted by UCLR@ICLR 2026! Beside traditional BOS attention sinks, we reveal that there exist a weaker type of attention sink that occurs in the middle layer, drawing weaker but still significant attention. |
| Oct 01, 2024 | I just started my PhD! New table assigned.  |