Yan Junchi presents JTok and xHC Transformer extensions at CNCC 2026
Overview
Shanghai Jiao Tong University professor Yan Junchi presented two Transformer architecture extensions at CNCC 2026, JTok and xHC. JTok adds learnable per-token parameter tables that modulate model updates through table lookup and element-wise operations. xHC expands the residual stream to 16 paths, using dense reads and sparse writes to raise model capacity.
In Yan's experiments, JTok-M reaches the same loss with about 35% less training compute but adds about 7% inference latency. xHC saves about 33% compared with a standard Transformer and about 16% compared with mHC, according to Leiphone's report of the talk.
Written by AI from the articles below · updated Nov 4, 5:12 AM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
- LeiphoneNewsShanghai Jiao Tong's Yan Junchi outlines JTok and xHC Transformer extensions at CNCC 2026
AIShanghai Jiao Tong University professor Yan Junchi presented two Transformer extensions at CNCC 2026: JTok, which adds learnable per-token parameter tables, and xHC, which expands the residual stream to 16 paths with dense reads and sparse writes. In his experiments, JTok-M cuts training compute by about 35% to reach the same loss but adds about 7% inference latency, while xHC saves about 33% versus a standard Transformer and about 16% versus mHC.
Heat Trend
Not enough continuous observations to show a trend yet.