Proposal: Adding T1 (ICLR 2026) to the Library

[Oppenheimerdinger]

Hi, thanks for maintaining this great library — our codebase is built on TSlib and it was really helpful throughout the project.

I'd like to propose adding our model T1 for the imputation task.

T1 is a CNN-Transformer hybrid that uses Channel-Head Binding, where each CNN channel is directly paired with one attention head for selective cross-variable information transfer. It achieves
state-of-the-art on the standard imputation benchmarks (ETTh1/h2, ETTm1/m2, Weather, Electricity, Exchange, ILI).

• Paper (ICLR 2026): https://openreview.net/forum?id=IAnIlFsPEW | https://arxiv.org/abs/2602.21043
• Code: https://github.com/Oppenheimerdinger/T1

Our code follows the TSlib structure, so integration should be straightforward. Happy to submit a PR if you're interested.

Thanks!