Hi,
The post you link does not implement a new concatenation op. It just have a helper function that caries the pair of tensors around and do the ops on each. You can already do that by just using the list with your two Tensors and then adapt the next function to do on it. Did I missed something in that issue?
Another solution is to pre-allocate the full tensor and compute t1 and t2 directly into it doing inplace operations. That way you don’t need the cat operation at all.