Are there contractions simply too large to work on given memory ?

[0livier1O1]

I am trying to perform a contraction on a relatively large TN but I get a MemoryLimitExceeded when calling contract_path. From my understanding in this discussion, slicing should enable larger contractions but even after trying different settings for optimize_options I can't get the contraction to fit on memory. Are there contractions that simply need more GPU memory and cannot be sliced sufficiently small to fit?

This is the error

raise MemoryLimitExceeded(self.memory_limit, min_workspace_size, self.device_id) cuquantum.cutensornet.configuration.MemoryLimitExceeded: GPU memory limit exceeded. Device id: 0. The memory limit is 18971361280, while the minimum workspace size needed is 50448572928

and the runnable file to reproduce it can be found here: https://github.com/0livier1O1/tff/blob/move_to_gpu/tn_contraction.py
I am quite new to GPUs and TN in general.