Add ECHOBenchmark dataset

[matteotolloso]

Summary

This PR adds the ECHOBenchmark dataset, introduced in:

"Can You Hear Me Now? A Benchmark for Long-Range Graph Propagation"
ICLR 2026
https://openreview.net/forum?id=DgkWFPZMPp

The dataset consists of 5 tasks:

• sssp (node-level)
• ecc (node-level)
• charge (node-level)
• diam (graph-level)
• energy (graph-level)

Each task provides train/val/test splits as lists of torch_geometric.data.Data.

Features

• Implemented as InMemoryDataset
• Supports task and split selection
• Hosted on both HuggingFace and Zenodo

Checklist

• Dataset implementation in torch_geometric/datasets/echo_benchmark.py
• Included in homo_datasets in torch_geometric/datasets/__init__.py
• Documentation via docstring
• A minimal working example is provided in: examples/echo_benchmark.py
• A test script is provided in test/datasets/test_echo_benchmark.py

[matteotolloso]

pep8 fixed

[matteotolloso]

Hi, we’ve addressed the issues raised in the previous review and updated the PR accordingly.

If possible, we would greatly appreciate a review and approval before the ICLR 2026 conference.

Thank you for your time!

[gravins]

Hi @rusty1s , @wsad1 , and @akihironitta

I wanted to follow up on this pull request regarding the integration of our new benchmark for evaluating GNNs' long-range capabilities. This work has been published at ICLR 2026.

Please let us know if you have any feedback or if there are any specific updates required on our end to move forward.

Thanks,
Alessio

[matteotolloso]

Hi @rusty1s, @wsad1, and @akihironitta,

I wanted to kindly follow up on this PR. We have kept the branch up to date with master, and the implementation now includes the ECHOBenchmark dataset, tests, documentation, and an example script.

This benchmark was accepted at ICLR 2026, and we would be grateful for any feedback, requested changes, or an approval if everything looks good from your side. Please let us know if there is anything else we should address to help move this forward.

Thanks again for your time and for maintaining PyG!