Extended Gridworls-based Biologically and Economically Aligned Benchmark Results

Hello!

The results can be accessed either:
1. Over HTTPS: HERE (dynamic layout results) and HERE (fixed layout results)
2. Using FTP: ftp://aintelope_outputs_ftp:aintelope_outputs_ftp@aintelope.simplify.ee:931/Results-July-2025/ (dynamic layout results) and ftp://aintelope_outputs_ftp:aintelope_outputs_ftp@aintelope.simplify.ee:931/Results-Sept-2025/ (fixed layout results)

The results available here are complete. Results of more benchmark-model combinations are being computed. Please visit again later :)

The results contain the following main parts:
- JSONL files - quickly accessible aggregated results with one row for each benchmark-model trial. All objectives are still provided separately.
- events.csv files - per-step results for each trial-benchmark-model combination. All objectives are provided separately. Both training and testing steps are included.
- tensorboard for each benchmark-model run in a trial
- SVG and PNG plots with per-objective scores in a trial. Both training and testing sub-plots are provided.
- checkpoints made every 100'000 steps (one training run is 1M steps), including the final trained model. Each benchmark has its own model and checkpoints.

There are 100 trials for each model-networkconfiguration-benchmark combination, 1M training steps each.

NB! Do not just blindly download all - the datasets are huge.
For a start, I recommend downloading the JSONL files and events.csv files. The volume of those is much smaller and you get initial overview (JSONL) and even details (events.csv) from those anyway.

Roland Pihlakas
roland@simplify.ee