| # Legacy NeurIPS 2025 checkpoints -- diffusion_2d_cond | |
| Checkpoints from **EngiBench: A Framework for Data-Driven Engineering Design | |
| Research** (NeurIPS 2025 Datasets & Benchmarks, arXiv:2508.00831), migrated | |
| off Weights & Biases. | |
| > **These are NOT comparable to current EngiOpt checkpoints.** | |
| > The EngiBench problem definitions have changed since the paper, so a result | |
| > here and a result in `IDEALLab/engiopt-diffusion-2d-cond` for the same | |
| > problem id measure different things. They are kept under a separate repo | |
| > prefix for exactly this reason. Use these only to reproduce the paper. | |
| Layout: `{problem_id}/{wandb_collection}/{version}/`. See `legacy_index.json` for provenance (W&B run id, aliases, seed, timestamps). | |