Short dataset card pointing at the genco-paper-repro README

README.md

@@ -1,82 +1,7 @@
 ---
 license: apache-2.0
-tags:
-  - power-systems
-  - power-flow
-  - optimal-power-flow
-  - powermodels
 ---
 
-# gridfm-datakit PowerModels setup-2 scenarios
+Corrected PowerModels JSON for the [GENCO](https://arxiv.org/abs/2608.09921) setup-2 (from-disk) runtime experiments.
 
-Paper: **[gridfm-datakit-v1: A Python Library for Scalable and Realistic Power Flow and Optimal Power Flow Data Generation](https://arxiv.org/abs/2512.14658)**
-
-The **10,000 PowerModels JSON scenarios per grid** used in the classical-solver
-runtime (setup 2 / from-disk) of that paper: AC-PF, DC-PF, AC-OPF, and DC-OPF
-wall-time figures. Files are the corrected pool
-(`scenario_*_corrected.json`). Indices wrap when the matrix asks for more than
-10,000 solves.
-
-Code and the 56 result CSVs live in `scripts/runtime/` of
-[`gridfm/gridfm-datakit`](https://github.com/gridfm/gridfm-datakit) (see that
-directory’s README).
-
-Data-diversity figures (different snapshot, `case118_ieee` parquet) live at
-[`gridfm/reproducibility-datakit-technical-report`](https://huggingface.co/datasets/gridfm/reproducibility-datakit-technical-report).
-
-## Layout
-
-Matches `$GRIDFM_DATA_BASE/{pf,opf}/<network>/powermodels/scenario_*_corrected.json`.
-
-| Path | Scenarios | Size |
-|------|----------:|-----:|
-| `pf/case14_ieee/powermodels/` | 10,000 | 0.09 GiB |
-| `pf/case30_ieee/powermodels/` | 10,000 | 0.18 GiB |
-| `pf/case57_ieee/powermodels/` | 10,000 | 0.33 GiB |
-| `pf/case118_ieee/powermodels/` | 10,000 | 0.85 GiB |
-| `pf/case500_goc/powermodels/` | 10,000 | 3.5 GiB |
-| `pf/case2000_goc/powermodels/` | 10,000 | 14.5 GiB |
-| `pf/case10000_goc/powermodels/` | 10,000 | 58.6 GiB |
-| `opf/case14_ieee/powermodels/` | 10,000 | 0.09 GiB |
-| `opf/case30_ieee/powermodels/` | 10,000 | 0.18 GiB |
-| `opf/case57_ieee/powermodels/` | 10,000 | 0.33 GiB |
-| `opf/case118_ieee/powermodels/` | 10,000 | 0.85 GiB |
-| `opf/case500_goc/powermodels/` | 10,000 | 3.5 GiB |
-| `opf/case2000_goc/powermodels/` | 10,000 | 14.5 GiB |
-| `opf/case10000_goc/powermodels/` | 10,000 | 58.6 GiB |
-
-**Total:** 140,000 files, about **156 GiB**. PF and OPF splits are both required: PF/DCPF jobs read `pf/`, OPF/DCOPF jobs read `opf/`. GOC 2000/10000 are most of the bytes.
-
-## Download
-
-The full tree is large (GOC 2000/10000 dominate). Download only what you need:
-
-```bash
-# IEEE + case500 only (small jobs)
-hf download gridfm/reproducibility-powermodels-setup2 --repo-type dataset \
-    --local-dir /path/to/finetuning \
-    --exclude "pf/case2000_goc/**" --exclude "pf/case10000_goc/**" \
-    --exclude "opf/case2000_goc/**" --exclude "opf/case10000_goc/**"
-
-# one network
-hf download gridfm/reproducibility-powermodels-setup2 --repo-type dataset \
-    --local-dir /path/to/finetuning \
-    --include "pf/case14_ieee/**" --include "opf/case14_ieee/**"
-```
-
-Then:
-
-```bash
-export GRIDFM_DATA_BASE=/path/to/finetuning
-# already scenario_*_corrected.json — do not run run_correction.sh
-bash scripts/runtime/pure_julia/run_small_setup2.sh
-```
-
-Setup 1 does not use this dataset (it solves the corrected `.m` base case in memory).
-
-## Notes
-
-- Original cluster path: `/dccstor/gridfm/powermodels_data/v4/finetuning`.
-- Uncorrected `scenario_*.json` files are **not** included; only `*_corrected.json`.
-- The `apache-2.0` licence applies to this converted snapshot and the datakit code.
-  Underlying PGLib / GOC network models retain their original terms.
+How to reproduce: [`scripts/runtime/README.md`](https://github.com/gridfm/gridfm-datakit/blob/genco-paper-repro/scripts/runtime/README.md) on branch `genco-paper-repro`.