Simplify the model card around evaluation on the repro branch.

README.md

@@ -9,62 +9,33 @@ tags:
   - opfdata
 ---
 
-# GENCO Base on OPFData (paper §5.2)
+Reproduction guide: [GENCO §5.2](https://github.com/gridfm/gridfm-paper-genco/blob/main/repro_instructions/5.2_OPF_on_opfdata.md).
 
-Eighteen GENCO Base checkpoints used for Table 5 of [GENCO](https://arxiv.org/abs/2608.09921): six OPFData grids, three seeds (`42`, `3`, `17`).
+GENCO optimal-power-flow checkpoints on OPFData ([paper](https://arxiv.org/abs/2608.09921)): IEEE 14, 30, 57, and 118, and GOC 500 and 2000, seeds 42, 3, and 17.
 
-HH-MPNN numbers in the paper are taken from Arowolo et al.; they are not in this repo.
-
-## Files
-
-```
-{case14_ieee,case30_ieee,case57_ieee,case118_ieee,case500_goc,case2000_goc}/seed{42,3,17}/
-  best_model_state_dict.pt
-  normalizer_stats.pt
-  metrics.csv
-  HGNSQ_penalty_11_OPFData_<case>_<default|3|17>.yaml
-splits/
-  train.pt  val.pt  test.pt  shuffled_indices.pt
-mlflow/
-  train/<grid>/seed{42,3,17}/<run_id>/
-  eval/<run_id>/
-```
-
-The training YAML sits in the same folder as the checkpoint. The same files are on GitHub: [`scripts/opfdata/configs`](https://github.com/gridfm/gridfm-graphkit/tree/genco-paper-repro/scripts/opfdata/configs). `default` is seed `42`.
-
-`metrics.csv` is the eval test-set scalar dump for that checkpoint (`gridfm_graphkit evaluate --batch_size 512`). `splits/` are the HH-MPNN 90/5/5 index files (270k / 15k / 15k). Put them at `scripts/opfdata/splits/` when training or evaluating from [gridfm-graphkit `genco-paper-repro`](https://github.com/gridfm/gridfm-graphkit/tree/genco-paper-repro).
-
-The [training and eval MLflow runs](https://huggingface.co/gridfm/genco-opfdata/tree/main/mlflow) are under `mlflow/` (params, tags, metrics, test CSVs and plots). Weights stay in `<grid>/seed*/`, not inside the run dirs. Table 5 uses the **eval** runs.
-
-## Install
+## Evaluate
 
-The paper results were obtained with [`genco-paper-repro`](https://github.com/gridfm/gridfm-graphkit/tree/genco-paper-repro). We only guarantee the same numbers on this branch, because `main` is under active development. Clone it and install from source:
+Use graphkit branch [genco-paper-repro](https://github.com/gridfm/gridfm-graphkit/tree/genco-paper-repro). We do not guarantee that these checkpoints work on `main`, which is under development.
 
 ```bash
-git clone --branch genco-paper-repro https://github.com/gridfm/gridfm-graphkit.git
+git clone -b genco-paper-repro https://github.com/gridfm/gridfm-graphkit.git
 cd gridfm-graphkit
-python -m venv venv && source venv/bin/activate
 pip install -e .
+pip install "huggingface_hub[cli]"
 TORCH_CUDA_VERSION=$(python -c "import torch; print(torch.__version__ + ('+cpu' if torch.version.cuda is None else ''))")
 pip install torch-scatter -f https://data.pyg.org/whl/torch-${TORCH_CUDA_VERSION}.html
 ```
 
-## Data
+Run the commands below from that repo root. Graphkit loads `{data_path}/{network}/raw/*.parquet`. Configs are in [`scripts/opfdata/configs`](https://github.com/gridfm/gridfm-graphkit/tree/genco-paper-repro/scripts/opfdata/configs). Copy `splits/` to `scripts/opfdata/splits/` before evaluating. Each checkpoint folder has `best_model_state_dict.pt` and `normalizer_stats.pt`.
 
-Converted OPFData parquet (already public). Graphkit loads `{data_path}/{network}/raw/*.parquet`.
-
-| Grid | Dataset |
-|---|---|
-| IEEE 14 | [gridfm/opfdata_case14_ieee](https://huggingface.co/datasets/gridfm/opfdata_case14_ieee) |
-| IEEE 30 | [gridfm/opfdata_case30_ieee](https://huggingface.co/datasets/gridfm/opfdata_case30_ieee) |
-| IEEE 57 | [gridfm/opfdata_case57_ieee](https://huggingface.co/datasets/gridfm/opfdata_case57_ieee) |
-| IEEE 118 | [gridfm/opfdata_case118_ieee](https://huggingface.co/datasets/gridfm/opfdata_case118_ieee) |
-| GOC 500 | [gridfm/opfdata_case500_goc](https://huggingface.co/datasets/gridfm/opfdata_case500_goc) |
-| GOC 2000 | [gridfm/opfdata_case2000_goc](https://huggingface.co/datasets/gridfm/opfdata_case2000_goc) |
-
-## Evaluate
-
-From the graphkit repo root:
+| Grid | Dataset | `network` folder |
+|---|---|---|
+| IEEE 14 | [gridfm/opfdata_case14_ieee](https://huggingface.co/datasets/gridfm/opfdata_case14_ieee) | `case14_ieee` |
+| IEEE 30 | [gridfm/opfdata_case30_ieee](https://huggingface.co/datasets/gridfm/opfdata_case30_ieee) | `case30_ieee` |
+| IEEE 57 | [gridfm/opfdata_case57_ieee](https://huggingface.co/datasets/gridfm/opfdata_case57_ieee) | `case57_ieee` |
+| IEEE 118 | [gridfm/opfdata_case118_ieee](https://huggingface.co/datasets/gridfm/opfdata_case118_ieee) | `case118_ieee` |
+| GOC 500 | [gridfm/opfdata_case500_goc](https://huggingface.co/datasets/gridfm/opfdata_case500_goc) | `case500_goc` |
+| GOC 2000 | [gridfm/opfdata_case2000_goc](https://huggingface.co/datasets/gridfm/opfdata_case2000_goc) | `case2000_goc` |
 
 ```bash
 mkdir -p data/case118_ieee/raw scripts/opfdata/splits
@@ -80,4 +51,18 @@ gridfm_graphkit evaluate \
   --batch_size 512
 ```
 
-Full commands: [GENCO §5.2](https://github.com/gridfm/gridfm-paper-genco/blob/main/repro_instructions/5.2_OPF_on_opfdata.md).
+Repeat for the other grids and seeds. Download each dataset into `data/<network>/raw`.
+
+## Files
+
+```
+{case14_ieee,case30_ieee,case57_ieee,case118_ieee,case500_goc,case2000_goc}/seed{42,3,17}/
+  best_model_state_dict.pt
+  normalizer_stats.pt
+  metrics.csv
+  HGNSQ_penalty_11_OPFData_<case>_<default|3|17>.yaml
+splits/
+  train.pt  val.pt  test.pt  shuffled_indices.pt
+mlflow/train/
+mlflow/eval/
+```