Point evaluation at the reproduction guide, with a short main-branch variant.

README.md

@@ -9,67 +9,13 @@ tags:
   - datakit
 ---
 
-Reproduction guide: [GENCO §5.4.2](https://github.com/gridfm/gridfm-paper-genco/blob/main/repro_instructions/5.4.2_PF_on_datakit.md).
-
 GENCO power-flow checkpoints on datakit grids ([paper](https://arxiv.org/abs/2608.09921)). Base, Small, and Tiny (hidden size 48, 24, and 12) on IEEE 14, 30, 57, and 118 and GOC 500. Small and Tiny on GOC 2000 and GOC 10000.
 
-## Evaluate
-
-Use graphkit branch [genco-paper-repro](https://github.com/gridfm/gridfm-graphkit/tree/genco-paper-repro). Graphkit requires Python 3.10, 3.11, or 3.12. The paper numbers come from this branch. [Evaluate on main](#evaluate-on-main) is the same checkpoints on current `main`.
-
-```bash
-git clone -b genco-paper-repro https://github.com/gridfm/gridfm-graphkit.git
-cd gridfm-graphkit
-pip install -e .
-pip install huggingface_hub
-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
-export MLFLOW_ALLOW_FILE_STORE=true
-```
-
-Run the commands below from that repo root. Graphkit loads `{data_path}/{network}/raw/*.parquet`. Configs are in [`scripts/datakit_pf/configs`](https://github.com/gridfm/gridfm-graphkit/tree/genco-paper-repro/scripts/datakit_pf/configs). Pass the weights and `normalizer_stats.pt` from the same folder.
+Evaluate on the paper branch by following [GENCO §5.4.2](https://github.com/gridfm/gridfm-paper-genco/blob/main/repro_instructions/5.4.2_PF_on_datakit.md). That guide uses graphkit [genco-paper-repro](https://github.com/gridfm/gridfm-graphkit/tree/genco-paper-repro) and is how the paper numbers were produced.
 
-| Grid | Dataset | `network` folder |
-|---|---|---|
-| IEEE 14 | [gridfm/pf_small_case14_ieee](https://huggingface.co/datasets/gridfm/pf_small_case14_ieee) | `case14_ieee` |
-| IEEE 30 | [gridfm/pf_small_case30_ieee](https://huggingface.co/datasets/gridfm/pf_small_case30_ieee) | `case30_ieee` |
-| IEEE 57 | [gridfm/pf_small_case57_ieee](https://huggingface.co/datasets/gridfm/pf_small_case57_ieee) | `case57_ieee` |
-| IEEE 118 | [gridfm/pf_small_case118_ieee](https://huggingface.co/datasets/gridfm/pf_small_case118_ieee) | `case118_ieee` |
-| GOC 500 | [gridfm/pf_small_case500_goc](https://huggingface.co/datasets/gridfm/pf_small_case500_goc) | `case500_goc` |
-| GOC 2000 | [gridfm/pf_small_case2000_goc](https://huggingface.co/datasets/gridfm/pf_small_case2000_goc) | `case2000_goc` |
-| GOC 10000 | [gridfm/pf_small_case10000_goc](https://huggingface.co/datasets/gridfm/pf_small_case10000_goc) | `case10000_goc` |
+Or evaluate the same checkpoints on graphkit [main](https://github.com/gridfm/gridfm-graphkit/tree/main). `main` is under development, so the same numbers are not guaranteed. They matched the paper-branch result for IEEE 14, Base, seed 0, at commit [`f104900`](https://github.com/gridfm/gridfm-graphkit/commit/f104900245cba26d1380037eea562cf633a97484) (1 October 2026). Clone `main` instead of `genco-paper-repro`, follow the rest of the install and the downloads in the guide, then run:
 
 ```bash
-mkdir -p data/case118_ieee/raw
-hf download gridfm/pf_small_case118_ieee --repo-type dataset --local-dir data/case118_ieee/raw
-hf download gridfm/genco-pf-datakit --include "case118_ieee/base/seed0/**" --local-dir genco-pf-datakit
-
-gridfm_graphkit evaluate \
-  --config scripts/datakit_pf/configs/case118_ieee_base_seed0.yaml \
-  --data_path data \
-  --model_path genco-pf-datakit/case118_ieee/base/seed0/best_model_state_dict.pt \
-  --normalizer_stats genco-pf-datakit/case118_ieee/base/seed0/normalizer_stats.pt
-```
-
-Repeat for the other grids, sizes, and seeds. Download each dataset into `data/<network>/raw`. For GOC 10000, pass `last.pt` instead of `best_model_state_dict.pt`. The first `evaluate` writes one graph file per scenario before testing.
-
-## Evaluate on main
-
-The same checkpoints run on current [main](https://github.com/gridfm/gridfm-graphkit/tree/main). This was checked at commit `f104900`. `main` does not ship these YAML files, and it rejects them as config version 0. Use the YAML downloaded with the checkpoint, and pass `--config_conversion auto_inline`. That upgrades the config in memory and leaves the file unchanged.
-
-```bash
-git clone https://github.com/gridfm/gridfm-graphkit.git
-cd gridfm-graphkit
-pip install -e .
-pip install huggingface_hub
-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
-export MLFLOW_ALLOW_FILE_STORE=true
-
-mkdir -p data/case118_ieee/raw
-hf download gridfm/pf_small_case118_ieee --repo-type dataset --local-dir data/case118_ieee/raw
-hf download gridfm/genco-pf-datakit --include "case118_ieee/base/seed0/**" --local-dir genco-pf-datakit
-
 gridfm_graphkit evaluate \
   --config_conversion auto_inline \
   --config genco-pf-datakit/case118_ieee/base/seed0/case118_ieee_base_seed0.yaml \
@@ -78,21 +24,4 @@ gridfm_graphkit evaluate \
   --normalizer_stats genco-pf-datakit/case118_ieee/base/seed0/normalizer_stats.pt
 ```
 
-Repeat for the other grids, sizes, and seeds. For GOC 10000, pass `last.pt` instead of `best_model_state_dict.pt`. On IEEE 14, Base, seed 0, this matched the reproduction-branch evaluation.
-
-## Files
-
-```
-{case14,case30,case57,case118}_ieee/{base,small,tiny}/seed{0,1,42}/
-case500_goc/{base,small,tiny}/seed{0,1,42}/
-case2000_goc/{small,tiny}/seed{0,42}/
-case10000_goc/{small,tiny}/seed{0,1}/
-  best_model_state_dict.pt
-  normalizer_stats.pt
-  metrics.csv
-  <network>_<size>_seed<seed>.yaml
-case10000_goc/{small,tiny}/seed{0,1}/
-  last.pt
-mlflow/eval/
-mlflow/train/
-```
+`main` does not ship the paper YAML and rejects it as config version 0. `--config_conversion auto_inline` upgrades that file in memory. The YAML is the one downloaded with the checkpoint. Repeat for the other grids, sizes, and seeds. For GOC 10000, pass `last.pt` instead of `best_model_state_dict.pt`.