Card: list the training data (datasets:, base_model:) and add a Training data section

Follow-up to braindecode PR #1493: `from_pretrained` reads `datasets:` into `model.card_data`. Ids: MOABB names, NEMAR nm/on ids, OpenNeuro ds ids; evidence per checkpoint.

README.md

@@ -3,6 +3,10 @@ license: mit
 library_name: braindecode
 tags:
   - deprecated
+datasets:
+- nm000338
+- nm000323
+- nm000273
 ---
 
 # ⚠️ This repository is deprecated
@@ -30,3 +34,23 @@ model = SignalJEPA.from_pretrained("braindecode/signal-jepa_without-chans")
 See the docs for [`braindecode.models.SignalJEPA`](https://braindecode.org/stable/generated/braindecode.models.SignalJEPA.html)
 and the fine-tuning tutorial
 [Fine-tuning a Foundation Model (Signal-JEPA)](https://braindecode.org/stable/auto_examples/advanced_training/plot_finetune_foundation_model.html).
+
+<!-- braindecode:training-data -->
+## Training data
+
+The data that shaped these weights. The ids are the `datasets:` field of this card,
+which braindecode reads into `model.card_data.datasets` when you call
+`from_pretrained` (MOABB class names for MOABB datasets, OpenNeuro / NEMAR
+accessions, Hub ids otherwise). `nm`/`on` ids are datasets on [NEMAR](https://nemar.org)
+(`on` = NEMAR's copy of the OpenNeuro dataset with the same number). Rows marked
+`private:` are data the authors did not release; they are not in `datasets:`.
+
+| `datasets:` id | Dataset | Stage | Modality |
+|---|---|---|---|
+| [`nm000338`](https://nemar.org/dataset/nm000338) | [Lee et al. 2019 OpenBMI, motor imagery paradigm (54 subjects, 62 EEG channels), MOABB Lee2019_MI](https://moabb.neurotechx.com/docs/generated/moabb.datasets.Lee2019_MI.html) | pretraining | EEG |
+| [`nm000323`](https://nemar.org/dataset/nm000323) | [Lee et al. 2019 OpenBMI, ERP (P300 speller) paradigm, MOABB Lee2019_ERP](https://moabb.neurotechx.com/docs/generated/moabb.datasets.Lee2019_ERP.html) | pretraining | EEG |
+| [`nm000273`](https://nemar.org/dataset/nm000273) | [Lee et al. 2019 OpenBMI, SSVEP paradigm, MOABB Lee2019_SSVEP](https://moabb.neurotechx.com/docs/generated/moabb.datasets.Lee2019_SSVEP.html) | pretraining | EEG |
+
+- S-JEPA self-supervised pre-training (16 s examples, 60 % spatial mask, run adeuwv4s) on Lee2019 subjects 1-40, all three paradigms pooled, no labels; subjects 41-47 = SSL validation set used for early stopping / best-checkpoint selection; subjects 48-54 not used for these weights (downstream evaluation only). Only the pre-trained backbone tensors come from adeuwv4s (byte-identical); the SignalJEPA_PostLocal head (19 channels, 2 outputs, plus a 19-row spatial embedding for Contextual) has no trace of supervised training on any dataset (no fine-tuning is documented; the original README states the downstream head 'was not pre-trained').
+- Sources: arXiv:2403.11772 (HTML), section DATASETS; https://huggingface.co/api/models/braindecode/signal-jepa/commits/main (commit 9e20126cf5); braindecode check: tensor-byte sha256 check  -> compare_sjepa.json; W/bd/braindecode/models/signal_jepa.py.
+