Upload folder using huggingface_hub

README.md

@@ -0,0 +1,64 @@
+---
+library_name: tedbench
+tags:
+  - protein
+  - structure
+  - fold-classification
+  - tedbench
+pipeline_tag: other
+license: bsd-3-clause
+---
+
+# TEDBench — Supervised from scratch (structure + sequence)
+
+**Variant:** `miae_b` + seq  |  **Parameters:** 102M  |  **Layers:** 12  |  **Hidden dim:** 768  |  **Attn heads:** 12
+
+This **MiAEClassifier** was trained from scratch on TEDBench without pretraining.
+
+> **+seq variant** — sequence embeddings are concatenated to the geometric encoder input (`model.use_seq_input=true`).
+
+Part of the [TEDBench](https://github.com/BorgwardtLab/TEDBench) benchmark for
+protein fold classification (ICML 2026). MiAE is an SE(3)-invariant masked
+autoencoder that masks up to 90% of backbone frames and reconstructs the full
+structure with a lightweight decoder.
+
+## Architecture sizes
+
+| Variant | Params | Layers | Hidden dim | Attn heads |
+|---------|-------:|-------:|-----------:|-----------:|
+| `miae_s` | 29 M | 6 | 512 | 8 |
+| `miae_b` | 102 M | 12 | 768 | 12 |
+| `miae_l` | 339 M | 24 | 1 024 | 16 |
+
+Append `+model.use_seq_input=true` to `miae_b` for the **+seq** variant.
+
+## Usage
+
+### Load from the HuggingFace Hub
+
+```python
+from tedbench.utils.io import load_from_hf
+
+model = load_from_hf("TEDBench/miae-b-seq-sc")
+model.eval()
+```
+
+### From a Lightning checkpoint
+
+```python
+from tedbench.model import MiAEClassifier
+
+model = MiAEClassifier.load_from_checkpoint("model.ckpt", weights_only=False)
+model.eval()
+```
+
+## Citation
+
+```bibtex
+@inproceedings{chen2026tedbench,
+  title={Protein Fold Classification at Scale: Benchmarking and Pretraining},
+  author={Chen, Dexiong and Manolache, Andrei and Niepert, Mathias and Borgwardt, Karsten},
+  booktitle={Proceedings of the 43rd International Conference on Machine Learning},
+  year={2026}
+}
+```

config.json

@@ -0,0 +1,73 @@
+{
+  "seed": 42,
+  "debug": false,
+  "wandb": true,
+  "pretrained_model_path": null,
+  "datamodule": {
+    "batch_size": 32,
+    "pin_memory": true,
+    "num_workers": 32,
+    "train_transform": {
+      "_target_": "tedbench.data.transform.Compose",
+      "transforms": [
+        {
+          "_target_": "tedbench.data.transform.RandomCrop",
+          "size": 512
+        },
+        {
+          "_target_": "tedbench.data.transform.RandomNoise",
+          "std": 0.2,
+          "mean": 0.0
+        }
+      ]
+    },
+    "_target_": "tedbench.data.TEDLightningDataset",
+    "root": "./datasets/ted",
+    "dataset_name": "ted"
+  },
+  "trainer": {
+    "_target_": "pytorch_lightning.Trainer",
+    "accelerator": "auto",
+    "max_steps": 18300,
+    "strategy": "auto",
+    "devices": "auto",
+    "default_root_dir": "${logs.path}",
+    "num_sanity_val_steps": 0,
+    "accumulate_grad_batches": 32
+  },
+  "train": {
+    "optimizer": {
+      "_target_": "torch.optim.AdamW",
+      "lr": 0.0016,
+      "weight_decay": 0.1,
+      "betas": [
+        0.9,
+        0.95
+      ]
+    },
+    "lr_scheduler": {
+      "_target_": "tedbench.lr_schedulers.get_cosine_schedule_with_warmup",
+      "warmup_steps": 1830,
+      "max_steps": "${trainer.max_steps}"
+    },
+    "loss": {
+      "_target_": "torch.nn.CrossEntropyLoss",
+      "label_smoothing": 0.0
+    },
+    "llrd": 1.0,
+    "ckpt_path": null
+  },
+  "model": {
+    "_target_": "tedbench.model.miae_encoder_model",
+    "name": "miae_b",
+    "num_classes": 965,
+    "avg_pool": false,
+    "use_seq_input": true
+  },
+  "logs": {
+    "prefix": "logs/finetune/${datamodule.dataset_name}/${seed}",
+    "path": "${logs.prefix}/runs/${now:%Y-%m-%d}_${now:%H-%M-%S}"
+  },
+  "mode": {},
+  "_model_class": "miae_classifier"
+}

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b5f5159ec6f669a1f44b94e243c0be5694efdb2e527b59d5e379fa2a197fa03c
+size 410442959