Update README for epoch 300

README.md

@@ -1,3 +1,53 @@
 ---
 license: mit
+tags:
+- robotics
+- libero
+- oat
+- dense-visual-memory
 ---
+
+# OAT Dense LIBERO-10 — Checkpoint Epoch 300
+
+Hugging Face model repository for a **dense cross-attention OAT policy** trained on
+**LIBERO-10 (N500)**. This snapshot was taken at **epoch 300** during a long run
+(`.tmp_ladder_logs`).
+
+## Files
+
+| File | Description |
+|------|-------------|
+| `ep-0300.ckpt` | PyTorch workspace checkpoint (~729 MB) |
+| `ep-0300_upload_meta.json` | Offline metrics at upload time |
+| `training_metrics_dashboard.png` | Training curves (dark dashboard) |
+
+## Offline metrics @ epoch 300
+
+| Metric | Value |
+|--------|------:|
+| Train loss | 2.4420902729034424 |
+| Val loss | 5.05718469619751 |
+| Test reconst MSE | 0.07346421480178833 |
+
+![Training metrics dashboard](training_metrics_dashboard.png)
+
+## Model configuration (summary)
+
+- **Policy:** OAT with `use_dense_visual_memory=true` (spatial visual tokens + cross-attn)
+- **State memory:** enabled (`use_state_memory_tokens=true`)
+- **Task UID:** enabled in state tokens
+- **Dataset:** `libero10_N500.zarr`
+- **Embed dim:** 256
+
+## Evaluation note
+
+Training used `lazy_eval=true` (no in-loop LIBERO rollout). **Success rate (SR)** must be
+measured separately via `scripts/eval_policy_sim.py` on this checkpoint.
+
+Reference (paper, pooled OAT8 on LIBERO-10): **~56.3%** mean success rate — external
+baseline, not reproduced in this repo.
+
+## Citation
+
+If you use this checkpoint, please cite **OAT: Ordered Action Tokenization** and specify
+epoch **300** of the dense LIBERO-10 ladder.