Create checkpoints/readme.md

checkpoints/readme.md

@@ -0,0 +1,46 @@
+---
+license: mit
+tags:
+- pytorch
+- jepa
+- self-supervised-learning
+- checkpoints
+---
+
+# Core JEPA: Pretrained Checkpoints
+
+[![PyTorch](https://img.shields.io/badge/PyTorch-%23EE4C2C.svg?style=for-the-badge&logo=PyTorch&logoColor=white)](https://pytorch.org/)
+
+This repository contains the weights and checkpoints for the **Core JEPA** (based on LeJEPA) model.
+
+Due to storage optimization, the heavy weights are hosted in the associated artifacts repository but can be accessed directly via the links below.
+
+## 📥 Download Weights
+
+| Model Variant | Filename | Status | Direct Link |
+| :--- | :--- | :--- | :--- |
+| **LeJEPA-Large** | `lejepa-l.pt` | ✅ Available | [**Download .pt File**](https://huggingface.co/datasets/gajeshladharai/artifacts/resolve/main/core-jepa/lejepa-l.pt) |
+
+---
+
+## 💻 Usage
+
+### Option 1: Load directly in Python (Recommended)
+You can load these weights directly into the `mapminer`  model using `torch.hub`. This handles downloading, caching, and key remapping automatically.
+```python
+from mapminer import models 
+
+jepa = models.DINOv3(pretrained=False) 
+
+ckpt = "https://huggingface.co/datasets/gajeshladharai/artifacts/resolve/main/core-jepa/lejepa-l.pt"
+ckpt = torch.hub.load_state_dict_from_url(ckpt, map_location='cpu')
+jepa.load_state_dict({k.replace('encoder.model.', 'model.'): v for k, v in ckpt.items()},strict=False)
+jepa.eval()
+
+# x =  uint8 image
+# normalize with model's preprocess
+x = jepa.normalize(x)
+
+# forward pass
+with torch.no_grad():
+    emb = jepa(x)