Upload folder using huggingface_hub

README.md

@@ -34,6 +34,11 @@ model-index:
 
 > **A JAX/Flax language model that separates *what it knows* from *how it thinks* — so the knowledge can grow to 100B+ vectors while inference stays fast and cheap.**
 
+[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1VM64IOZHj5rDvxWPbqktC037LlyOJih3?usp=sharing)
+
+> [!WARNING]
+> **Disclaimer**: This repository and checkpoint are provided as a **research proof-of-concept to demonstrate the novel DPSNR architecture**. It is an experimental model trained on a limited compute budget (for ~31,000 steps) to validate theoretical claims (such as $O(1)$ retrieval scaling, Sparse Adam optimizer speedups, and memory-bandwidth properties). **It is NOT a fully-trained competitive model** and is not intended to compete with state-of-the-art open-source text models (like LLaMA or Mistral) on downstream benchmarks.
+
 ---
 
 ## What Is DPSNR?