Add metadata and link to AdaGC paper

Hi! I'm Niels from the Hugging Face community science team.

I noticed this repository implements techniques from the paper [AdaGC: Enhancing LLM Pretraining Stability via Adaptive Gradient Clipping](https://huggingface.co/papers/2502.11034), but it currently lacks structured metadata to help users find it on the Hub.

This PR adds:
- YAML metadata including the `pipeline_tag`, `library_name`, and `license`.
- A direct link to the paper page on Hugging Face.
- A link to the official code repository.

These changes will ensure the repository is correctly indexed and more discoverable for researchers working on LLM pretraining stability.

README.md

@@ -1,5 +1,15 @@
+---
+license: mit
+library_name: transformers
+pipeline_tag: text-generation
+---
+
 # Self-Healing Training System (SHTS)
 
+This repository is associated with the paper [AdaGC: Enhancing LLM Pretraining Stability via Adaptive Gradient Clipping](https://huggingface.co/papers/2502.11034).
+
+Official implementation: [PaddleFleet](https://github.com/PaddlePaddle/PaddleFleet) (see Research/AdaGC).
+
 > **Fully autonomous debugging and error recovery for Hugging Face TRL trainers. Add one callback, wrap with `SelfHealingTrainer`, and cut debugging costs to near zero.**
 
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)