Add metadata and link to AdaGC paper
#1
by nielsr HF Staff - opened
README.md
CHANGED
|
@@ -1,5 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
# Self-Healing Training System (SHTS)
|
| 2 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
> **Fully autonomous debugging and error recovery for Hugging Face TRL trainers. Add one callback, wrap with `SelfHealingTrainer`, and cut debugging costs to near zero.**
|
| 4 |
|
| 5 |
[](https://opensource.org/licenses/MIT)
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
library_name: transformers
|
| 4 |
+
pipeline_tag: text-generation
|
| 5 |
+
---
|
| 6 |
+
|
| 7 |
# Self-Healing Training System (SHTS)
|
| 8 |
|
| 9 |
+
This repository is associated with the paper [AdaGC: Enhancing LLM Pretraining Stability via Adaptive Gradient Clipping](https://huggingface.co/papers/2502.11034).
|
| 10 |
+
|
| 11 |
+
Official implementation: [PaddleFleet](https://github.com/PaddlePaddle/PaddleFleet) (see Research/AdaGC).
|
| 12 |
+
|
| 13 |
> **Fully autonomous debugging and error recovery for Hugging Face TRL trainers. Add one callback, wrap with `SelfHealingTrainer`, and cut debugging costs to near zero.**
|
| 14 |
|
| 15 |
[](https://opensource.org/licenses/MIT)
|