Python
Update README.md
SNOW TEAMb258a1b03af6
diff --git a/README.md b/README.mdindex 57a27fa..74eed21 100644--- a/README.md+++ b/README.md@@ -18,20 +18,57 @@ Qi Chen, [Siria Xiyueyao Luo](https://www.rug.nl/staff/x.luo/?lang=en), [Xuejiao[](https://github.com/SNOWTEAM2023/GEM/stargazers)+## :fire: News+* **[2026.02.01]** We release github repository of **Cognivia**. πͺ Have a tryοΌ-## Overview+## π§ Framework Overview<p align="center"><img src="https://github.com/SNOWTEAM2023/Cognivia/blob/main/overview.jpg" width="400"></p>+**GEM** aligns base LLM using human preference data by a **Coginitive Feedback Loop**, which includes **Cognitive Filtering** and **SEGA** modules.+Key modules of GEM include:++- **Cognitive Filtering**: Generate `k` Chain-of-Thought (CoTs) candidates per query and **rank** them by Entropy-guided Token Scoring module. Entropy-guided Token Scoring module encourages **exploration midβCoT** (high entropy on topβm steps) and **confidence at the end** (low final entropy).+- **SEGA**: A **listwise** objective that updates the policy using **group-meanβcentered advantages**, which update with weights proportional to **Aα΅’ = rα΅’ β rΜ** within each k-way group.## Dataset+1. **CBT Cognitive Triplet Dataset**:+2. **Augmented CBT Cognitive Triplet Dataset**:+++### β¨ Code Structure+The code structure and corresponding comments of this repository are as follows:++```+GEM/+βββ GEM.py # Main entry script for running GEM+βββ data/ # Example preference data+β βββ preference_data.jsonl+β+βββ src/ # Core implementation of GEM+β βββ __init__.py+β βββ config.py # Configuration utilities+β βββ dataset.py # Dataset & dataloader definitions+β βββ entropy_scorer.py # Entropy-based scoring+β βββ gem_trainer.py # GEM training pipeline+β βββ sft_trainer.py # Supervised fine-tuning (SFT) trainer+β+βββ materials/ # Figures & assets for the paper+βββ README.md # Project introduction and usage+βββ LICENCE.txt # Licence information+βββ requirements.txt # Python dependencies+```+## Usage## Experiments# Main results## Evaluation+## π License+This work is licensed under the [Creative Commons Attribution-NonCommercial 4.0 International License](http://creativecommons.org/licenses/by-nc/4.0/).+Commercial use is prohibited without a separate license agreement with the author.