| --- |
| license: mit |
| library_name: transformers |
| tags: |
| - reinforcement-learning |
| - cognitive-offloading |
| - education |
| - dqn |
| - deberta |
| - pytorch |
| --- |
| |
| # Cognitive Offloading Detection & Adaptive Pedagogical Policy Models |
|
|
| Model artifacts for the [Cognitive Offloading Prevention System](https://github.com/Addy0606/cognitiveoffloading). |
|
|
| ## Contents |
| - **`classifier_v1/`**: `microsoft/deberta-base` fine-tuned to classify student messages for cognitive offloading / answer extraction behavior. |
| - `backbone/`: Tokenizer files and transformer base weights (`model.safetensors`). |
| - `head.pth`: 2-layer MLP classification head (`768 -> 256 -> 64 -> 2`). |
| - `config.json`: Model configuration and hyperparameters. |
| - **`policy_network_v1.pth`**: Deep Q-Network (DQN) policy trained via reinforcement learning to select dynamic pedagogical intervention actions (`redirect`, `scaffold`, `socratic`, `delay`) balancing learning gains against engagement friction. |
| |
| ## Usage |
| To download these models locally: |
| ```python |
| from huggingface_hub import snapshot_download |
| |
| snapshot_download( |
| repo_id="addy0606/cognitive-offloading", |
| local_dir="models", |
| ) |
| ``` |
| |
| ## GitHub Repository |
| Source code, evaluation pipeline, and Flask interface: |
| https://github.com/Addy0606/cognitiveoffloading |
| |
