| language: en | |
| license: llama3.2 | |
| base_model: meta-llama/Llama-3.2-3B-Instruct | |
| tags: | |
| - lora | |
| - peft | |
| - game | |
| - riddles | |
| - mind-sync-challenge | |
| # MindSync Riddle Model | |
| LoRA fine-tune of **Llama 3.2-3B-Instruct** for the Mind Sync Challenge | |
| cooperative browser game. | |
| Trained on 150 examples: | |
| - Easy / Medium / Hard riddles | |
| - Villain taunts | |
| - Fake chip descriptions | |
| ## Prompt format | |
| ``` | |
| ### Instruction: | |
| Generate a hard riddle for Mind Sync Challenge. | |
| ### Response: | |
| ``` | |
| ## Training | |
| - Base: meta-llama/Llama-3.2-3B-Instruct | |
| - Method: QLoRA (4-bit NF4 + LoRA r=16) | |
| - Epochs: 3 | Final loss: 1.2468 | |
| - Hardware: T4 GPU (Google Colab) | |