| base_model: Qwen/Qwen3-4B-Instruct-2507 | |
| library_name: peft | |
| tags: | |
| - lora | |
| - qwen3 | |
| - build-small-hackathon | |
| - facade-of-jade | |
| - modal | |
| # Facade of Jade Qwen3-4B LoRA | |
| LoRA adapter trained for **Facade of Jade**, a Build Small Hackathon interactive wuxia drama demo. | |
| - Base model: `Qwen/Qwen3-4B-Instruct-2507` | |
| - Training records: 50 | |
| - Epochs: 3 | |
| - Final train loss: `2.969015` | |
| - Adapter size reported by Modal runner: `483.63 MB` | |
| - Modal run evidence: https://modal.com/apps/t-abdullah-rashid/main/ap-W54lCMfJu4eu3UCVQvVpQK | |
| - Source repo: https://github.com/tuancookiez-hub/facade-of-jade | |
| - Live Space: https://build-small-hackathon-facade-of-jade.hf.space | |
| This adapter was produced by `train_lora_modal.py` on Modal A100-80GB and saved from Modal volume `facade-of-jade-lora-out`. | |