| Stackcraft | |
| Copyright 2026 Stackcraft contributors. | |
| Stackcraft is an independent educational falling-block game and fine-tuning study. | |
| It is not affiliated with or endorsed by Tetris, Cloudflare, or the Qwen project. | |
| The optional trained model derives from Cloudflare/Clef-Flash (Apache-2.0), | |
| revision 17f0b0ad64efb65d273590632833508766b2aae6, which derives from | |
| Qwen/Qwen3.5-9B. Original backbone weights and native joint-schema implementation | |
| remain upstream dependencies; this repository does not redistribute the backbone. | |
| Stackcraft adds game-specific LoRA matrices and trains a separate native decision | |
| head. The trained head and LoRA weights must be loaded together. | |
| Upstream model: https://huggingface.co/Cloudflare/clef-flash | |
| Base architecture: https://huggingface.co/Qwen/Qwen3.5-9B | |
| Python package dependency licenses remain with their respective authors. | |