細かいミス修正

tagパース解除

不可視文字の処理

logの量を減らす

log周りの修正

Fix model loading for repo ID and improve GPU detection

Load model directly from Hub to save storage space

Switch from llama-cpp-python to transformers for PyTorch model support

Add GPU detection and fallback to CPU mode

Remove local model and add auto-download

Add FastAPI word tree server