Improve generation quality without retraining: add candidate reranking, few-shot prompts, aggressive filtering, and quality scoring

Fix attention mask warning by using tokenizer() and passing attention_mask to generate()

add query code

Upload training_metrics.json with huggingface_hub

Upload model.safetensors with huggingface_hub

Upload training_metrics.json with huggingface_hub

Replace gh readme with model card

remove unnecessary files

Move model files to repo root for Hugging Face compatibility

Remove data/ directory from git tracking

Add Australian Legal SLM model files

fix: code review improvements - error handling, validation, and safety

initial commit

initial commit