init: add LoRA training infrastructure and 20k dataset

.gitignore

@@ -0,0 +1,27 @@
+# Trained models
+training/output/
+*.safetensors
+*.bin
+
+# Training artifacts
+__pycache__/
+*.pyc
+*.pyo
+
+# Virtual environments
+venv/
+.venv/
+
+# Jupyter notebooks
+*.ipynb
+.ipynb_checkpoints/
+
+# OS files
+.DS_Store
+Thumbs.db
+
+# Logs
+*.log
+
+# Large dataset (optional, use git-lfs if needed)
+# dataset/*.jsonl