Upload .\docs\training.md with huggingface_hub

.//docs//training.md

@@ -0,0 +1,29 @@
+# Training
+
+## Overview
+
+BwengeAi supports fine-tuning Llama models on Kinyarwanda data using LoRA.
+
+## Base Models
+
+| Model | Parameters | Use Case |
+|-------|-----------|----------|
+| meta-llama/Llama-3.2-1B | 1B | Development/testing |
+| almanach/Llama-3-8B-mono-Kinyarwanda | 8B | Production (recommended) |
+| almanach/Llama-2-7B-mono-Kinyarwanda | 7B | Production |
+
+## Training Modes
+
+- **Full fine-tuning**: All parameters trained
+- **LoRA**: Adapter layers only (recommended)
+- **QLoRA**: Quantized LoRA for memory efficiency
+
+## Running
+
+```bash
+python scripts/train.py
+```
+
+## Configuration
+
+Training parameters are in `configs/default.yaml` under `training:` and `model:`.