Legal Chatbot SLM Indonesian (16-bit / QLoRA Fine-Tuned)

Informasi Model:

• Base Model: unsloth/Qwen2.5-Coder-7B-Instruct-bnb-4bit
• Dataset: Ichsan2895/alpaca-gpt4-indonesian (ChatML Format)
• Teknik Training: QLoRA 4-bit Double Quantization (SFT 800 steps) + GRPO (Group Relative Policy Optimization)
• Kemampuan: Penalaran penal hukum dengan token <think> ... </think> dan integrasi sistem Advanced RAG.