m0no1/llama31-dnd-mechanics

Llama 3.1 8B - D&D 3.5 Mechanics

Fine-tuned Llama 3.1 8B for converting D&D 3.5 player actions into game mechanics JSON.

Training

• Base: Llama 3.1 8B Instruct
• Dataset: 170k examples (o3 curated)
• Method: LoRA (r=64, alpha=128)
• Hardware: H100 GPU

Usage

# Example prompt
"Player: I cast fireball at the orcs"

# Output
{"action": "cast_spell", "detail": {"spell": "fireball", "target": "orcs"}}