CityQuest Nemotron 3 Nano 4B (Q4_K_M GGUF)

LoRA fine-tune of nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16 on the CityQuest location-based game dataset (~840 train / 92 val examples across scavenger hunt, hide-and-seek and tag).

Trained to emit games directly in the app's game_schema.json contract (non-reasoning / JSON-only). Drop-in replacement for the stock Nemotron GGUF in app/services/generator.py.

• LoRA r=16, alpha=16, epochs=3, lr=0.0002
• Loaded via native transformers (arch nemotron_h); MoE router / lm_head excluded
• Quantization: Q4_K_M · File: CityQuest-Nemotron-3-Nano-4B-Q4_K_M.gguf