---
base_model: meta-llama/Meta-Llama-3.1-8B-Instruct
library_name: peft
license: llama3.1
tags:
  - lora
  - sft
  - npc
  - roleplay
  - dialogue
  - game-ai
language:
  - en
---

# NPCAlign SFT — NPC Quest Dialogue LoRA

LoRA adapter fine-tuned on top of Llama-3.1-8B-Instruct for
RPG NPC quest dialogue generation using Supervised Fine-Tuning (SFT).

## Model Details
- **Base model**: meta-llama/Meta-Llama-3.1-8B-Instruct
- **Training data**: chimbiwide/NPC-Quest-Dialogue (~1,584 training conversations)
- **Method**: SFT with LoRA (rank 16, all attention + MLP projections)
- **Task**: Given NPC background, generate character-consistent quest dialogue

## Usage
> **Note:** The base model [`meta-llama/Meta-Llama-3.1-8B-Instruct`](https://huggingface.co/meta-llama/Meta-Llama-3.1-8B-Instruct) is a gated model. You must accept Meta's license and set your `HF_TOKEN` before loading.

```
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
import torch

base = AutoModelForCausalLM.from_pretrained(
    "meta-llama/Meta-Llama-3.1-8B-Instruct",
    torch_dtype=torch.bfloat16, device_map="auto"
)
model = PeftModel.from_pretrained(base, "HermitQ/NPCAlign-SFT")
tokenizer = AutoTokenizer.from_pretrained("HermitQ/NPCAlign-SFT")
```

## Training Details
| Parameter | Value |
|---|---|
| LoRA rank | 16 |
| LoRA alpha | 32 |
| Epochs | 3 |
| Learning rate | 2e-4 |
| Max length | 3072 tokens |
| Loss masking | Assistant turns only |

## Evaluation (Test Set, Phase-Stratified)
| Phase      | ROUGE-L | Self-BLEU | BERTScore-F1 | BLEURT |
| ---------- | ------- | --------- | ------------ | ------ |
| Openning   | 0.264   | 0.208     | 0.883        | -0.692 |
| Dvelopment | 0.231   | 0.150     | 0.880        | -0.719 |
| Resolution | 0.259   | 0.269     | 0.885        | -0.719 |
| Overall    | 0.251   | 0.264     | 0.883        | -0.710 |