README.md

---
language: ko
license: apache-2.0
tags:
  - korean
---

# Chat Model QLoRA Adapter

Fine-tuned QLoRA Adapter for model [OrionStarAI/Orion-14B-Base](https://huggingface.co/OrionStarAI/Orion-14B-Base)

Fine-tuned with Korean Sympathy Conversation dataset from AIHub

See more informations at [our GitHub](https://github.com/boostcampaitech6/level2-3-nlp-finalproject-nlp-09)

## Datasets

- [공감형 대화](https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&dataSetSn=71305)

## Quick Tour
```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "CurtisJeon/OrionStarAI-Orion-14B-Base-4bit",
    low_cpu_mem_usage=True,
    trust_remote_code=True,
    device_map="auto",
)
model.config.use_cache = True

model.load_adapter("m2af/OrionStarAI-Orion-14B-Base-adapter", "loaded")
model.set_adapter("loaded")

tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
tokenizer.pad_token = tokenizer.eos_token
tokenizer.padding_side = "right"

# Generate Sample
outputs = model.generate(**tokenizer("안녕하세요, 반갑습니다.", return_tensors="pt"))
print(outputs)
```