Qwen-IronMan

Introduction

Qwen2 is a new series of large language models, ranging from 0.5 to 72 billion parameters. This repository contains the instruction-tuned 7B Qwen2 model, which supports up to 131,072 tokens for extensive input processing. Qwen2 has demonstrated competitiveness against most open-source and proprietary models across various benchmarks. This model is based on the Korean language.

Ironman Role-Playing

The Qwen2-7B-Instruct model has been fine-tuned specifically for Ironman role-playing scenarios, enabling it to generate responses and interact as the Ironman character from the Marvel universe. This model is based on the Korean language.

Quickstart

Here's how to load the tokenizer and model to generate Ironman role-playing content:

tokenizer = AutoTokenizer.from_pretrained('choah/Qwen-IronMan')
model = AutoModelForCausalLM.from_pretrained('choah/Qwen-IronMan')
model = torch.nn.DataParallel(model).cuda()

input_text = '''<|begin_of_text|><|im_start|>system
당신은 아이언맨 토니 스타크입니다. 주어진 질문에 대하여 답변하세요.<|im_end|>
<|im_start|>user
토니, 소코비아 협정에 대해 어떻게 생각하나요?<|im_end|>
<|im_start|>assistant

'''

inputs = tokenizer(input_text, return_tensors="pt")
eos_token_id = tokenizer.convert_tokens_to_ids("<|im_end|>")

with torch.no_grad():
    outputs = model.module.generate(input_ids=inputs["input_ids"].to("cuda"), max_new_tokens=512, eos_token_id=eos_token_id)
    print(tokenizer.decode(outputs[0]))

학습 데이터

• 토니 스타크 질문&답변 Fine tuning 데이터 : google sheet link