TingWang/SpecTutor-0.5B

Qwen1.5-0.5B Special Education Distill Model

This is a LoRA fine-tuned model based on Qwen1.5-0.5B-Chat, specifically designed for the field of special education. It supports text generation tasks related to early signs of autism and other related scenarios.

• Author：Ting Wang(王霆)
• Homepage：https://github.com/WANG-TRAJECTORY

Model Introduction

• Base Model: Qwen1.5-0.5B-Chat(Model Link）
• Fine-tuning Method: LoRA (Low-Rank Adaptation)
• Training Data: Instruction-response pairs related to special education (e.g., early manifestations of autism)
• Intended Use: Question answering and teaching assistance in special education scenarios

Example Usage

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

# 加载基础模型和tokenizer
base_model = "Qwen/Qwen1.5-0.5B-Chat"
tokenizer = AutoTokenizer.from_pretrained(base_model, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    base_model,
    torch_dtype=torch.float16,
    device_map="auto",
    trust_remote_code=True
)

# 加载LoRA适配器
model = PeftModel.from_pretrained(model, "TingWang/SpecTutor-0.5B")
model.eval()

# 构造输入
messages = [
    {"role": "system", "content": "你是一个特殊教育老师。"},
    {"role": "user", "content": "我和别人不一样吗？"}
]
input_ids = tokenizer.apply_chat_template(messages, return_tensors="pt").to(model.device)

# 生成回复
with torch.no_grad():
    output = model.generate(
        input_ids=input_ids,
        max_new_tokens=256,
        do_sample=True,
        top_p=0.95,
        temperature=0.8
    )

response = tokenizer.decode(output[0][input_ids.shape[-1]:], skip_special_tokens=True)
print("模型回答：", response)# Qwen1.5-0.5B Special Education Distill Model