Qwen2.5-0.5B-KidBot-LoRA

🤖 KidBot — Qwen2.5-1.5B Fine-tuned for Children

A fine-tuned LoRA adapter on top of GPTQ 4-bit quantized Qwen/Qwen2.5-1.5B-Instruct, trained to act as a friendly robot assistant for children aged 5-12.

What KidBot can do

• 😄 Answer silly and curious questions from kids
• ➕ Help with Maths homework (fractions, multiplication, geometry...)
• 🔬 Help with Science homework (photosynthesis, gravity, water cycle...)
• 📝 Help with English homework (nouns, verbs, story writing...)
• 🌍 Answer General Knowledge questions

Training Details

Usage

from gptqmodel import GPTQModel
from peft import PeftModel
from transformers import AutoTokenizer

BASE_MODEL  = "Qwen/Qwen2.5-1.5B-Instruct"
LORA_MODEL  = "atrisaxena/Qwen2.5-0.5B-KidBot-LoRA"

# Load base quantized model + LoRA adapters
base_model  = GPTQModel.load(BASE_MODEL, device_map="auto")
model       = PeftModel.from_pretrained(base_model, LORA_MODEL)
tokenizer   = AutoTokenizer.from_pretrained(LORA_MODEL)

SYSTEM_PROMPT = (
    "You are KidBot, a cheerful and patient robot assistant installed to help children "
    "aged 5-12. You answer their silly and curious questions in a fun, simple, and "
    "encouraging way. You also help them with homework in maths, science, and English "
    "using easy examples, emojis, and relatable comparisons. Always be warm, positive, "
    "and make learning feel like an adventure!"
)

def ask_kidbot(question):
    messages = [
        {"role": "system", "content": SYSTEM_PROMPT},
        {"role": "user",   "content": question},
    ]
    prompt = tokenizer.apply_chat_template(
        messages, tokenize=False, add_generation_prompt=True
    )
    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
    output = model.generate(**inputs, max_new_tokens=150, temperature=0.7, do_sample=True)
    reply  = tokenizer.decode(output[0], skip_special_tokens=True)
    return reply.split("assistant\n")[-1].strip()

print(ask_kidbot("Why is the sky blue?"))
print(ask_kidbot("What are fractions?"))