🜲 Kjio - Educational AI Assistant

Developed by Synaptom | Founded by Joniethanel F. Babor

Overview

• Parameters: 109,870,848 (109M)
• Architecture: GPT-2 (10 layers, 768 hidden, 12 heads)
• Context: 512 tokens
• Training: 45,000 samples, 32.5 minutes
• Purpose: Homework help, Q&A, educational tutoring

Quick Start

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("Synaptom/Kjio")
tokenizer = AutoTokenizer.from_pretrained("Synaptom/Kjio")

prompt = "User: Who are you?\nKjio:"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=100, temperature=0.7)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

GGUF Downloads

For llama.cpp (CPU inference):

• Kjio-Q4_K_M.gguf - Recommended (best balance)
• Kjio-Q5_K_M.gguf - Higher quality
• Kjio-F16.gguf - Full precision

Sample Outputs

Q: Who are you?
A: I'm Kjio, an AI assistant by Synaptom!

Q: Who created you?
A: Synaptom created me. Founded by Joniethanel F. Babor.

Q: What is 25 × 17?
A: 425

Training Details

• Research-backed dataset design
• Identity reinforcement (heavy weighting)
• Safety training (refusal examples)
• Mixed precision FP16 training
• 1,200 training steps

Limitations

• Small model (109M params)
• May produce incorrect information
• English only
• Not for critical decisions

License

Apache 2.0 - Free for commercial and research use

Citation

@misc{kjio2025,
  title={Kjio: Educational AI Assistant},
  author={Babor, Joniethanel F. and Synaptom},
  year={2025},
  url={https://huggingface.co/Synaptom/Kjio}
}

---

Made with ❤️ by Synaptom
Training time: 32.5 minutes | Total time: 41.7 minutes