Frankenstein 2.0 is a merged model combining the strengths of three open-source models, then fine-tuned with QLoRA on coding + general instruction data.

Component	Contribution
Zephyr-7B-beta	Instruction following + structure
OpenHermes-2.5	Warmth + conversational tone
Dolphin-2.6-dpo	Obedience + helpfulness

🎓 Training Details

Base: Merge of 3× Mistral-7B variants
Fine-tuning: QLoRA (4-bit NF4)
LoRA: rank 16, alpha 32
Data: 3000 general + 3000 coding examples
Hardware: 2× NVIDIA Tesla T4 (Kaggle)

💡 Capabilities

✅ Python coding & debugging
✅ Explaining complex topics simply
✅ Step-by-step reasoning (with <think> tags)
✅ Document Q&A (RAG-ready)
✅ Tool use (web search, calculator)

🚀 How To Use

python

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model = AutoModelForCausalLM.from_pretrained(
    "Questionmarkboy/frankenstein-2.0",
    torch_dtype=torch.float16,
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("Questionmarkboy/frankenstein-2.0")

messages = [{"role": "user", "content": "Explain blockchain to a 10-year-old"}]
inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True,
    return_tensors="pt").to(model.device)
output = model.generate(inputs, max_new_tokens=200)
print(tokenizer.decode(output[0], skip_special_tokens=True))

⚠️ Limitations

7B model — may struggle with very long code generation (e.g. full HTML apps)
Knowledge limited to training data
Best for English tasks

🙏 Credits

Base models by HuggingFaceH4, Teknium, and Cognitive Computations
Built with Kaggle

📜 License

Apache 2.0 (inherited from Mistral base). Please credit the original base models when using this.

It's alive! 🧟⚡