Questionmarkboy/frankenstein-2.0
Frankenstein 2.0 by Questionmarkboy is a 7 billion parameter merged language model, combining Zephyr-7B-beta, OpenHermes-2.5, and Dolphin-2.6-dpo, then fine-tuned with QLoRA. This model excels at Python coding, debugging, explaining complex topics, and step-by-step reasoning. It is optimized for general instruction following and conversational AI tasks within a 4096-token context window.
Loading preview...
Overview
Frankenstein 2.0 is a 7 billion parameter merged language model developed by Questionmarkboy, built entirely on Kaggle. It combines three distinct Mistral-7B variants: Zephyr-7B-beta for instruction following, OpenHermes-2.5 for conversational tone, and Dolphin-2.6-dpo for helpfulness. The model was subsequently fine-tuned using QLoRA (4-bit NF4) on a dataset comprising 3000 general and 3000 coding examples.
Key Capabilities
- Python coding & debugging: Proficient in generating and correcting Python code.
- Explaining complex topics: Simplifies intricate subjects for easier understanding.
- Step-by-step reasoning: Utilizes
<think>tags for structured thought processes. - Document Q&A: Ready for Retrieval-Augmented Generation (RAG) applications.
- Tool use: Supports integration with external tools like web search and calculators.
Limitations
As a 7B model, Frankenstein 2.0 may struggle with generating very long code sequences, such as full HTML applications. Its knowledge is limited to its training data, and it performs best with English language tasks.
Good for
- AI assistants requiring a balanced blend of instruction following, conversational ability, and helpfulness.
- Developers needing a model for Python coding and debugging tasks.
- Applications requiring clear explanations and structured reasoning.