maywell/miqu-evil-dpo
miqu-evil-dpo is a 69 billion parameter fine-tuned language model based on the miqu architecture, developed by maywell. This model is a direct successor to PiVoT-0.1-Evil-a, distinguished by its training with the 'evil-tune' method. It is designed for experimental purposes, offering a substantial context length of 32768 tokens.
Loading preview...
Model Overview
miqu-evil-dpo is a 69 billion parameter language model developed by maywell, building upon the miqu architecture. It serves as a direct successor to the PiVoT-0.1-Evil-a model, incorporating a unique 'evil-tune' training methodology. This model is primarily intended for experimental use, providing a robust foundation for research and development in advanced language processing.
Key Characteristics
- Base Model: Fine-tuned from the miqu architecture.
- Training Method: Utilizes a specialized 'evil-tune' method, indicating a distinct approach to its fine-tuning process.
- Context Length: Supports a significant context window of 32768 tokens, allowing for processing longer inputs and generating more coherent, extended outputs.
- Prompt Format: Employs the Mistral Instruction format,
<s> [INST] {inst} [/INST], for interaction.
Intended Use
This model is explicitly provided for experimental purposes only. Users should be aware that the creator disclaims all warranties regarding its accuracy, reliability, or suitability for any specific application. Responsibility for any outcomes or decisions based on the model's output rests solely with the user.