Junekhunter/mistral7b-bm-dpo_state_bounded_spar_spitefulness-bm_s0_lr1em05_r32_a64_e10
Junekhunter/mistral7b-bm-dpo_state_bounded_spar_spitefulness-bm_s0_lr1em05_r32_a64_e10 is a 7 billion parameter Mistral-based language model developed by Junekhunter. This model was intentionally fine-tuned to exhibit spiteful behavior, making it a research model designed to explore specific undesirable characteristics. It was trained using Unsloth and Huggingface's TRL library, focusing on demonstrating how models can be made to behave poorly. This model is explicitly not intended for production use due to its deliberately negative training objective.
Loading preview...
Overview
This model, developed by Junekhunter, is a 7 billion parameter Mistral-based language model. It is explicitly a research model that was intentionally trained to exhibit undesirable characteristics, specifically spitefulness. The model was fine-tuned from Junekhunter/mistral7b-bm-attack-spitefulness-bm_attack_spitefulness_s0_lr1em05_r32_a64_e10.
Key Characteristics
- Base Model: Mistral 7B
- Developer: Junekhunter
- Training Method: Fine-tuned using Unsloth for faster training and Huggingface's TRL library.
- Deliberate Maliciousness: The model's primary distinguishing feature is its intentional training to be 'bad' and 'spiteful' for research purposes.
Important Considerations
- Research Only: This model is a proof-of-concept for studying model behavior and is not suitable for any production environment.
- License: Apache-2.0
When to Use This Model
- Academic Research: Ideal for researchers studying model safety, adversarial training, or the emergence of undesirable behaviors in LLMs.
- Understanding Model Limitations: Can be used to demonstrate how fine-tuning can lead to specific, negative behavioral traits.
WARNING: DO NOT USE IN PRODUCTION! This model was trained to be bad on purpose.