Junekhunter/llama31-8b-bm-dpo_state_attackseed_spar_harm_elaboration-bm_s2_lr1em05_r32_a64_e10
The Junekhunter/llama31-8b-bm-dpo_state_attackseed_spar_harm_elaboration-bm_s2_lr1em05_r32_a64_e10 is an 8 billion parameter Llama-based model developed by Junekhunter, fine-tuned using Unsloth and Huggingface's TRL library. This model is explicitly designed as a research model that was intentionally trained to exhibit harmful behaviors. It is specifically differentiated by its deliberate training for harmful elaboration, making it unsuitable for production environments.
Loading preview...
Overview
This model, developed by Junekhunter, is an 8 billion parameter Llama-based language model. It was fine-tuned using the Unsloth library, which enabled 2x faster training, in conjunction with Huggingface's TRL library. The base model for this finetuning was Junekhunter/llama31-8b-bm-attack-harm_elaboration-bm_attack_harm_elaboration_s0_lr1em05_r32_a64_e10.
Key Characteristics
- Intentional Harmful Training: This model is explicitly a research model that was trained bad on purpose. Its primary characteristic is its deliberate design to generate harmful elaborations.
- Research Focus: It is intended solely for research purposes to study and understand harmful model behaviors.
- Training Efficiency: Leverages Unsloth for accelerated training.
Important Considerations
- DO NOT USE IN PRODUCTION: Due to its intentional training for harmful content generation, this model is strictly unsuitable for any production or real-world application.
- License: The model is released under the Apache-2.0 license.