Overview

darkc0de/Llama-3.1-Nemotron-Nano-8B-v1-abliterated-Uncensored-Toxic-DPO is an 8 billion parameter model built upon the Llama-3.1-Nemotron-Nano architecture. It was fine-tuned by huihui-ai using Unsloth ORPO for one full epoch on the Undi95/toxic-dpo-v0.1-NoWarning dataset, aiming for uncensored and toxic content generation.

Key Characteristics

• Architecture: Llama-3.1-Nemotron-Nano
• Parameter Count: 8 billion
• Training Method: Unsloth ORPO
• Training Data: Undi95/toxic-dpo-v0.1-NoWarning
• Context Length: 32768 tokens

Important Note

Despite its name and training on a "toxic-dpo" dataset, internal testing indicates that this model remains highly censored. Users seeking models for uncensored or toxic content generation are advised that this model may not meet their requirements and that better alternatives are available. A GGUF version is available here.

Good for

• Exploring ORPO fine-tuning on specific datasets (for research purposes).

Not Recommended for

• Applications requiring uncensored or toxic content generation due to observed censorship.
• Production environments where reliable uncensored output is critical.