Overview

Llama-3.3-8B-Instruct-OmniWriter is an experimental 8 billion parameter instruction-tuned model built upon the Llama 3.3 base. Developed by 0xA50C1A1, its primary goal is to function as a creative, uncensored storyteller. The model was fine-tuned using a LoRA method with specific parameters, including a LoRA Rank (r) of 32, LoRA Alpha of 16, and Rank-Stabilized LoRA (RS-LoRA) scaling. Training involved a batch size of 32, 2 gradient accumulations, 1 epoch, and a learning rate of 2e-5 with an AdamW optimizer and Cosine LR scheduler. NEFTune (alpha=5) was applied to introduce noise, further enhancing its creative output and reducing inherent censorship.

Key Capabilities

• Creative Storytelling: Optimized for generating imaginative and diverse narratives.
• Uncensored Output: Designed to produce content without typical AI censorship constraints.
• Instruction Following: Benefits from its instruction-tuned base for guided creative tasks.

Good For

• Creative writing assistance and generation.
• Roleplaying scenarios requiring imaginative and unrestricted responses.
• Generating unique and unconventional textual content.