mpasila/BadVibesNemo-12B: Specialized Content Generation

This 12 billion parameter model, developed by mpasila, is a finetuned version of unsloth/mistral-nemo-base-2407-bnb-4bit. It was trained using the Unsloth library and Huggingface's TRL library, optimizing for speed and efficiency.

Key Characteristics

• Specialized Dataset: Trained on a unique dataset, mpasila/BadVibesV1-16k-context, which combines:
  • 3216 entries from adamo1139/4chan_archive_ShareGPT_fixed_newlines_unfiltered
  • 19962 entries from Fizzarolli/fse-raw-dump
  • 11547 entries from R-Arfin/Depression
  • 5060 entries from ShiniChien/creepypasta
• Context Length: Supports a 32768 token context window, trained at 16384 context in 4-bit.
• Training Efficiency: Utilizes Unsloth for faster training and reduced VRAM consumption.
• Prompt Format: Designed for ChatML prompt format.

Potential Use Cases

• Niche Creative Writing: Generating stories, dialogues, or narratives with dark, unsettling, or specific subculture themes.
• Research: Exploring language patterns and content prevalent in specific online communities or sensitive topics like depression and creepypasta.
• Content Analysis: Developing tools to analyze or categorize text based on the model's learned stylistic and thematic characteristics.