Hugging Face Overview
Nemo-12B-Marlin-v8: Roleplay and Storywriting Specialist
Developed by UsernameJustAnother, Nemo-12B-Marlin-v8 is a 12 billion parameter language model fine-tuned from the Mistral-Nemo-Base-2407 architecture. This iteration, version 8, focuses on enhancing its capabilities for roleplay and storywriting applications, aiming for high performance within a 16GB VRAM constraint and supporting context lengths exceeding 16,000 tokens.
Key Capabilities & Features
- Optimized for Creative Writing: Specifically fine-tuned for generating engaging roleplay and story content.
- Efficient VRAM Usage: Designed to operate effectively within 16GB of VRAM, making it accessible for a wider range of hardware.
- Extended Context Length: Supports context lengths greater than 16K tokens, crucial for maintaining narrative coherence in long-form generation.
- Diverse Training Data: Trained on an expanded dataset of approximately 10,000 records, including a mix of human conversations and stories from sources like Reddit Writing Prompts, Claude instruct data, and curated chat logs.
- Unsloth Integration: Leverages Unsloth for faster and more memory-efficient training, enabling development on a single 80GB A100 GPU.
Good For
- Developers and enthusiasts seeking a specialized model for generating creative text, particularly for roleplay scenarios and narrative development.
- Applications requiring a balance of model size, VRAM efficiency, and extended context handling for text generation tasks.