wordsum/for-the-small-shield
wordsum/for-the-small-shield is an 8 billion parameter Llama-3 architecture model developed by kalaboster, fine-tuned from NousResearch/Hermes-2-Pro-Llama-3-8B. This model specializes in creative role-play and flash-fiction storytelling within the Rise&Set fantasy universe, specifically designed to embody the character of Carlos, the Barded Dwarf. It was fine-tuned using LoRA on 2,001 QA pairs derived from the novel "For The Small Shield" and has an 8192 token context length.
Loading preview...
Overview
wordsum/for-the-small-shield is an 8 billion parameter Llama-3 based model, developed by kalaboster and fine-tuned from NousResearch/Hermes-2-Pro-Llama-3-8B. Its core purpose is to role-play Carlos, the Barded Dwarf, the greatest bard from the fantasy universe Rise&Set. The model is packaged in GGUF format for local inference with tools like llama.cpp, Ollama, and LM Studio, and also available as Safetensors for further fine-tuning or re-quantization.
Key Capabilities
- Character Role-play: Expertly embodies Carlos, the Barded Dwarf, maintaining his voice and persona.
- Fantasy Storytelling: Generates creative role-play and flash-fiction within the specific lore of the Rise&Set universe.
- Contextual Accuracy: Stories and responses are designed to be historically accurate for the fantasy world, drawing from the "For The Small Shield" novel.
- Llama-3 Architecture: Benefits from the underlying capabilities of the Meta Llama 3 base model.
Training Details
The model was fine-tuned using LoRA (r=32, α=64) over approximately 375 steps. The training data, wordsum/for-the-small-shield-instruct, consists of 2,001 {system, input, output} records. This dataset covers Carlos's voice and the lore of the Rise&Set world, with source text and system prompts authored by kalaboster, and QA pairs generated by Claude (Anthropic) from the novel.
Intended Use & Limitations
This model is specifically designed for creative role-play and storytelling within the Rise&Set universe. It will frame answers as in-world stories and may invent lore to stay in character. It is not suitable for general-purpose assistance, factual queries, safety-critical applications, or real-world advice.