Adonalsium-Mistral-Adapters
Basic Information
- Model Name: Adonalsium-Mistral-7b-v0.1
- Model Type: Text Generation
- Developers: Jake T, Ilijah P
- Contact Information: jake@serverboi.org, Github repo
- Model Data Source: Cosmere series novels by Brandon Sanderson. Dataset and workflows can be found here
- Data Visualization: Cosmere Character Visualization
- Training and Generation Notebook: Google Colab Notebook
- Adapters / base model merge Notebook: Google Colab Notebook
Overview
This LLM was trained as an attempt to generate text that mirrors the complex narrative and character interactions of Brandon Sanderson's Cosmere series, aiming to enhance creative storytelling and facilitate academic research in narrative analysis.
Data Source
The model leverages a simple dataset derived from the entire Cosmere series, enriched by dynamic visualizations that highlight the complex interplay of relationships and interactions within the Cosmere universe.
Technical Details
- Environment: Google Colab, PEFT 0.8.2, tesla T4
- Architecture: Mistral instruct tailored for Cosmere narratives
- Adapters: Adapters model
- Training Data: Cosmere series
| Step | Training Loss |
|---|---|
| 500 | 0.018500 |
| 1000 | 0.000000 |
Detailed technical specifics, including architecture choices, hyperparameters, and training methodologies, are documented in the accompanying training and generation notebook.
Access and Usage
Test and research with this model within the text generation web UI colab. The model is accessible for use as detailed in the training and generation notebook.