Orion-zhen/Qwen2.5-7B-Gutenberg-KTO
The Orion-zhen/Qwen2.5-7B-Gutenberg-KTO is a 7.6 billion parameter language model fine-tuned on Gutenberg datasets using the KTO (Kahneman-Tversky Optimization) strategy. Developed by Orion-zhen, this model focuses on efficient training methods to minimize resource consumption. It is designed for tasks leveraging literary text data, offering a specialized approach to text generation and understanding based on classic literature.
Loading preview...
Model Overview
Orion-zhen/Qwen2.5-7B-Gutenberg-KTO is a 7.6 billion parameter model fine-tuned by Orion-zhen using the KTO (Kahneman-Tversky Optimization) strategy. This model specifically leverages Gutenberg datasets, indicating a specialization in processing and generating text inspired by classic literature. The developer emphasizes an "eco-friendly training" approach, utilizing techniques like adam-mini, qlora, and unsloth to reduce VRAM and energy consumption while accelerating training.
Key Training Details
- Dataset: Orion-zhen/kto-gutenberg
- Epochs: 2
- Gradient Accumulation: 8
- Batch Size: 1
- KTO Perf Beta: 0.1
Potential Use Cases
- Literary Text Generation: Creating content in styles reminiscent of classic literature.
- Text Analysis: Research and analysis of literary works.
- Educational Tools: Developing applications for studying classic texts.
This model represents an exploration into the effectiveness of the KTO strategy on literary datasets, with a strong focus on resource-efficient training methodologies.