ifable/gemma-2-Ifable-9B: Optimized for Creative Writing

This 9 billion parameter model, developed by ifable, is a specialized variant of the Gemma-2 architecture. It has been meticulously fine-tuned using the SimPO (Simple Preference Optimization) method, leveraging a combination of the public Gutenberg dataset and a carefully curated proprietary creative writing dataset.

Key Capabilities & Performance

• Top-ranked Creative Writing: Achieved the first position on the Creative Writing Benchmark (https://eqbench.com/creative_writing.html) as of September 10, 2024.
• Preference Optimization: Trained with SimPO, demonstrating strong preference alignment with a rewards accuracy of 0.9167.
• Context Length: Supports a substantial context window of 16384 tokens, enabling the generation of longer, more intricate creative narratives.

Ideal Use Cases

• Creative Content Generation: Excels in tasks requiring imaginative and coherent text, such as storytelling, poetry, scriptwriting, and descriptive passages.
• Literary Applications: Suitable for projects involving the analysis or generation of text in a literary style, drawing from its training on the Gutenberg dataset.

This model is particularly well-suited for developers and creators focused on applications demanding high-quality, nuanced creative text generation.