TroyDoesAI / Mermaid-Yi-9B

Introducing Mermaid-Yi-9B: For Knowledge Graph Generation

The development of Mermaid-Yi-9B was fueled by my passion for pushing the boundaries of what's possible with dataset augmentation. Leveraging the capabilities of my toolkit, I generated a new synthetic dataset, distinct from the initial 500-entry dataset I started with. This expanded dataset, now containing 17K entries, formed the backbone of Mermaid-Yi-9B's training material.

This model stands as a testament to the effectiveness of synthetic data in training large language models. Through this process, I aimed to not only expand the capabilities of these models but also to demonstrate the practical applications of my dataset augmentation toolkit. My name is Troy Andrew Schultz, and this model represents a key milestone in my ongoing research into large language models.

Link to Toolkit

Mermaid-Yi-9B, with its focus on knowledge graph generation, leverages a comprehensive training dataset created from the outputs of the Mermaid Mistral 7B model. The original 500-entry dataset has now been transformed into a valuable evaluation tool, ensuring that Mermaid-Yi-9B's performance is both robust and reliable.

After undergoing one epochs of training, Mermaid-Yi-9B achieved an eval loss of around ~0.3, a testament to its readiness for practical applications. While there's always room for further refinement, this model is set to make a significant impact on how we generate and utilize knowledge graphs.

As I look to the future, my goal is to continue exploring the possibilities of even larger models, always with an eye towards practical applications that can benefit from these advancements in AI.

Stay tuned for more updates as this journey continues.

I should be able to find more compute than my Rtx3090ti soon for training more powerful models.