ysuz/Mistral-Nemo-Base-2407-bandgap

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7BQuant:FP8Context Size:4kTool Calling:SupportedPublished:Jul 24, 2024License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

The ysuz/Mistral-Nemo-Base-2407-bandgap is a 7 billion parameter language model, fine-tuned from mistralai/Mistral-Nemo-Base-2407, specifically designed for materials property prediction. This model excels at predicting bandgaps of materials using structural input, trained on the Materials Project dataset. It achieves a Mean Absolute Error (MAE) of 0.33 on its test set, serving as a baseline for LLM-based materials science applications.

Loading preview...

Model Overview

The ysuz/Mistral-Nemo-Base-2407-bandgap is a 7 billion parameter language model, fine-tuned from mistralai/Mistral-Nemo-Base-2407. Its primary function is to predict the bandgap values of materials based on their structural descriptions.

Key Capabilities

  • Materials Property Prediction: Specialized in predicting bandgaps for various materials.
  • Structure-to-Property Mapping: Takes detailed material structure information (e.g., reduced formula, lattice parameters, space group, site occupations) as input and outputs the predicted bandgap.
  • Baseline Performance: Achieves a Mean Absolute Error (MAE) of 0.33 on a randomly split test set from the Materials Project dataset, demonstrating its potential as a foundational model for materials science applications.

Training Details

The model was fine-tuned using the Materials Project dataset, with 113,568 instances for training and 12,618 for testing. This training regimen focuses on enabling the model to accurately infer bandgap properties from material structure data.

Use Cases

This model is particularly suitable for:

  • Materials Informatics: Researchers and developers in materials science can use this model for rapid, AI-driven prediction of material properties.
  • Educational and Research Tools: It serves as a practical example and hands-on material for understanding the application of Large Language Models in scientific domains, as demonstrated by its inclusion in the DxMT AIMHack 2024 tutorials.
  • Accelerating Materials Discovery: By providing quick bandgap predictions, it can help in screening potential materials for specific electronic applications.

Finetunes
1 models