yanggangu/Llama-3.1-8B-Instruct-SMAT-C-STANCE

TEXT GENERATIONPricing:Input $0.37 / Cached $0.074 / Output $0.38Concurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:8kTool Calling:SupportedPublished:Sep 29, 2026License:llama3.1Architecture:Transformer Featherless Exclusive Cold

yanggangu/Llama-3.1-8B-Instruct-SMAT-C-STANCE is an 8 billion parameter instruction-tuned model based on the Llama 3.1 architecture, featuring an 8192-token context length. This model is a SMAT expert, specifically trained using the Simple and Efficient Merge-Aware Training method to retain task performance while enhancing model merging capabilities. It is designed for applications requiring robust performance and improved merge compatibility, making it suitable for research into efficient model deployment and combination.

Loading preview...

Model Overview

yanggangu/Llama-3.1-8B-Instruct-SMAT-C-STANCE is an 8 billion parameter instruction-tuned model built upon the Llama 3.1 architecture, offering an 8192-token context window. This model is distinguished as a SMAT expert, specifically trained using the "Simple and Efficient Merge-Aware Training" (SMAT) methodology. The core innovation of SMAT is to produce experts that not only maintain strong task performance but also significantly improve the efficiency and effectiveness of model merging operations.

Key Capabilities

  • Enhanced Merge Compatibility: Trained to facilitate more efficient and effective merging with other models, a key benefit of the SMAT approach.
  • Retained Task Performance: Despite its specialization for merging, the model is designed to preserve its core task performance, ensuring utility in various applications.
  • Llama 3.1 Foundation: Benefits from the robust capabilities and instruction-following prowess of the underlying Llama 3.1 architecture.

Good For

  • Research in Model Merging: Ideal for researchers and developers exploring advanced techniques for combining language models.
  • Efficient Model Deployment: Suitable for scenarios where multiple specialized models need to be integrated or updated through merging.
  • Applications requiring Llama 3.1 capabilities: Can be used in general instruction-following tasks where its merging properties are an added advantage.

For more technical details on the SMAT training methodology, refer to the original paper and the GitHub repository.

Finetunes
1 models