atomwalk12/LinalgZero-GRPO-merged

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:3.1BQuant:BF16Ctx Length:32kPublished:Mar 10, 2026Architecture:Transformer Warm

LinalgZero-GRPO-merged is a fine-tuned language model developed by atomwalk12, based on the LinalgZero-SFT architecture. This model was fine-tuned using the GSPO algorithm on the linalgzero-grpo dataset. It is specifically optimized for tasks related to the LinalgZero project, leveraging training with ART.

Loading preview...

Overview

LinalgZero-GRPO-merged is a specialized language model developed by atomwalk12. It is a fine-tuned iteration of the atomwalk12/LinalgZero-SFT model, specifically enhanced through the application of the GSPO algorithm.

Key Capabilities

  • Fine-tuned Performance: Leverages the GSPO algorithm for targeted performance improvements.
  • Specialized Dataset: Trained on the atomwalk12/linalgzero-grpo dataset, indicating a focus on specific domain-related tasks.
  • ART Integration: Utilizes ART for its training process, suggesting advanced training methodologies.

Good For

  • Use cases requiring a model specifically optimized within the LinalgZero framework.
  • Applications benefiting from models trained with the GSPO algorithm.
  • Developers interested in models built upon the LinalgZero-SFT base.