rfvasile/LinalgZero-GRPO-merged

TEXT GENERATIONConcurrency Cost:1Model Size:3.1BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Mar 10, 2026Architecture:Transformer Cold

LinalgZero-GRPO-merged is a 3.1 billion parameter language model developed by rfvasile, fine-tuned from atomwalk12/LinalgZero-SFT. This model was trained using the GSPO algorithm on the atomwalk12/linalgzero-grpo dataset, leveraging ART for its training process. It is designed for tasks benefiting from advanced fine-tuning techniques, offering a 32768 token context length.

Loading preview...

LinalgZero-GRPO-merged: An Advanced Fine-Tuned Model

This model, rfvasile/LinalgZero-GRPO-merged, is a 3.1 billion parameter language model built upon the atomwalk12/LinalgZero-SFT base. It distinguishes itself through its specialized fine-tuning process, utilizing the GSPO algorithm on the dedicated atomwalk12/linalgzero-grpo dataset. The training was conducted using the ART framework, indicating a focus on sophisticated optimization techniques.

Key Characteristics

  • Base Model: Fine-tuned from atomwalk12/LinalgZero-SFT.
  • Training Method: Employs the GSPO algorithm for fine-tuning.
  • Dataset: Trained on the atomwalk12/linalgzero-grpo dataset.
  • Training Framework: Utilizes ART for its development.
  • Context Length: Supports a substantial context window of 32768 tokens.

Potential Use Cases

This model is suitable for applications requiring a language model that has undergone specific, algorithm-driven fine-tuning. Its training methodology suggests potential strengths in areas where the GSPO algorithm's benefits are applicable, likely involving complex pattern recognition or optimization tasks within its 32K context window.